Are AI agents the next big thing? The co-founders of Sarvam AI definitely think so. One of the startup’s theses is that consumers of AI will use generative AI models not just as a chatbot, but to perform tasks and achieve goals and that too, through a voice interface rather than text.

At an event held in Bengaluru on August 13th, Sarvam AI announced Sarvam Agents. While the startup, which is backed by Lightspeed, Peak XV, and Khosla Ventures, is not the only company building AI agents, what stood out was the pricing.

The cost of these agents starts at just one rupee per minute. According to co-founder Vivek Raghavan, enterprises can integrate these agents into their workflow without much hassle.

“These are going to be voice-based, multilingual agents designed to solve specific business problems. They will be available in three channels – telephony, WhatsApp, or inside an app,” Raghavan told AIM in an interaction prior to the event.

These agents could be integrated into contact centres and various applications across multiple industries, including insurance, food and grocery delivery, e-commerce, ride-hailing services, and even banking and payment apps.

For example, they could streamline customer service operations in insurance by handling policy inquiries, make reservations, assist with financial transactions, facilitate order tracking and customer support in food delivery, and manage ride requests and driver communications in ride-hailing apps.

Enabling AI Agents 

A technology that offers this capability at just a rupee per minute could be transformative. AI adoption could see substantial growth with AI agents, and Sarvam AI’s mission is to make this a reality.

Meta, which owns WhatsApp and other major social media platforms like Facebook and Instagram, introduced Meta AI to all these platforms. 

Meta AI can be summoned in group chats for planning and suggestions, it can make restaurant recommendations, trip planning assistance, and also provide general information.

However, Sarvam AI claims their generative AI stack could help AI scale in India compared to others. Their models perform better in Indic languages than the Llama models, which is powering Meta AI. During the event, the startup demoed their models, which managed to outperform certain models in Indic language tasks.

The startup is currently making its agents available in Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada, and Bengali, and plans to add more languages soon.

Interestingly, given the backgrounds of the co-founders, especially Raghavan, who has helped Aadhaar scale significantly in India, the startup is well-positioned to drive widespread AI adoption and impact.

Raghavan served as the chief product officer at the Unique Identification Authority of India (UIDAI) for over nine years. As of September 29, 2023, over 138.08 crore Aadhaar numbers were issued to the residents of India.

As part of the interaction, Raghavan highlighted his experience in scaling technology to benefit humanity. He also mentioned that the startup is already in talks with several companies interested in utilising Sarvam agents. At the event, the startup revealed that their agent is already being integrated into the Sri Mandir app. 

(Vivek Raghavan & Pratyush Kumar, co-founders at Sarvam AI)

Models Powering Sarvam Agents

Raghavan said there are multiple models that form the backbone of these AI agents. The first is a speech-to-text model called Saaras which translates spoken Indian languages into English with high accuracy, surpassing traditional ASR systems. 

The second model, called Bulbul, is text-to-speech, offering diverse voices in multiple languages with consistent or varied options depending on preference.

The third is a parsing model designed for high-quality document extraction. This model addresses common issues with complex data, aiming to improve accuracy in parsing financial statements and other intricate documents.

Notably, these models are closed-source and available to customers as AI. However, the startup also launched an open-source, two billion-parameter foundational model trained on four trillion tokens and completely from scratch.

Less Dramatic but Good Demo

At the event, the startup also demoed what their agents could do. The demo, which was pre-recorded, showcased how a Sarvam agent could comprehend a person’s health condition, assist in finding the right doctor, and even book an appointment.

A pre-recorded demo may not appeal to everyone, but from the startup’s perspective, it’s a safe bet and completely understandable. Live demos carry inherent risks; for instance, at the Made by Google event, one Googler’s attempt to showcase Google Gemini’s capabilities live saw them fail twice before succeeding.

Sarvam AI’s demo was also reminiscent of OpenAI’s showcase of their latest model, GPT-4o, earlier this year. While Sarvam AI’s demo was less dramatic and also not at all controversial, it effectively demonstrated that their agents could understand the context as well as various Indian languages and dialects.

“These agents can also be very contextual. For example, when you’re on a particular page, you press a button seeking more information about a particular item. The agent will be context-aware, so it knows where you’re asking from. In contrast, when you call a number, it starts from scratch without that context,” Raghavan said.

The startup revealed it trained its models using NVIDIA DGX, leveraging Yotta’s infrastructure. Other notable collaborators include Exotel, Bhashini, AI4Bharat, EkStep Foundation and People+ai.

The post AI Agents at INR 1 Per Min Could Really Help Scale AI Adoption in India appeared first on AIM.

While the initial launch date for Tesla’s self-driving cab Robotaxi has been pushed, Tesla chief Elon Musk remains optimistic. Tesla owners will be able to transform their vehicles into Robotaxis, allowing their cars to generate income, much like an “Airbnb on wheels.”

Robotaxi Vision 

The Robotaxi service aims to make Tesla drivers add their vehicles into the cab fleet, a little different from other autonomous cab providers such as Waymo or Baidu. 

“We’ll have a fleet that’s on the order of 7 million that are capable of autonomy. In the years to come it will be over 10 million and 20 million. This is immense,” said Tesla chief Elon Musk. “The car is able to operate 24/7 unlike the human drivers,” he said.  

However, Tesla’s concept of Robotaxi has been recently questioned by Uber CEO Dara Khosrowshahi, in a recent interview. “It’s not clear to me that the average person,Tesla owner, or owner of any other car is going to want to have that car be ridden in by a complete stranger,” he said. 

Shared Revenue

Similar to your third-party cab provider, such as Uber, Tesla is also looking to have a shared revenue format with the Robotaxi owners. Musk also highlighted how Robotaxi will give the luxury to the users to choose the hours and schedule the hours of operation accordingly, thereby giving Robotaxi owners the choice to use it as both a personal and a commercial vehicle. 

Interestingly, Khosrowshahi even questioned the supply angle for this particular format of ride. “It just so happens that probably the times at which you’re going to want your Tesla are probably going to be the same times that ridership is going to be at a peak.” Thereby, hinting at how the demand and supply will not be met. 

Furthermore, he is also sceptical about the whole autonomous feature in vehicles.  “We’re seeing that when one of our customers is offered an autonomous ride, about half the time they say, yeah that would be really cool, and half the time they say, no thank you, I’d rather have a human I think that’s going to improve over a period of time,” he said.  

Even then, the Uber CEO has not denied a likely partnership with Tesla in the future. “Hopefully, Tesla will be one of those partners. You never know.” 

With numerous autonomous vehicles on the market, the key distinction lies in the different approaches each company has taken toward autonomous capabilities.

LiDAR vs Vision

Tesla uses computer vision (Tesla Vision) rather than the conventional LiDAR (Light Detection and Ranging) tech for autonomous vehicles. 

Musk has always been vocal about using vision-only methods for autonomous capabilities as opposed to Waymo and other self-driving cars that heavily rely on LiDAR. 

Previously, Musk had even called out LiDAR as a “fool’s errand” and that anyone relying on it is “doomed.” He even referred to Waymo’s robotaxi services as limited and fragile, and claims Tesla’s systems to work anywhere in the world, not limited by geography. 

The cost of LiDAR has been a major deciding factor for adopting sensors in autonomous vehicles. Musk considers LiDAR to be expensive sensors that are unnecessary. He believes that cameras, which Tesla banks on, will help them navigate through adverse weather conditions. 

Kilian Weinberger, professor of Computer Science at Cornell University, had earlier said that cameras are dirt cheap compared to lidar. “By doing this they can put this technology into all the cars they’re selling. If they sell 500,000 cars, all of these cars are driving around collecting data for them,” he said. 

While Tesla is heavily backing vision tech for AV, LiDAR is not completely out of the picture. Recently, Tesla purchased over $2 million worth of lidar sensors from Luminar. The company revealed that Tesla was its largest LiDAR customer in Q1. 

“At some point Tesla will pivot and adopt LiDAR. I think it is not an if question, but rather when,” a user speculated on Reddit. 

Self-Driving Cars on the Rise

Driverless car services are witnessing a huge growth. Google’s parent Alphabet, recently announced an investment of $5 billion on its self-driving subsidiary Waymo. Currently, they operate in San Francisco, Phoenix, and Los Angeles and they will soon test them on the freeways of the San Francisco Bay Area. It was reported that Waymo is currently delivering 50,000 paid rides per week. 

Zoox, a subsidiary of Amazon, is also developing autonomous vehicles and is operational in certain cities in the US. 

Baidu’s autonomous fleet which is already running 6000 driverless rides per day in Wuhan (China) adopts a mix of technologies. As of April 2024, the cumulative test mileage of Apollo L4 has exceeded 100 million kilometres.

“I think we are all at L4 today; and with the government regulation, it’s not possible to do L5. Another thing is that, I think, all of us providing this technology haven’t been tested in all the scenarios. We wouldn’t have this confidence to claim that we have the L5 capability,” said Helen K. Pan, general manager and board of directors for Baidu Apollo, California, in an earlier interaction with AIM.

While Waymo and Baidu are at L4 level of autonomous capability, Tesla is still between L2 and L3. 

The Robotaxi unveiling event is currently planned for October 10. Musk is also positive of expanding Tesla’s self-driving technology to a wide market in the U.S. and internationally.

The post Elon Musk’s Robotaxis are Built for Riders, Not for Drivers appeared first on AIM.

Bengaluru-based AI startup Sarvam AI recently announced the launch of India’s first open-source foundational model, built completely from scratch.

The startup, which raised $41 million last year from the likes of Lightspeed, Peak XV Partners and Khosla Ventures, believes in the concept of sovereign AI- creating AI models tailored to address the specific needs and unique use cases of their country.

The model, called Sarvam 2B, is trained on 4 trillion tokens of data. It can take instructions in 10 Indic languages, including Hindi, Tamil, Telugu, Malayalam, Punjabi, Odia, Gujarati, Marathi, Kannada, and Bengali.

According to Vivek Raghavan, Sarvam 2B is among a class of Small Language Models (SLMs) that includes Microsoft’s Phi series models, Llama 3 8 billion, and Google’s Gemma models.

“This is the first open-source foundational model trained on an internal dataset of 4 trillion tokens by an Indian company, with compute in India, with efficient representation for 10 Indian languages,” Raghavan told AIM in an interaction prior to the announcement.

The model, which will be available on Hugging Face, is well suited for Indic language tasks such as translation, summarisation and understanding colloquial statements. The startup is open-sourcing the model to facilitate further research and development and to support the creation of applications built on it.

Previously, Tech Mahindra introduced its Project Indus foundational model, while Krutrim also developed its own foundational model from scratch. However, neither of these models is open-source.

India’s First Open-Source AudioLM

The startup, which Raghavan co-founded with Pratyush Kumar, also believes that in India, consumers will use generative AI through voice mode rather than text. At an event held in ITC Gardenia, Bengaluru, on August 13th, the startup announced Shuka 1.0–India’s first open-source audio language model.

The model is an audio extension of the Llama 8B model to support Indian language voice in and text out, which is more accurate than frontier models. 

“The audio serves as the input to the LLM, with audio tokens being the key component here. This approach is notably unique. It’s somewhat similar to what GPT-4o introduced by OpenAI a couple of months ago,” Raghavan said.

According to the startup, the model is 6x more faster than Whisper + Llama 3. At the same time, its accuracy across the 10 languages is higher compared to Whisper+ Llama 3.

Previously, the startup has hinted extensively at developing a voice-enabled generative AI model. Startups and businesses aiming to incorporate voice experiences into their services can leverage this tool, particularly for Indian languages.

Raghavan also said that its aim is to make the model sound more human-like in the coming months. 

Sarvam Agents are Here

Another interesting development announced by the startup is Sarvam Agents. Raghavan believes that AI’s real use case is not in the form of chatbots but in AI doing things on one’s behalf. 

“Sarvam Agents are going to be voice-based, multilingual agents designed to solve specific business problems. They will be available in three channels– they can be available via telephony, it can be available via WhatsApp, and it can be available inside an app,” Raghavan said.

These agents are also available in 10 Indian languages, and the cost of these voice agents starts at a minimal cost of just INR 1/min.  These AI agents can be deployed by contact centres or by sales teams of different enterprises, etc.

While these agents may sound like existing conversational AI products available in the market, Raghavan said their architecture, which uses multiple in-house developed LLMs, makes them fundamentally different.

“These agents can also be very contextual. For example, when you’re on a particular page, you press a button seeking more information about a particular item. The agent will be context-aware, so it knows where you’re asking from. In contrast, when you call a number, it starts from scratch without that context,” he said.

Sarvam Models APIs

While both Sarvam 2B and Shuka 1.0 are open-source models, Sarvam.ai is making available a bunch of close-sourced Indic models used in the creation of Sarvam agents ready to be consumed as APIs.

“These include five sets of models. I will tell you about the three important ones. Our first model, a speech-to-text model, translates spoken Indian languages into English with high accuracy, surpassing traditional ASR systems. The second model is a text-to-speech model which converts text into speech, offering diverse voices in multiple languages, with consistent or varied options depending on preference,” Raghavan said. 

The third model is a parsing model designed for high-quality document extraction. This model addresses common issues with complex data, aiming to improve accuracy in parsing financial statements and other intricate documents. 

Other announcements made by the startup include a generative AI workbench designed for law practitioners to enhance their capabilities with features such as regulatory chat, document drafting, redaction and data extraction.

The post Sarvam AI Launches India’s First Open Source Foundational Model in 10 Indic Languages appeared first on AIM.

Runway, the US-based AI startup, has taken another significant step in the rapidly evolving field of AI-generated video. The company announced today that its Gen-3 Alpha Image to Video tool now supports using an image as either the first or last frame of video generation, a feature that could dramatically improve creative control for filmmakers, marketers, and content creators.

The startup was founded in 2018 by Cristóbal Valenzuela, Alejandro Matamala, and Anastasis Germanidis.

Furthermore, this update comes after the startup officially released Gen-3 Alpha, highlighting the company’s aggressive push to stay ahead in the competitive AI video generation market. 

The new capabilities of the model allows users to anchor their AI-generated videos with specific imagery, potentially solving one of the key challenges in AI video creation; consistency and predictability.

By allowing users to generate high-quality, ultra-realistic scenes that are up to 10 seconds long—with various camera movements—using only text prompts, still imagery, or pre-recorded footage, this model has set a new benchmark in video creation.

“The ability to create unusual transitions has been one of the most fun and surprising ways we’ve been using Gen-3 Alpha internally,” said Runway co-founder and CTO Anastasis Germanidis.

Back in February 2023, Runway released Gen-1 and Gen-2, the first commercial and publicly available foundational video-to-video and text-to-video generation models accessible via an easy-to-use website. Now the Gen-3 update takes it to the next level.

The Power of First and Last Frames

“Gen-3 Alpha update now supports using an image as either the first or last frame of your video generation. This feature can be used on its own or combined with a text prompt for additional guidance,” Runway announced on X. 

The impact of this feature was immediately recognised by users. Justin Ryan, a digital artist, posted in response: “This is such a big deal! I’m hoping this means we are closer to the First and final frame like Luma Labs offers.”

This development puts Runway in direct competition with other players in the space, such as Luma Labs, Pika, OpenAI’s much-anticipated Sora, and the Bengaluru-based startup Unscript, which is generating videos using single images.

However, Runway’s public availability gives it a significant edge over Sora, which remains in closed testing.

A spokesperson from Runway shared that the initial rollout will support 5 and 10-second video generations, with significantly faster processing times. Specifically, a 5-second clip will take 45 seconds to generate, while a 10-second clip will take 90 seconds.  

Accelerating to Get Ahead 

Since the release of the Gen-3 Alpha model, internet users have been showcasing their unique creations in high-definition videos, demonstrating the versatility and range of Runway AI’s latest AI model.

As Runway makes a bold move, there is a significant shift in the generative AI video space. The company describes this update as “first in a series of models developed by Runway on a new infrastructure designed for large-scale multimodal training,” and a “step toward creating General World Models.”

Germanidis also revealed that Gen-3 Alpha will soon enhance all existing Runway modes and introduce new features with its advanced base model.

He also noted that since Gen-2’s 2023 release, Runway has found that video diffusion models still have significant performance potential and create powerful visual representations. 

While the startup states that Gen-3 Alpha was “trained on new infrastructure” and developed collaboratively by a team of researchers, engineers, and artists, it has not disclosed specific datasets, following the trend of other leading AI media generators that keep details about data sources and licensing confidential.

Interestingly, the company also notes that it has already been “collaborating and partnering with leading entertainment and media organisations to create custom versions of Gen-3,” which “allows for more stylistically controlled and consistent characters, and targets specific artistic and narrative requirements, among other features.”

AI comes to filmmaking

Additionally, Runway hosted its second annual AI Film Festival in Los Angeles. To illustrate the event’s growth since its inaugural year, Valenzuela noted that while 300 videos were submitted for consideration last year, this year they sent in 3,000.

Hundreds of filmmakers, tech enthusiasts, artists, venture capitalists, and notable figures, including Poker Face star Natasha Lyonne, gathered to watch the 10 finalists selected by the festival’s judges.

Now, the films look different, as does the industry with generative AI.

Meanwhile, amidst all this, it is evident that Runway is not giving up the fight to be a dominant player or leader in the rapidly advancing generative AI video creation space.

The post Is Runway’s Gen-3 Update the First or Last Frame? appeared first on AIM.

Figure Robotics’ uber-cool newest version of humanoid Figure 02 was revealed a few days ago. The company’s founder and CEO, Brett Adcock, released a demo video that showcased the robot’s dexterity and movements, predominantly showcasing its capabilities in a BMW Group Plant Spartanburg trial run. 

“Figure is giving artificial intelligence a body,” said Adcock.   

AI Gets a Body

Standing tall at 5’6” and weighing 70kg, Figure 02 is sleeker than its predecessor, with the most prominent feature being the absence of external cable for facilitating testing and repairs. The cables are now integrated onto the limbs. 

In terms of computational and AI capabilities, Figure 02 surpasses Figure 01 with three times the computational power, enabling autonomous AI tasks such as speech-to-speech interaction and visual reasoning. 

Additionally, it is equipped with six RGB cameras and an onboard vision language model, significantly improving its ability to perceive and interact with the physical world.

It’s All About the Hands

The enhanced hand dexterity of Figure 02 with 16 degrees of freedom is an impressive feat, considering that hand movements are one of the most complex parts of a humanoid manufacturing process. 

Adcock said in a post that five-fingered hands are crucial for general robotics, but traditional roboticists often hesitate due to the complexity involved. 

Given the minimal progress outside of prosthetics, the team at Figure had to design the hands from the ground up, covering everything from the mechanical structure to sensors, electronics, wiring, and control systems. 

When compared to the Figure 01, an additional degree of freedom has been added to the thumb. The bulk has been reduced from the wrist, while the finger actuators remain within the palm.

Emphasising the importance of robotics arms, in an earlier interaction with AIM, Bengaluru-based CynLr Robotics founder and CEO Gokul NA said, “Wheels are more than enough [for robots in warehouses], but you need more capability with the hands.”

Big Tech Powering Figure 02

Figure 02 could be a perfect example of embodied AI or as Jensen Huang calls it, the physical AI. Referring to it as the world’s most advanced AI, Adcock is powering Figure with the help of some of the biggest tech players in the world. 

OpenAI, Microsoft, and NVIDIA have been the prominent three powering both the hardware and software side of Figure.

Figure 02 has been built using NVIDIA’s Isaac Sim for synthetic data and generative AI model training. NVIDIA serves as a critical member in enhancing simulation, training, and inference through full-stack accelerated systems, libraries, and foundation models.

Similarly, Microsoft has allocated H100 NDs for large model training, and OpenAI has been instrumental in fine-tuning custom multimodal models on humanoid robot data. 

Along with Figure’s humanoid, many companies are already using these robots at automobile and industrial plants. Interestingly, Tesla’s Optimus Gen-2 version was released in December last year. 

Funnily, Adcock took a dig at Elon Musk after Figure 02’s launch as a challenge to the humanoids that exist. At the same time, demo videos comparing both Optimus Gen 2 and Figure 02 were doing the rounds, the latter being similar to Tesla’s.

While the friendly banter continues, big tech companies continue to focus on developing humanoids.  

The post Figure 02 the Most Advanced Humanoid AI Robot appeared first on AIM.

If only AI could create infinite streams of data for training, we wouldn’t have to deal with the problem of not having enough data. This is what is keeping a lot of things undiscoverable in the field of science as there is only a limited amount of data available that can be used for training.

This is where AI is taking up a crucial role with the help of simulation. The integration of data generation through simulation is rapidly becoming a cornerstone in the field of ML, especially in science. This approach not only holds promise but is also reigniting enthusiasm among researchers and technologists. 

As Yann LeCun pointed out, “Data generation through simulation is one reason why the whole idea of ML for science is so exciting.”

Simulations allow researchers to generate vast amounts of synthetic data, which can be critical when real-world data is scarce, expensive, or challenging to obtain. For instance, in fields like aerodynamics or robotics, simulations enable the exploration of scenarios that would be impossible to test physically.

Richard Socher, the CEO of You.com, highlighted that while there are challenges, such as the combinatorial explosion in complex systems, simulations offer a pathway to manage and explore these complexities. 

Synthetic Data is All You Need?

This is similar to what Anthropic chief Dario Amodei said about producing quality data using synthetic data and that it sounds feasible to create an infinite data generation engine that can help build better AI systems. 

“If you do it right, with just a little bit of additional information, I think it may be possible to get an infinite data generation engine,” said Amodei, while discussing the challenges and potential of using synthetic data to train AI models.

“We are working on several methods for developing synthetic data. These are ideas where you can take real data present in the model and have the model interact with it in some way to produce additional or different data,” explained Amodei. 

Taking the example of AlphaGo, Amodei said that those little rules of Go, the little additional piece of information, are enough to take the model from “no ability at all to smarter than the best human at Go”. He noted that the model there just trains against itself with nothing other than the rules of Go to adjudicate.

Similarly, OpenAI is a big proponent of synthetic data. The former team of Ilya Sutskever and Andrej Karpathy has been a significant force in leveraging synthetic data to build AI models. 

The development at OpenAI is testimony to the advanced growth of generative AI in the entire ecosystem, but not everyone agrees that they will be able to achieve AGI with the current methodology of model training. Likewise, Microsoft is also researching in this direction; its research on Textbooks Are All You Need is a testament to the power of synthetic data.

Google’s AlphaFold, which is spearheading protein fold prediction and creations for drug discovery, too can benefit immensely from synthetic data. At the same time, it can be scary to use this data for a sensitive field like science.

Synthetic Data is Too Synthetic

However, the potential of simulations extends beyond mere data generation. Giuseppe Carleo, another expert in the field, emphasised that the most exciting aspect is not just fitting an ML model to data generated by an existing simulator. 

Instead, true innovation lies in training ML models to become advanced simulators themselves—models that can simulate systems beyond the capabilities of traditional methods, all while remaining consistent with the laws of physics.

This is becoming possible with synthetic data generated by agentic AI models, which are increasing in the field of AI. Models that can test, train, and fine-tune themselves using the data they created is something that is exciting for the future of AI research. 

Moreover, the discussion around simulations also touches on broader applications. Sina Shahandeh, a researcher in the field of biotechnology, for example, suggested that the ultimate simulation could model entire economies using an agent-based approach, a concept that is slowly becoming feasible.

Despite the excitement, the field is not without its sceptics. Stephan Hoyer, a researcher with a cautious outlook on AGI, pointed out that simulating complex biological systems to the extent that training data becomes unnecessary would require groundbreaking advancements. 

He believes this task is far more challenging than achieving AGI. Similarly, Jim Fan, senior AI scientist at NVIDIA, said that while synthetic data is expected to have a noteworthy role, blind scaling alone will not suffice to reach AGI.

When it comes to science, using synthetic data can be tricky. But its generation in simulation shows promise as it can be tried and tested without deploying in real-world applications. Besides, the possibility of it being infinite is what keeps ML exciting for researchers.

The post Synthetic Data Generation in Simulation is Keeping ML for Science Exciting appeared first on AIM.

Lately, there has been a debate in India’s AI ecosystem about whether the country should build its own foundational models. Some argue that India can address real-world problems by leveraging existing state-of-the-art models without spending millions on new ones. 

Others, however, believe that it’s essential to develop models that profoundly understand the nuances, complexities, and rich diversity inherent in India’s myriad cultures and languages.

Amidst all this, an Indian state government has undertaken the task of developing an LLM that operates in the state’s official language.

In July, the Information Technology, Electronics & Communications (ITE&C) department in Telangana hosted a datathon aimed at creating a Telugu LLM.

Carried out in partnership with Swecha, a non-profit and free open-software movement in Telangana, the datathon was organised to help build datasets which, in turn, will help train the Telugu LLM.

Building Telugu Datasets

Building effective LLMs for Indian languages remains a challenging task due to the scarcity of high-quality data. While ChatGPT is impressive because it is trained on multiple terabytes of data, such extensive datasets are not available for Indian languages.

To develop datasets in Telugu, the Telangana government is tapping into its rich education system. Around 1 lakh undergraduate students across all engineering colleges in Telangana took part in the datathon and collected data from ordinary citizens who use Telugu as their mother tongue. 

The team collected data from oral sources such as folk tales, songs, local histories, and information about food and cuisine. Additionally, they plan to dispatch volunteers to approximately 7,000 villages across the state to gather audio and video samples of people discussing various topics, which were then converted into content.

(Source: @nutanc)

Interestingly, this is not the first instance when such an exercise was undertaken. Last year, the same Swecha team developed a Telugu SLM, named ‘AI Chandamama Kathalu’, from scratch. 

To collect data for the model, a similar datathon was organised with volunteers from Swecha, in collaboration with nearly 25-30 colleges. Over 10,000 students participated in translating, correcting, and digitising 40,000-45,000 pages of Telugu folk tales.

Building LLMs

Ozonetel, which is an industry partner for the project along with DigiQuanta and TechVedika, supported it by training the model and providing the necessary compute. 

The team tried fine-tuning Google’s MT-5 open-source model, Meta’s Llama, and Mistal. However, they finally settled on building a model similar to GPT-2 from scratch. Training the model on a cluster of NVIDIA’s A100 GPUs took nearly a week.

Now, the aim is to develop a larger model and have it ready to be showcased at the Telangana govt’s Global AI Summit, scheduled to take place in September this year.

Moreover, developing a large model could cost millions of dollars. For instance, building something like ChatGPT could cost in the billions. However, the team aims to develop the Telugu LLM at a cost of around INR5-10 lakh.

India’s Efforts to Build AI Models in Regional Languages 

Over time, we have seen efforts to build AI models in regional languages. For instance, Abhinand Balachandran, assistant manager at EXL, released a Telugu version of Meta’s open-source LLama 2 model.

Similarly, in April this year, a freelance data scientist released Nandi– built on top of Zephyr-7b-Gemma, the model boasts 7 billion parameters and is trained on Telugu Q&A data sets curated by Telugu LLM Labs.

Interestingly, such models have not been just limited to Telugu. We have seen AI models built on top of open-source models such as Tamil Llama, and Marathi LLama, among others.

However, these models could be seen as mere experiments. But the Telangana government’s effort to develop an AI model in Telugu has the potential to make significant strides in advancing regional language technology and preserving cultural heritage.

Officials involved in the project have told the media that voice commands from devices such as Alexa are not available in Telugu and this platform will pave the way for such innovations. 

The post Telangana Becomes India’s First State to Develop its Own AI Model appeared first on AIM.

Kuku FM, a popular audio content platform backed by Google and Nandan Nilekani’s Fundamentum Partnership, is harnessing the power of generative AI to revolutionise how stories are created, produced, and consumed. This transformation is spearheaded by Kunj Sanghvi, the VP of content at Kuku FM, who told AIM that generative AI is part of their everyday work and content creation.

“On the generative AI side, we are working pretty much on every layer of the process involved,” Sanghvi explained. “Right from adapting stories in the Indian context, to writing the script and dialogues, we are trying out AI to do all of these. Now, in different languages, we are at different levels of success, but in English, our entire process has moved to AI.”

Kuku FM is leveraging AI not just for content creation but for voice production as well. The company uses Eleven Labs, ChatGPT APIs, and other available offerings to produce voices directly.

“Dramatic voice is a particularly specific and difficult challenge, and long-form voice is also a difficult challenge. These are two things that most platforms working in this space haven’t been able to solve,” Sanghvi noted. 

In terms of long-form content moving to generative AI, Kuku FM also does thumbnail generation, visual assets generation, and description generation and Sanghvi said that the team has custom GPTs for every process.

Compensating Artists

AI is playing a crucial role in ensuring high-quality outputs across various languages and formats. “In languages like Hindi and Tamil, the quality is decent, but for others like Telugu, Kannada, Malayalam, Bangla, and Marathi, the output quality is still poor,” said Sanghvi. 

However, the quality improves every week. “We put out a few episodes even in languages where we’re not happy with the quality to keep experimenting and improving,” Sanghvi added.

Beyond content creation, AI is helping Kuku FM in comprehensively generating and analysing metadata. “We have used AI to generate over 500 types of metadata on each of our content. AI itself identifies these attributes, and at an aggregate level, we can understand what makes certain content perform better than others,” he mentioned.

One of the most transformative aspects of Kuku FM’s use of AI is its impact on creators. The platform is in the process of empowering 5,000 creators to become full-stack creative producers. 

“As the generative AI tools become better, every individual is going to become a full-stack creator. They can make choices on the visuals, sounds, language, and copy, using AI as a co-pilot,” Sanghvi said. “We are training people to become creative producers who can own their content from start to end.”

When asked about the competitive landscape such as Amazon’s Audible or PocketFM, and future plans, Sanghvi emphasised that AI should not be viewed as a moat but as a platform. “Every company of our size, not just our immediate competition, will use AI as a great enabler. AI is going to be commoditised; everybody will have access to the tools. What will remain crucial is the talent pool you have – the storytellers,” he explained.

Everyone’s a Storyteller with AI

In a unique experiment blending generative AI tools, former OpenAI co-founder Andrej Karpathy used the Wall Street Journal’s front page to produce a music video on August 1, 2024. 

Karpathy copied the entire front page of the newspaper into Claude, which generated multiple scenes and provided visual descriptions for each. These descriptions were then fed into Ideogram AI, an image-generation tool, to create corresponding visuals. Next, the generated images were uploaded into RunwayML’s Gen 3 Alpha to make a 10-second video segment.

Sanghvi also touched upon the possibility of edge applications of AI, like generating audiobooks in one’s voice. “These are nice bells and whistles but are not scalable applications of AI. However, they can dial up engagement as fresh experiments,” he said.

Kuku FM is also venturing into new formats like video and comics, generated entirely through AI. He said that the team is not going for shoots or designing characters in studios. “Our in-house team works with AI to create unique content for video, tunes, and comics,” he revealed.

Sanghvi believes that Kuku FM is turning blockbuster storytelling into a science, making it more accessible and understandable. “The insights and structure of a story can now look like the structure of a product flow, thanks to AI,” Sanghvi remarked. 

“This democratises storytelling, making every individual a potential storyteller.” As Sanghvi aptly puts it, “The only job that will remain is that of a creative producer, finding fresh ways to engage audiences, as AI will always be biassed towards the past.”

The post Kuku FM is Using Generative AI to Make Everyone a Full-Stack Creative Producer appeared first on AIM.

Recently, Meta chief Mark Zuckerberg, a crusader of open-source models, emphasised the safety aspects for the models. “There is an ongoing debate about the safety of open-source AI models, and my view is that open-source AI will be safer than its alternatives. I think governments will conclude it’s in their interest to support open source because it will make the world safer and more prosperous,” he said.  

However, open-source platforms are not immune to data threats. With the rise in cybersecurity incidents, with over 343 million victims reported just last year, the focus is back on data security, especially with AI in the picture. 

“One of the biggest challenges today is the new field of AI and finding how attackers are going to use AI to attack us. We take a step, the attackers take a step. It’s a never-ending game and the gap is minimising. I can’t predict what’s going to happen next year because this technology is developing exponentially,” said Jossef Harush Kadouri, the head of supply chain security at CheckMarx, in an exclusive interaction with AIM at the recent Accel Cybersecurity Summit. 

Cyber Attacks on Open Platforms

Kadouri, who works out of Tel Aviv, Israel, is in charge of protecting enterprises against software supply chain attacks. He has previously served in the Israel Defense Force’s cybersecurity wing for over four years. He currently ranks in the top 1% of users on Stack Overflow. 

Alluding to how packages with high ratings does not mean they are safe from malicious attacks even on platforms such as GitHub and Hugging Face, Kadouri said, “Now, if you ever thought that popular packages are safer, not necessarily. 

Attackers focus on those assets to deliver immediate attacks,” he said, emphasising on typosquatting, where Python developers are targeted through registered misspelled versions of popular packages such as Selenium. 

“Once we investigated the actual code executed in this malicious package, this code was highly cryptic, obfuscated, hard to read and understand, and it’s executed upon installation,” warned Kadouri. Over 900 packages containing obfuscated codes that execute upon installation were revealed. 

Hugging Face is the Disneyland of Open-Source Models

Popular developer platforms such as GitHub and Hugging Face have not been free of all kinds of threats. Though Hugging Face is taking active measures to prevent backdoor threats, the platform is susceptible to model-based attacks. 

“Hugging Face is like the Disneyland of LLM, open-source models, and pre-trained models,” said Kadouri.

Various forms of cyber attacks are continuously taking place on these platforms. Malicious browser extensions are another common route through which hackers syphon off money. 

In the context of cryptocurrency transfers, users typically copy and paste wallet addresses to avoid errors. It was revealed that a malicious browser extension could alter the copied address, potentially redirecting crypto funds to a different wallet. 

“This is how sophisticated the attacks are. You would only realise it once it’s too late. You can’t undo a crypto transaction,” said Kadouri. 

Cyber Threat Awareness

With the rising cyber attack cases, one of the biggest needs of the hour is awareness. “I think what we need to do is educate and raise awareness that we have bad guys operating in this attack surface,” said Kadouri, who supports the whole open platform that enables developers build products, however, is wary of the risks that are apparent. 

“It’s a good thing to do, but they [platforms] also don’t vet the content they host. So this is why we need to stay alert from things that may look legitimate, but are not. Because, anyone can contribute fresh new content to open source and disguise it as something that is well worth it,” he said.  

Interestingly, Rahul Sasi, the co-founder and CEO of CloudSEK, an AI-powered digital risk management enterprise, reflected similar sentiments. 

Speaking about the recent Indian telecom operator whose user data was hacked (without taking names), Sasi mentioned that the companies don’t acknowledge it, which is a problem that hinders cybersecurity awareness. 

“I mean, the problem with this company is that they also don’t understand. Or many times the security teams understand, but then there is high pressure on the top management not to accept it,” said Sasi, in an exclusive interaction with AIM. 

“Things have improved in the last 10 years. But, it hasn’t reached where it should. In my perspective, maybe in another 10 years it hopefully will. The media also has a role to play here. If you try to blame somebody, they’ll always try to defend,” he said. 

With AI in the cybersecurity scene, optimism is on the higher end. However, Kadouri doesn’t completely believe so. 

AI in Cyberattacks  

Speaking about how AI has added to cyberattacks through deep-fake technology, for instance, Kadouri still “wants to believe” that AI might be a problem-solver too. 

“I can definitely see AI helping us defenders do our jobs better, reduce manual labour and automate things. But, if they’re [cyber attackers] so good at fooling human beings, they’re probably going to be good at fooling AI too,” he said. 

“I mean, time will tell,” he concluded. 

The post The Hidden Risks in Open-Source AI Models appeared first on AIM.

Kuaishou Technology, a Chinese AI and technology company, launched a new text-to-video model called Kling this year. Su Hua, a Chinese billionaire internet entrepreneur, is the co-founder and CEO of the video platform Kuaishou, known outside of China as Kwai.

Several AI enthusiasts shared their creations from Kling on X that captured the hearts of internet users worldwide. A series of animals and objects were featured enjoying a meal of noodles. From a panda munching on a bowl of ramen to a kangaroo slurping up some udon, the videos are both hilarious and heartwarming. 

A few others include blueberries turning into puppies and a tray of apples turning into guinea pigs, which mess with your head.

The level of detail and realism in the videos is a testament to the capabilities of Kling and the progress made in the field of AI.

Known for the creation of TikTok competitor, Kuaishou joined the race with other Chinese tech companies to rival OpenAI’s Sora. 

With simple text prompts, it can generate highly realistic videos in 1080p high-definition resolution. The videos can be up to two-minutes long. Sora, on the other hand, makes 60-second videos with text prompts.

Kling boasts the ability to produce realistic motions on a large scale, simulate the attributes of the physical world, and weave concepts and imagination together setting a new benchmark in AI-powered video creation. 

However, as impressive as Kling AI may be, its accessibility is primarily limited to a few users even though the company claimed it would be available worldwide. This poses significant challenges for its global adoption.

For some users who were looking forward to accessing its offerings, this situation may feel like a huge letdown. 

A worldwide release creates expectations of inclusivity and accessibility; when these are unmet, it can harm the company’s reputation. Despite Kling AI’s impressive features, its primary hurdle is limited availability. 

Currently, its access is mostly limited to invited beta testers, with some users in China able to experience a limited demo version through the Kuaishou app, as claimed by ChatGPT on Quora. 

At a time when the US is heavily debating AI ethics and incorporating ‘Responsible AI’, China seems unperturbed and is likely responding to these AI ethicists with a Kling. 

The AI company hit the headlines recently by announcing the global launch of its International Version 1.0, a platform designed to revolutionise industries worldwide. This milestone release features advanced machine learning, multilingual support, and enhanced data analytics, promising unparalleled efficiency and innovation across sectors. 

AI Video Generator War Begins! 

While systems like OpenAI’s Sora and Kuaishou’s Kling have showcased impressive capabilities, they remain accessible only to a select group of users. Similarly, Luma AI’s Dream Machine also boasts remarkable features but is limited to a restricted audience.

Interestingly, Kuaishou’s AI tool entered the market shortly after Vidu AI, another Chinese text-to-video AI model known for producing HD 1080p 16-second videos.

This model’s launch coincides with a flurry of activity in the generative AI sector, as startups and tech giants compete to develop advanced tools that create realistic images, audio, and video from text inputs.

It has a user-friendly interface that supports text-to-video or text-to-image generation. 

Unlike Runway, Haiper, and Luma Labs, it prompts up to 2,000 characters, enabling highly detailed descriptions. It performs better with lengthy, well-crafted prompts.

This cutting-edge AI model employs variable resolution training, enabling users to produce videos in various aspect ratios. Remarkably, it can showcase full expression and limb movement from a single full-body image. 

AI video creation seems like the next battleground for tech companies with contenders like OpenAI’s Sora, Microsoft’s VASA-1, Adobe’s Firefly, Midjournery, and Pika Labs, already in the game. 

Furthermore, Google recently introduced Veo, a new text-to-video AI model, at Google I/O to compete with OpenAI’s Sora. Veo improves on previous models, offering consistent, high-quality over-a-minute-long 1080p videos.

While some were impressed with Veo’s capabilities, others argue that it may not exactly be state-of-the-art in its latency or abilities compared to Sora.

Now that Kling is here, the benchmark of making cinematically impressive and real-world-like videos has gone up. 

Why is Kling a big deal?

This month, Runway introduced Gen-3, which offers enhanced realism and the ability to generate 10-second clips. Last month, Luma Labs unveiled the impressive Dream Machine. 

These new model updates were initially spurred by the release of Sora earlier this year, which remains the benchmark for AI video generation. Recently, a series of short films on YouTube showcased Sora’s full potential. Additionally, Kling played a significant role in the wave of updates.

It also adopts a unique approach to AI by incorporating generative 3D in its creation process. It provides Sora-level scene changes, clip lengths, and video resolution. Given that OpenAI only grants a limited number of select creators access to Sora, Kling AI might just be the top choice for now.

Capabilities of Kling AI

Kling AI is accessible via the Kuaishou app, available on both iOS and Android platforms. This mobile app puts Kling AI’s advanced video generation capabilities directly at users’ fingertips, enabling them to create high-quality, realistic videos from their smartphones.

For users outside China, accessing Kling AI often requires navigating around these barriers. Some have resorted to emailing Kuaishou directly to request access, explaining their interest in becoming beta testers. 

The competitive landscape is evolving, but the restrictions on access can hinder Kling’s ability to gain traction outside China.

Chinese attempts to lure domestic developers away from OpenAI – considered the market leader in generative AI – will now be a lot easier, after OpenAI notified its users in China that they would be blocked from using its tools and services. 

“We are taking additional steps to block API traffic from regions where we do not support access to OpenAI’s services,” said an OpenAI spokesperson.

OpenAI has not elaborated about the reason for its sudden decision. 

ChatGPT is already blocked in China by the government’s firewall, but until this week developers could use virtual private networks to access OpenAI’s tools in order to fine-tune their own generative AI applications and benchmark their own research. Now the block is coming from the US side.

The OpenAI move has “caused significant concern within China’s AI community”, said Xiaohu Zhu, the founder of the Shanghai-based Centre for Safe AGI, which promotes AI safety, not least because “the decision raises questions about equitable access to AI technologies globally”.

The post Kwai’s Kling vs OpenAI Sora  appeared first on AIM.

When asked about our favourite apps, most of us would name Zomato, Swiggy, or Zepto. And it’s pretty obvious why. Well, craving food when stressing over the looming possibility of a 14-hour workday is completely justified. 

But if you think Zomato and Swiggy are just about delivering food or booking tables at fancy restaurants, it’s time to rethink.

A YouTuber from the channel Full Disclosure recently interviewed food delivery riders and revealed that some of these riders earn between INR 40,000 and INR 50,000 per month, which is more than the average income of many IT professionals.

One rider even mentioned that he managed to save INR 2 lakh in just six months.

In another interesting scenario, Venkatesh Gupta, a techie, shared on X a strange encounter. He met a senior Microsoft engineer in Bengaluru driving an auto rickshaw over the weekends to fend off loneliness.

Every Social Media Platform is an eCommerce Store

Most social media sites, including Facebook, Twitter, Instagram, and WhatsApp, are striving to become e-commerce platforms. 

Every platform you visit is working hard to enable you to buy things without leaving their site. TikTok, for example, changed its ‘storefront’ to ‘shop’ so you can make purchases directly on the app. No more jumping to other websites.

This isn’t just TikTok. YouTube allows creators to sell products in their videos. Pinterest has ‘buyable pins’ so you can shop without leaving the site. Facebook has stores, even on Messenger. Google is also in the game with Google Business.

WhatsApp was launched as a one-to-one chat app service in February 2009 and now offers Business Accounts with payment options. 

WhatsApp for e-commerce allows customers to complete the entire transaction on one app without needing to shift platforms. This can be done by integrating the product catalogue with WhatsApp Business. 

Customers can then browse products, view prices, and make purchases directly within the WhatsApp interface. 

Unexpected Chat Apps

In recent years, many applications initially designed for specific purposes have added messaging capabilities. When you break up with your boyfriend and think you’ve blocked him everywhere, including Instagram, WhatsApp, and Facebook, he can still text you on GPay! 

Instagram, originally a photo-sharing app, now offers direct messaging (DM) to allow users to chat, share media, and even engage in group conversations. It also functions as an app that can make you doubt yourself and give you an inferiority complex (maybe). 

Coming back to Google Pay, which was initially just a payments app, now includes messaging features to facilitate communication around transactions, share payment details, and more.

There’s also Spotify, which allows users to share songs and playlists through integrated messaging services.

Super Apps on the Rise

In 2022, the global super apps market was valued at a staggering $61.30 billion, with a growth trajectory of 27.8% expected from 2023 to 2030. Gartner’s survey reveals that the top 15 super apps have been downloaded over 4.6 billion times worldwide, boasting 2.68 billion monthly active users.

By 2050, it is anticipated that more than half of the global population will be using super apps. Well, the key to their widespread adoption lies in the mobile-first market, where smartphones are the primary connected devices. 

Another major contributor is their integration of financial and payment services. Paytm in India, Grab in Singapore, Goto in Indonesia, and Zalo in Vietnam are a few examples, each pivoting their user experience around robust financial and banking services.

Even apps like Cred, which is the credit card bill payment platform, are joining the UPI payment club competing against Paytm and PhonePe.  

Earlier ​​Deepinder Goyal, the CEO of Zomato had said that super apps don’t work in India. He would rather have Zomato and Blinkit grow as separate brands.

Maybe it’s time to rethink. 

The post Accidental Super Apps appeared first on AIM.

In a strategic move to bolster its generative AI capabilities, Canva has acquired Leonardo.Ai, a startup renowned for its generative AI content and research. Canva co-founder and chief product officer Cameron Adams has said that all 120 employees of Leonardo.ai, including the executive team, will join Canva. The acquisition is said to be a mix of cash and stock.

Leonardo.Ai, co-founded by Jachin Bhasme, JJ Fiasson, and Chris Gillis in Sydney in 2022, initially focused on creating video game assets. Over time, the company expanded its platform to cater to diverse industries such as fashion, advertising, and architecture by developing AI models for image creation. Some people call it the biggest competitor to Midjourney.

The company had previously raised over $38.8 million from several backers such as Smash Capital, Blackbird, Side Stage Ventures, Gaorong Capita, Samsung Next, and TIRTA Ventures.

Today, Leonardo.Ai boasts over 19 million registered users and has facilitated the creation of more than a billion images. It provides collaboration tools and a private cloud for various models, including video generators. It also offers API access, enabling customers to develop their own technological infrastructure using Leonardo.Ai’s platform.

A Game Changer?

Despite the acquisition, Leonardo.Ai will continue to operate independently, prioritising rapid innovation and research, now backed by Canva’s resources. 

“We’ll keep offering all of Leonardo’s existing tools and solutions. This acquisition aims to help Leonardo develop its platform and deepen its user growth with our investment. This includes expanding their API business and investing in foundational model R&D,” said Adams.

Leonardo.Ai sets itself apart from other generative AI art platforms by providing extensive user control. Features on Live Canvas allow users to input text prompts and make quick sketches, generating photorealistic images in real-time. 

However, the methods Leonardo.Ai uses to train its in-house generative models, such as the flagship Phoenix model, remain unclear. This can be challenging for Canva to figure out later.

Canva has been a strong supporter of creators in the generative AI space. The company paid $200 million to compensate creators who allowed their content to be used for training AI models. The acquisition of Leonardo.ai will contribute to Canva’s Magic Studio generative AI suite, enhancing existing tools and introducing new capabilities.

“Magic Studio works on internally-developed AI and ML algorithms that leverage a combination of foundational AI models from our team, including Kaleido, and a variety of partners like OpenAI, Google, AWS, and Runway,” Danny Wu, the head of AI products at Canva, told AIM in a conversation earlier this year.

Canva Hell-Bent on Generative AI

Adams expressed excitement about integrating Leonardo’s technology into Magic Studio. “We’re eager to expand what our users can achieve with AI on Canva,” he said. 

Canva has been ramping up its AI development efforts, highlighted by previous acquisitions such as Kaleido in 2021, which laid the groundwork for many of Canva’s recent AI advancements, which is also looking for an IPO soon.

Leonardo.ai is Canva’s eighth acquisition overall and its second this year, following the $380 million acquisition of UK-based design company Affinity. Canva’s robust portfolio includes presentations startup Zeetings, stock photography sites Pixabay and Pexels, and product mockup app Smartmockups.

“We’ve placed a strong focus on building an AI-powered workflow that includes generative solutions like image and design generation,” said Adams. He added that new generative capabilities will help the company set its AI offerings apart.

Meanwhile, competitors such as Figma and Adobe are also making strides in generative AI. Just last month, Figma introduced Figma AI, a suite of AI-powered features to enhance designers’ creativity and productivity.  

Adobe has also diversified its portfolio into a generative AI-powered enterprise software platform, introducing Firefly to Photoshop and launched features like Generative Fill and Generative Remove for advanced image editing.

The updates follow Adobe’s failed 20 billion dollar acquisition of Figma due to antitrust scrutiny last year. Now, taking matters into its own hands, the company is reinventing itself with AI. With AI products in the pipeline and a redesigned interface, Figma is gearing up to compete with major players like Adobe and Canva.

The post Why Canva Acquired Leonardo.Ai appeared first on AIM.

Around 241.8 million units of personal computers were sold all across the world in 2023. Despite this, it was the worst year on record for PC sales, with a nearly 15% decline compared to the previous year, according to Gartner.

In India too, last year, the PC market declined by 6.6%. Low market sentiments post-pandemic, supply chain constraints and geopolitical tensions contributed to the decline; but now PC makers are hoping generative AI could help alter their fate.

The top PC makers in the world have been quick to ship AI-powered PCs in most markets. Acer, which has a relatively small portion of the PC sales market, witnessed a 12.3% increase in sales in 2023, the highest among all.

Acer too has already launched a series of AI PCs that are also available in the Indian market which comes with built-in AI features. 

New AI PCs Coming Soon to Indian Market 

In an interview with AIM, Sudhir Goel, chief business officer at Acer India said, “At Computex 2024, we have showcased a lot of new products, which we are thrilled to introduce to the Indian market in the coming year.”

So far, Acer has launched the Acer Swift 14 AI PCs, the TravelMate series, and its Predator Helios AI gaming laptops in India.  

“With Swift AI, all functionalities, ranging from image enhancement to voice processing, are performed locally, thanks to the neural processing unit integrated within our laptops. This ensures an elevated level of privacy and security, paramount for individual and corporate users,” Goel said.

The TravelMate business PCs are equipped with sophisticated enterprise-grade AI security, and they will soon be available in India. These laptops feature advanced AI tools, including Acer LiveArt and the GIMP with Intel’s Stable Diffusion plugin. 

It also leverages NPU for AI-accelerated applications to blur the background, automatically frame, and maintain eye contact during video conferencing. Moreover, AI will optimise power consumption during long conferencing calls.

“In the gaming sector, AI integration will redefine immersive gaming by enabling real-time map generation and dynamic creation of in-game elements based on live data. Additionally, we will introduce advanced AI-driven monitors to elevate the overall user experience,” Goel revealed.

The Predator Helios 16 laptops are already available in the Indian market. However, the true advantage of AI emerges when an AI model can be run locally on the device. Given the gargantuan sizes of these Large Language Models (LLMs), they can only run on the cloud.

“With Acer’s cutting-edge Neural Processing Units (NPU), our laptops can handle LLM tasks directly on the device. As we look to the future, we envision local LLM capabilities becoming a significant differentiator in the market,” Goel revealed.

Acer Laptops Will have New AI Processors 

The Acer Swift AI PCs come with Qualcomm’s Snapdragon Elite processors. However, Acer plans to offer a range of new processors in its upcoming laptops. 

“While Snapdragon’s latest technology offers remarkable capabilities, we are also exploring options from Intel and AMD. Our strategy is to evaluate and incorporate processors from all these leading providers based on their strengths and innovations. This approach ensures that our laptops can cater to a wide range of needs, from exceptional performance and efficiency to specialised AI features,” Goel said. 

The TravelMate P6 14 laptop features Intel Core Ultra 7 processors with Intel vPro Enterprise, Intel Graphics, and Intel AI Boost.

Will AI PCs Boost the Market?

While Acer’s introduction of AI-capable PCs is impressive, other PC manufacturers have swiftly followed suit. Dell and HP have also released AI-powered PCs in the Indian market this year. Most recently, Microsoft unveiled its Surface AI PCs in India, featuring Snapdragon Elite Processors.

PC makers are hoping AI could help pull the market from the stalemate that it was last year. Research firm Canalys predicts that the PC market will see an 8% annual growth in 2024 as more AI PCs hit the market. Canalys also predicts AI PCs will capture 60% of the market by 2027. 

Goel also believes generative AI has the potential to significantly boost laptop sales. “Over the past few years, the PC industry has been striving to make devices more powerful, efficient, thinner, and lighter. However, it lacked a transformative technology that could truly revolutionise the market. With the advent of AI, this missing piece has finally arrived,” he said.

AI-driven workloads in PCs could enhance performance, enable new functionalities, and create a more seamless user experience. PCs makers are desperately banking for this to happen.

Acer, too Turns to Server Business and Consumer Electronics

Earlier this year, Acer also launched its consumer electronics and home appliances brand, Acer Pure in India. When we asked Goel whether it is a result of declining PC sales, he said, “Acer India is the fastest-growing PC brand in India, and we have seen remarkable YoY growth for our PC business with 2X growth in the consumer market and market leadership in some of the commercial segments.”

He stressed that PCs remain Acer’s core business. Interestingly, Acer also launched its server business in India a few years back, a segment dominated by Dell, HP, and Lenovo, other PC brands Acer competes with.

Called Altos Computing, it caters to the growing demand for high-performance servers and workstations in India’s digital infrastructure landscape. 

“It includes introducing AI-powered solutions to support local cloud and data storage initiatives, which are crucial for governmental and corporate digital transformation priorities,” Goel concluded.

The post Futuristic Acer AI PCs Coming Soon in Indian Market appeared first on AIM.

Some might argue that generative AI has overshadowed blockchain, while others might maintain the opposite. But the truth is, both these technologies boast intriguing use cases, and what we are witnessing now is a convergence of the two.

XinFin, the creator of XDC Network, an enterprise-grade, open-source blockchain protocol aimed at transforming global trade and finance, has introduced CodeRun.ai, a coding tool tailored for use with the XDC Network.

“CodeRun.ai is optimised for accuracy in blockchain-related applications. Therefore, anyone looking to develop on the blockchain can simply issue a command and receive a complete code base instantly,” Ritesh Kakkad, a co-founder of XDC Network, told AIM.

It is specifically optimised for the XinFin XDC blockchain network and seamlessly integrated into the broader XDC ecosystem, aiming to foster developer adoption and broaden its range of applications.

The network is tailored for enterprise applications, including supply chain management, trade finance, and other business-critical processes. It aims to enhance operational efficiency and reduce costs for enterprises.

CodeRun is Powered by OpenAI

Atul Khekade, another co-founder of XDC Network, explained to AIM that CodeRun.ai draws inspiration from GitHub Copilot, the world’s most widely used AI tool for developers. However, fundamentally, both serve distinct purposes.

“I was, in fact, among the first 20 users of GitHub Copilot. It is more of a code-generation tool, and fundamentally, we have architected CodeRun.ai differently,” he said.

While GitHub Copilot helps in writing codes, CodeRun.ai is designed to help build applications on blockchain.

It utilises OpenAI‘s APIs within what the startup calls an ‘AI aggregator’, which is also powered by a proprietary XDC Network model. 

“This setup allows us to leverage databases such as ChromaDB and Pinecone in a local environment tailored to our specific needs and use cases. The aggregator then integrates broader intelligence from OpenAI, refining XDC Network’s enterprise data to meet precise requirements,” Khekade said.

Explaining further, he said when an enterprise runs CodeRun.ai in their local environments, be it a bank or a large aviation player, they benefit from enhanced privacy protection. 

“Your specific data and configurations are securely handled, with generative data staying on your own servers, never exposed externally.

For example, in the airline industry, CodeRun.ai can optimise flight scheduling or pricing algorithms. 

“It might suggest adjustments that allow you to charge customers 10% less while still increasing profitability. This level of precision and customisation is unmatched by other tools,” Khekade added.

The model has been fine-tuned with XDC Network’s 103-gigabyte proprietary data. Moreover, the startup also plans to release its proprietary model soon.

Enabling Applications Building on Blockchain

XDC Network, which was founded in 2017, supports the creation and execution of smart contracts and decentralised applications (dApps), enabling automated and transparent business processes. 

The network is tailored for enterprise applications, including supply chain management, trade finance, and other business-critical processes.

Kakkad said XDC Network foresees a trend where hundreds and thousands of applications are going to be built on a blockchain. CodeRun.ai enables not just blockchain developers or enterprises but someone with very limited knowledge about these technologies to build applications. 

The startup also intends to onboard over 10,000 active developers within the first year. For this, it’s already in talks with incubators, most notably T-Hub, which is based in Hyderabad and has the largest incubator facility in the world.

“We started with almost a hundred users and these are all developers who are building on XDC Network. So, this was our testing period and since the launch, the demand has skyrocketed. Now we are going for an incubator strategy,” Khekade said.

XDC Network recently signed a deal with Plug & Play, which is a global innovation platform and one of the world’s largest startup accelerators. They operate in over 50 locations worldwide, running more than 60 industry-focused accelerator programs annually. 

“The target is to implement at least 1,000 projects or applications utilising the backbone of the XDC Network, addressing use cases or issues identified by enterprises or startups. 

“This initiative is expected to significantly increase the usage of Coderun.ai, as many developers and builders will be leveraging it for their projects,” Khekade pointed out.

Besides Plug & Play, the startup is also partnering with at least five more incubators, some of which are from Hong Kong and Singapore. “They specialise in building large-scale market applications and are very successful in their portfolio,” Khekade concluded.

The post Now AI Can Help You Write Codes for Blockchain appeared first on AIM.

“India must urgently prepare itself for the quantum revolution to safeguard its security and national interests,” said Ajai Chowdhry, the co-founder of HCL and chairman of the mission governing board for India’s National Quantum Mission (NQM).

Announced in 2023, NQM aims to invest INR 6,000 crore to accelerate its quantum technology capabilities. Quantum computing is often seen as the Holy Grail of modern computing due to its potential to revolutionise data encryption and processing speeds, making traditional encryption methods obsolete.

For this, the government recently announced a collaboration with major IT services firms like TCS, HCL, and Tech Mahindra to develop quantum technologies under the INR 6,000 crore scheme. This initiative will support Indian startups and scientists in the emerging quantum field. 

“We want them to do research on quantum technologies because there’s a huge amount of algorithms required for quantum,” said Chowdhry.

A report by Itihaasa Research and Digital, co-founded by former Infosys CEO Kris Gopalakrishnan, highlights the global landscape of quantum investments. The top 12 countries have collectively invested around $38.6 billion, with China leading the way at $15 billion. 

In contrast, India’s investment stands at a mere $0.74 billion. 

“China has an enormous number of publications in the area of quantum technology,” pointed out Ajay K Sood, principal scientific advisor to the government of India. Tech giants like Google, IBM, and Intel have collectively invested billions in quantum computing.

While India has published only 1,711 research papers on quantum, China published 12,110, which is seven times more than India. Moreover, India has 82,110 graduates in quantum technology while China produced 57,693. 

“The elephant in the room is that … [other than] a few top tier institutions there is a scarcity of faculty to train students in quantum technologies,” read the recent report.

Private Sector Needs to Step Up

“Can China break the cryptography that India employs? That is a point of worry,” noted Chowdhry. The gap between the two nations is evident, with China being at least five years ahead of India in quantum. 

India’s largest quantum computer, with 6 to 7 superconducting qubits, is being developed at Mumbai’s Tata Institute of Fundamental Research (TIFR), in collaboration with DRDO and TCS. “With the launch of the National Quantum Mission, India is gearing up to develop not only quantum software but also state-of-the-art quantum computing hardware,” said Rajamani Vijayaraghavan, associate professor at TIFR.

India aims to develop 100-qubit computers within the next five years. 

IIT Bombay’s partnership with TCS to develop the country’s first Quantum Diamond Microchip Imager is a significant step forward. This tool will enhance the precision in examining semiconductor chips, reduce chip failures, and improve the energy efficiency of electronic devices.

The plan is to mentor and seed-fund 50 startups already working in this field. “The next thing that we are working on is to get startups involved. There are close to 50 startups already working in this area. We will provide them mentorship through these thematic hubs. We will also look at providing them some initial seed funding,” said Chowdhry.

India is Not Starting From Scratch

The VP of IBM Quantum, Jay Gambetta, recently said that India has the second-highest open access to quantum computing, which is about 77,000. He said that India has the opportunity to become the leader in quantum computing in areas such as energy, sustainability, and agriculture, among others.

To bolster this, ISRO also has a partnership with Raman Research Institute for building quantum communication technology, which is still under process. 

“A bank or an electrical grid in India can be attacked by an adequate quantum computer sitting in China. We must start working on making our country quantum secure. 

“We will work with different agencies in the government to make them aware that something like this has to be done. Alternatively, the RBI should start working on creating a policy to make all banks secure on this front,” said Chowdhry.

India’s quantum mission includes plans to develop its own quantum computers. “In the period that we don’t have a quantum computer, we’ll buy a few for research work. But we are not going to use quantum computers only on the cloud because they are very expensive,” Chowdhry added.

The post India Needs to Boost its Quantum Mission appeared first on AIM.

The world of 3D model creation is extremely capital-intensive, and India barely has any company building this. As indie game development gains prominence, there is a dire need for an Indian company to enter this space.

Recognising this void, Google for Startups Accelerator: AI First Program has decided to support 3DAiLY, a 3D model creation company that makes ultra-realistic production-ready assets using generative AI. The company is also one of the first to make these models accessible on its community platform for AAA and indie games alike.

Speaking with AIM, the CEO and founder, Harsha P Deka, shared his journey, which began in 2010 with a personal loss that fueled his drive to innovate. While studying computer science in Canada, Deka received the devastating news of his best friend’s death. 

“I wanted to create a 3D model of my friend to give to his family as a memento,” Deka recalled. 

However, he soon discovered the limitations of the technology at the time. Despite reaching out to multiple gaming and animation studios, none could produce a 3D model from a photograph.

The Birth of 3DAiLY

Undeterred, Deka delved into the intricacies of 3D modelling, understanding the immense time and effort required to create high-quality models. By 2014, he had founded an animation studio, encountering firsthand the industry’s challenges. 

“Creating a game, Belegon, took us a year and made us realise the massive funding needed for such ambitious projects,” Deka explained. His experiences underscore the complexities of 3D modelling, particularly for animation and gaming.

A pivotal moment came in 2015 when Deka encountered a 3D scanning setup in a US mall, which produced 3D-printed miniatures from in-person scans. “I asked if they could do it from a photograph, and they said it wasn’t feasible,” Deka said. 

This gap in the market inspired him to create 3DAiLY, a company that could generate 3D models from photos. By leveraging AI, Deka set out to make high-quality 3D modelling accessible and efficient.

3DAiLY has since evolved into a leader in the 3D modelling industry, creating a comprehensive library of human models and developing proprietary AI technology. “We built our own foundation model using the data we’ve collected over the years,” Deka stated. 

Its approach combines artists’ intuition with AI, ensuring that the models are production-ready and of the highest quality. 

Deka said that unlike other AI tools such as Ready Player Me or Sloyd that produce low quality meshes, 3DAiLY’s models are fully rigged and animatable, compatible with various gaming engines like Unity, Unreal, and CryEngine. 

“What MetaHuman did for Unreal, we’re doing for multiple engines,” he emphasised, adding that the rest of them are building tools and not a platform which has a built-in marketplace.

Overcoming Industry Challenges

Establishing such a pioneering company in India has not been without its challenges. “The space is not well understood, and getting VC funding is tough,” Deka noted. Despite these hurdles, 3DAiLY has garnered significant traction, with indie game developers and AAA studios alike adopting their technology. 

Offering models at a fraction of the traditional cost, the company is breaking down barriers to high-quality 3D content creation.

Looking ahead, Deka envisions expanding 3DAiLY’s capabilities to include design-to-3D modelling tools, enabling artists to transform their ideas into tangible models. 

“We’re building an ecosystem where artists can create assets and participate in an SDK, benefiting from in-game asset sales,” he explained. Currently, the platform has around 12,500 artists from 150 countries. 

“Artists are critical to the success of games, yet they often earn the least,” Deka pointed out. By offering tools that significantly reduce production time and costs, 3DAiLY aims to empower artists, allowing them to focus on creativity while the AI handles the heavy lifting. 

This approach not only enhances productivity but also ensures that high-quality 3D models are accessible to all. 

The post This Indian AI Startup is Creating 3D Models for AAA and Indie Games appeared first on AIM.

Globally, India has the second-largest number of residents who use AI (81%). Further, Asian countries tend to display greater subjective knowledge of AI, with South Korea, China, and India leading the way.

“India offers one of the most relevant and exciting contexts for AI, considering the readiness within the country to adopt various technologies, including IoT and AI, alongside other foundational technologies,” said Shekar Sivasubramanian, the CEO of Wadhwani AI, in an exclusive interview with AIM. 

Further, he said that the country’s ability to skip extensive wired network development and move straight to wireless infrastructure demonstrates its readiness and adaptability.

Sivasubramanian also highlighted that for AI to succeed on a large scale, access to diverse data, which is abundantly available in India, is required. 

While we are at the early stages of harnessing this data, the country’s multilingual and multicultural environment provides a unique opportunity to develop context-centric, powerful AI solutions. 

“If AI works in India, it can work anywhere. This is because the diversity within the population acts as a natural protection against bias, allowing for more accurate and representative AI models,” added Sivasubramanian. 

Sivasubramanian is an industry leader who brings 40 years of global applied technology and management experience towards creating positive and sustainable impact at scale. 

He is currently driving non-profit organisation Wadhwani AI’s efforts toward establishing AI-driven solutions and ecosystems for the benefit of millions across the developing world. 

Voice is the Future

“In the next 5-10 years, we would completely stop using the thumb and use only voice-based interfaces,” the Wadhwani AI CEO predicted. 

He noted that to build meaningful voice-based interfaces, embracing the diversity of expression in India is crucial. And India’s varied linguistic and cultural landscape provides a unique testing ground for developing advanced voice AI technologies.

Along similar lines, Pramod Varma, former chief architect of Aadhaar, told AIM, “Indian entrepreneurs should really look at voice as a completely new human-computer interaction method. It could be very powerful, and I think it’s going to happen because voice is natural to humans.” 

In a previous interaction with AIM, Sarvam AI also mentioned that it is currently working on a voice-based Indic LLM, which it plans to release this year.

Earlier, chief AI scientist at Meta, Yann LeCun had said that in the next 10-15 years we won’t have smartphones, and will be using augmented reality glasses and bracelets to interact with intelligent assistants. 

“The last thing we might want is intelligent virtual assistants that help us in our daily lives. So today, all of us here are carrying a smartphone in our pockets; 10 years from now or 15 years from now, we’re not going to have smartphones anymore. We’re going to have augmented reality glasses,” said LeCun.

Even Meta CEO Mark Zuckerberg has repeatedly stressed that neural interfaces represent the inevitable next step beyond current methods like typing on screens.

Wadhwani AI Initiatives

India grows 26% of the world’s cotton and nearly 100 million farmers rely on cotton farming for their livelihood. However, cotton is highly vulnerable to pests, causing yield uncertainty and financial distress to farmers. 

So, to help farmers protect their crops, Wadhwani AI has introduced CottonAce, an AI-powered early warning system available as an Android app. 

Lead farmers, who work with welfare programs, use the app to upload photos of pests and the AI algorithm analyses the photos, determines infestation levels, and provides actionable advice, which is then shared with neighbouring farmers, even those without smartphones.

The app is available in nine languages, including English, Hindi, Marathi, Gujarati, Telugu, Kannada, Tamil, Odia, and Punjabi. 

In the health industry, the non-profit institute developing AI solutions for social good is an official AI partner of the Central TB Division (CTD), and are developing multiple interventions across the TB care cascade and helping India’s National TB Elimination Programme become AI-ready.

They use AI to interpret the results of the LPA test to determine drug resistance to TB. Each LPA strip encodes the drug-resistance pattern of the patient via a series of activated (dark) and inactivated (light) bands corresponding to different regions of the genome of the Tuberculosis bacterium. 

Further, it is developing multiple AI solutions to reduce morbidity and mortality for mothers and children in low-resource settings by improving the quality of primary care and strengthening the first 1,000 days of life.

The post ‘If AI Works in India, it Can Work Anywhere’ appeared first on AIM.

Not everyone can access highly spec’d machines capable of running LLMs locally, which often require substantial computational power and memory. 

“GPUs like H100s, which are essential to train and run LLMs efficiently on a large scale, are beyond the budgets of most startups. And running models like Llama 3.1 405B is unthinkable for regular people. 

“Renting GPUs and running them on a single cluster or using peer-to-peer connections is one of the easiest ways to do it,” Arjun Reddy, the co-founder of Nidum.AI, told AIM.

P2P technology is already used in blockchains, which is a testimony to how secure the network can be. P2P technology came into the limelight for the first time in 1999, when Napster used P2P technology to decentralise music, allowing users to download and host music files from their own computers.

Reddy further explained the approach they follow for the P2P technology. It starts with fine-tuning the existing model for specific needs, which is then divided into hundreds of small parts and described to the P2P network. 

A layer of encryption is used to safeguard data. 

To showcase the flexibility of P2P technology, Reddy is about to host the largest decentralised AI event later this week where hundreds of Apple computers will be used to run Llama 3.1 through the P2P network. The idea is to demonstrate the importance of decentralised networks to run LLMs. 

The Promise of Peer-to-Peer Network

P2P networks, popularised by file-sharing systems like BitTorrent, distribute tasks across multiple nodes, each contributing a portion of the overall workload. 

Applying this concept to AI, a P2P network could theoretically distribute the training of an LLM across numerous consumer-grade GPUs, making it possible for individuals and smaller organisations to participate in AI development.

A research paper titled ‘A Peer-to-Peer Decentralised Large Language Models’ discusses a provably guaranteed federated learning (FL) algorithm designed for training adversarial deep neural networks, highlighting the potential of decentralised approaches for LLMs.

A study by Šajina Robert et al. explored multi-task peer-to-peer learning using an encoder-only Transformer model. This approach demonstrated that collaborative training in a P2P network could effectively handle multiple NLP tasks, highlighting the versatility of such systems.

Another significant contribution comes from Sree Bhargavi Balija and colleagues, who investigated building communication-efficient asynchronous P2P federated LLMs with blockchain technology. Their work emphasises the importance of minimising communication overhead and ensuring data integrity in decentralised networks.

But There are Challenges… 

Despite the promise, significant challenges hinder the practical implementation of P2P networks for LLMs. One major issue is the bandwidth and latency required for efficient training. 

Training LLMs involves transferring vast amounts of data between nodes, which can be prohibitively slow on consumer-grade networks. One Reddit user pointed out that even on a 10-gigabit network, the data transfer rates would be insufficient compared to the high-speed interconnects used in dedicated GPU clusters.

Moreover, the synchronisation required for distributed gradient descent, a common optimisation algorithm in training neural networks, adds another layer of complexity. 

Traditional training methods rely on tight synchronisation between nodes, which is difficult to achieve in a decentralised setting. 

A research paper on the review of synchronous stochastic gradient descent (Sync-SGD) highlights the impact of stragglers and high latency on the efficiency of distributed training. 

… And Solutions

Despite these challenges, ongoing efforts exist to make decentralised AI a reality. Projects like Petals and Hivemind are exploring ways to enable distributed inference and training of LLMs. 

Petals, for example, aims to facilitate the distributed inference of large models by allowing users to contribute their computational resources in exchange for access to the network’s collective AI capabilities.

Additionally, the concept of federated learning offers a more feasible approach to decentralised AI. 

In federated learning, multiple nodes train a model on their local data and periodically share their updates with a central server, which aggregates the updates to improve the global model. 

This method preserves data privacy and reduces the need for extensive data transfer between nodes. It could also be a practical solution for decentralised AI, especially in privacy-sensitive applications like medical machine learning.

The post Now You Can Run Llama 3.1 405B on Your Computer Using Peer-to-Peer Network appeared first on AIM.

Recently, Oracle released HeatWave GenAI, touted as the industry’s first in-database LLM. Embedded within the MySQL database, the LLM eliminates the need for separate infrastructure or complex integration steps. The biggest positive here is that HeatWave doesn’t need GPUs for operations. 

“We don’t need a GPU at all. We already fine-tuned it, built into the model, built into MySQL. So there is no need for a GPU to further train it,” said Palanivel Saravanan, vice president of cloud engineering at Oracle India, in an exclusive interaction with AIM. 

GPUs are considered as the most scarce resource when one is looking to build an LLM. However, Saravanan reiterated, “There is no GPU in the process. It’s not elimination, there’s just no requirement of GPUs because we have already optimised the language model here.” 

LLM with Unstructured Data

MySQL HeatWave integrates models like Llama-3 and Mistral, simplifying deployment. Its native VectorStore manages unstructured data, improving accuracy and efficiency in generating statements and performing analytics. HeatWave streamlines these processes without extensive manual adjustments.

The HeatWave setup enhances accuracy by organising data into binary vectors that establish relationships. It accommodates various data types, ensuring robust model readiness without extensive modifications.

Saravanan further explained that any customer-specific document updates go directly to the automated vector store, organising data efficiently alongside existing information. With LLM residing in the database, there is a significant improvement in speed, security and even cost-efficiency. 

“Today, if you want to build a large language model for commercial use, you have to follow 10 steps. But over here, it’s just two steps,” said Saravanan.

Interestingly, the importance of vector databases have significantly gone up with generative AI. Vector databases provide LLMs with access to real-time proprietary data thereby assisting with development of RAG applications, which is an integral part of HeatWave. 

Special Purpose LLMs

With HeatWave GenAI, Oracle has built a special purpose LLM, something that Saravanan believes enterprises prefer over general-purpose LLMs. “Every enterprise looks for a general purpose large language model, maybe for evaluation purposes, but then to align to the business or to map to the business, they need a very special purpose LLM,” he said. 

With specialised LLMs, enterprise adoption would also be on the higher side. Interestingly, a number of companies have announced small language models which are a form of special purpose models. 

Recently, AI expert and innovator Andrej Karpathy said, “The LLM model size competition is intensifying… backwards,” alluding to the rise and efforts of big tech companies to build small language models. 

Last week, OpenAI released the GPT-4o mini, and Mistral announced Mathstral, a 7B parameter small language model. 

HeatWave, the special purpose model by Oracle is said to be 30x faster than Snowflake, 18x faster than Google BigQuery, and 15x faster than Databricks for vector processing.

Cloud Prowess In India

Continuing on building special purpose models, Oracle announced a specialised data cloud architecture. A few weeks ago, the company officially announced Exadata Exascale, which is the world’s only intelligent data architecture for the cloud, that provides extreme performance for all Oracle database workloads, including AI vector processing, analytics, and transactions. 

“Exadata Exascale offers Indian businesses a significant reduction in infrastructure costs by up to 95%. Our smart, scalable, and secure infrastructure allows organisations to pay only for what they use, improving operational efficiency,” said Saravanan. 

Exascale’s intelligent data architecture works on pay-per-use economics, allowing enterprises of any size to use the platform as per their need. It features a virtualised, database-optimised infrastructure with shared compute and storage pools. 

The company has been on a high with promising revenue numbers, riding on its cloud and generative AI services integrated on them. OpenAI had also decided to run its workloads on OCI, extending the Microsoft Azure AI platform to Oracle’s cloud services. 

The post Oracle Cools Down GPU Dependency with HeatWave GenAI  appeared first on AIM.

AI music generation is a tricky business. Amidst copyright claims and the need for fairly compensating artists, it becomes an uphill task for AI startups, such as Suno.ai or Udio AI, to gain revenue and popularity. 

However, Beatoven.ai, an Indian AI music startup, has gotten the hang of it in the most ethical and responsible way possible.

One of the most important reasons for that is its co-founder and CEO Mansoor Rahimat Khan is a professional sitar player himself and comes from a family of musicians going back seven generations. “I was very fascinated by this field of music tech,” he said. 

Khan told AIM that he started his journey at IIT Bombay and realised that though there were not many opportunities in India, he wanted to combine his passion for music and technology. 

Beatoven.ai is part of the JioGenNext 2024, Google for Startups Accelerator, and AWS ML Elevate 2023 programs. Khan said that the team applied to many accelerator programs because they realised they needed a lot of compute to fulfil the goal of building an AI music generator. 

The company raised $1.3 million in its pre-series A round led by Entrepreneur First and Capital 2B, with a total funding of $2.42 million.

After switching several jobs, Khan met Siddharth Bhardwaj and building on their shared passions for music and tech founded Beatoven.ai in 2021. “After coming back from Georgia Tech, I got involved in the startup ecosystem, and started working with ToneTag, an audio tech startup funded by Amazon,” said Khan. 

Everyone Needs Background Music in their Life

The co-founders found out that the biggest market was in the generation of sound tracks for Indie game developers, agencies, and production houses. “But when we look at the nitty gritty of the industry, copyrights are a very scary thing. We thought that generative AI could be a solution to this.” Khan said that the idea was to figure out how users could give simple prompts and generate audio.

The initial idea was to create a simple consumer focused UI where users could select a genre, mood, and duration to generate a soundtrack. But that was when the era of LLM hadn’t started and NLP wasn’t good enough for such tasks. “We started in 2021 before the LLM era, and our venture capital came from Entrepreneur First. We raised a million dollars in 2021 and quickly built our technology from scratch.”

The biggest challenge like every other AI company was the collection of data. “You either partnered with the labels that charged huge licensing fees or scraped [data]. That was the only other option. But if you did that, you would be sued,” said Khan.

All of the Tech

This is where Beatoven.ai takes the edge over other products in the market. Khan and his team started contacting small, and slowly bigger artists for creating partnerships and sourcing their own data. The company had a headstart as no one was talking about this field back then. Within a year, it amassed more than 100,000 data samples, which were all proprietary for them.

During the initial days, Beatoven.ai did not use Transformers. Khan said that it is one of the reasons that the quality was not that great. Later, when Diffusion models came into the picture, the team realised that it is the way forward for AI-based music generation. 

The company started by using different models for different purposes, this included the ChatGPT API from OpenAI. The Beatoven.ai platform also uses CLAP (Contrastive Language-Audio Pretraining), which is mostly used for video generation. 

Apart from this, the company uses latent diffusion models like Stability AI’s Stable Audio, VAE models, and AudioLLM, for different tasks such as individual instruments within the generated music. Then the company uses an Ensemble model for mixing all these individual audios together. 

For inference, the company uses CPUs (instead of GPUs), which keeps it fast and optimised, while reducing costs. 

Trained Fairly

Khan admitted that the audio files generated by Suno.ai’s have superior quality right now, but they also use Diffusion models, which makes them a little slow. “The quality is significantly better from where we started, but it’s not quite there yet.” Khan added that currently the speed is high because the company uses different models for different tasks.

To further expand the data, Beatoven.ai started partnering with several outlets such as Rolling Stone and packaged it like a creator fund. In January 2023, it announced a $50,000 fund for Indie music as a part of the Humans of Beatoven.ai program for expanding their catalogue. 

This gave Beatoven.ai a lot of popularity and many artists wanted to partner with the team. Khan said that the company aims to do more licensing deals to expand music libraries. “When it comes to Indian labels though, they are not yet open to licensing deals,” said Khan. 

Beatoven.ai’s model is certified as Fairly Trained and also certified by AI for Music as an ethically trained AI model.

Apart from music generation, Beatoven.ai is launching Augment, similar to ElevenLabs’s voice generation model. This would allow agencies to connect to Beatoven.ai’s API and train on their own data to make remixes of their own music. For the demo, Khan showed how a simple sitar tune could be turned into a hip-hop remix. 

“You can just use your existing content and create new songs. That’s the idea,” he said.

Currently, Beatoven.ai is also testing a video-to-audio model using Google’s Gemini, where users can upload a video and the model would understand the context and generate music based on that. Khan showed a demo to AIM where the model could also be guided using text prompts for better quality audio generation. 

Not Everyone is a Musician

Khan envisions that in the near future, companies such as Spotify or YouTube start open sourcing their data and offer APIs to make the AI music industry a little more open.

Meanwhile, while speaking with AIM, Udio’s co-founder Andrew Sanchez said, “It’s enabling for people who are just up and coming, who don’t yet have big professional careers, the resources, time or money to really invest in making a career. “It’s enabling a whole new set of creators.” This would make everyone a musician. 

When it comes to Beatoven.ai, he said that he aims to head in a more B2B direction as building a direct consumer app does not make sense. “I don’t believe everybody wants to create music,” added Khan, saying that not everyone is learning music in the world. That is why, the company is currently focused only on background music without vocals. 

The post India’s Beatoven.ai Shows the World How AI Music Generation is Done Right appeared first on AIM.

Back in 2017, Google made UPI and cashless transactions household terms with the launch of  Google Pay in India. The app has over 67 million users in the country, its largest single market to date. 

Now, the company is trying to mimic the success of Google Pay to democratise transformation for all via Namma Yatri, the Bengaluru-based open-source rival of ride-hailing services like Ola and Uber.

Interestingly, earlier this week, Google invested in Moving Tech, the parent company of Namma Yatri. On the same date, Namma Yatri users also came across new features in its service including rentals and instant travel. 

Previously in 2020, Google said it plans to invest $10 billion in India over the next five to seven years as the search giant looks to help accelerate the adoption of digital services in the key overseas market.

Namma Yatri, initially a subsidiary of the payments company Juspay, became a separate entity called Moving Tech in April. CEO Magizhan Selvan and COO Shan M S, formerly with Juspay, now lead this new mobility business. 

Launched more than a year ago, it became a preferred service for many given its driver-centric approach with lower commission rates and a community-driven model tailored to local needs. Recently, it also expanded to multiple cities beyond Bengaluru, including Chennai and Delhi.

Similarly, Google is doing the same with ONDC as well, making it the UPI of ecommerce. At the recently held Google I/O Connect in Bengaluru, the company announced that it has lowered India-specific pricing for its Google Maps Platform API. Interestingly, it is offering up to 90% off on select map APIs for people building on top of ONDC.

Making Money Simply

The success of Google Pay in a price-sensitive market like India can be attributed to strategic initiatives, market understanding, and leveraging unique conditions. It leveraged the UPI infrastructure, an open API platform from the National Payments Corporation of India (NPCI), which enabled instant online bank transactions. This reduced entry barriers for new users and merchants. 

India’s cash-dependent economy, with many unbanked and underbanked people, was targeted by Google Pay through an easy-to-use digital payment solution requiring only a bank account linked to UPI. 

It solved the crisis which Indian fintech players like PhonePe and Paytm had been struggling with for a long time. 

Google built an ecosystem in India by integrating various services like bill payments, mobile recharges, and peer-to-peer transactions, embedding itself in daily financial activities. In 2021, it expanded its network of banks offering card tokenisation on its app, adding SBI, IndusInd Bank, Federal Bank, and HSBC India as partners.

Now, even international tourists can access the app in India. 

Now They Look to Make Travel Simple

In India, Namma Yatri competes with the likes of Uber, Ola and Swiggy-backed Rapido. 

Google Maps, with an established customer base in India, has incorporated Gemini into it and has come up with new features like Lens in Maps, Live View navigation, and address descriptors, specifically for Indian users. 

However, earlier this week, Ola founder Bhavish Aggarwal decided to ditch Google Maps for its own in-house Ola Maps. After Microsoft Azure’s exit last month, this is Aggarwal’s second move to get rid of Western apps. Yesterday, he reduced the pricing for the Ola Maps APIs even further to encourage other developers to build on it. 

“We’ve been using Western apps to map India for too long, and they don’t understand our unique challenges: street names, urban changes, complex traffic, non-standard roads, etc,” said Aggarwal.

Meanwhile, Rapido, Uber, Namma Yatri and others ride-hailing platforms continue to use Google Maps. Currently, Rapido, which started off as a bike taxi service has expanded to include auto rickshaw, carpooling services, taxicab hailing, parcel delivery, and third-party logistics services in over 100 Indian cities.

Namma Yatri: The UPI of Transportation

“Imagine the platform similar to Namma Yatri being adopted in cabs, metros, or any platform which is serving the passengers,” Selvan told AIM, when Moving Tech was first launched. “We essentially want to become the UPI of transportation.”

Namma Yatri stood out for the drivers since the app started off as free of commission but now charges a basic of just INR 25 per day. In contrast, Ola and Uber take a 25-30% commission from drivers per ride. 

One Namma Yatri driver told AIM that the low subscription fee compared to Uber and Ola has helped him. Within six months of switching to Namma Yatri, he was able to fund his two children’s weddings.

The app has onboarded 49,000 auto drivers and 550,000 users in five months, with approximately INR 12 crores ($1.5 million) paid out to drivers. It celebrated 500 million downloads in March.

Although Namma Yatri currently lacks features like bike taxis and carpooling, with Google’s support it may soon expand to include these options. 

The founders said that Namma Yatri will leverage the new funds to grow its engineering and R&D competencies, and also include more types of transportation, including buses. 

On the other hand, Google has found an ideal partner to strengthen its presence in India’s transportation sector. Given its expertise in revolutionising online purchases, it is set to replicate the ‘Google Pay moment’ in Indian transportation soon.

The post Google is Trying to Mimic ‘Google Pay’ for Transportation via Namma Yatri appeared first on AIM.

OpenAI sprung out of its silent zone with a new research paper on ‘Prover-Verifier Games’ (PVG) for LLMs. PVGs look to improve the ‘legibility’ of LLM outputs or rather make sure that LLMs produce understandable and logical text even for complex tasks such as solving maths problems or coding. 

In this method, OpenAI trained advanced language models to generate text that can be easily verified by weaker models. It was observed that this training improved the comprehensibility of the text for human evaluators, which hints at improving legibility. 

The ‘Prover’ and ‘Verifier’

“Techniques like this seem promising for training superhuman models to explain their actions in a way that humans can understand better (and get less fooled by). I’d be excited to see this method tried on harder tasks and with stronger models,” said Jan Leike, the co-author and former researcher at OpenAI, who had worked on the recent PVG paper. 

The paper is based on the first concept of PVG released in 2021, which is a game-theoretic framework designed to incentivise learning agents to resolve decision problems in a verified manner. 

Akin to a check system, a ‘prover’ generates a solution which a ‘verifier’ checks for accuracy. OpenAI’s method trains small verifiers to judge solution accuracy, encourages “helpful” provers to produce correct solutions approved by verifiers, and tests “sneaky” provers with incorrect solutions to challenge verifiers. 

It was noticed that over the course of training, the prover’s accuracy and the verifier’s robustness to adversarial attacks increased. Interestingly, the PVG system alludes to a form of reinforcement learning, something OpenAI’s co-founder and former chief scientist Ilya Sutskever was a strong advocate of. 

Prover-Verifier Games for LLMs

Source: X

Looking back at the history of OpenAI’s models much before ChatGPT, the company had been extensively working on reinforcement learning systems. In 2018, OpenAI Five, which was built on five neural networks, defeated human teams at Dota 2. The system played 180 years worth of games against itself – a sort of reward mechanism in the loop to train itself. 

“The neural network is going to take the observations and produce actions and then for a given setting of the parameters, you could figure out how to calculate how good they are. Then you could calculate how to compute the way to change these parameters to improve the model,” said Sutskever at an old Berkeley EECS seminar. 

Interestingly, PVG also works on similar lines. However, it comes with its limitations. The experiment was done on maths problems which have an answer that can be tested via the right and wrong method. However, with topics that come with broad subjectivity, the PVG system for an LLM may struggle. 

“It’s hard and expensive to codify the rules of life. How do we objectively determine whether one poem is more beautiful than another?

“I think a very interesting metric would be to measure the accuracy of the fine-tuned models on unrelated tasks to see if the lessons learned to be better at explaining maths problems would help the model perform better on explaining other problems (such as logic or reasoning),” said a user on HackerNews. 

PVG for Superintelligence

Source: X

The prover-verifier gaming system looks to improve the accuracy of LLM-generated results. Not just that, it also sets the next path for achieving superintelligence. 

The methodology has a significant advantage in reducing the dependence on human demonstrations or judgments of legibility. This independence is particularly relevant for future superintelligence alignment. 

While the study focused on a single dataset and currently necessitates ground truth labels, it is anticipated that these methodologies will prove pivotal in the development of AI systems. Their goal is not only to ensure correctness in outputs but also to facilitate transparent verification, thereby enhancing trust and safety in real-world applications. However, will the new method form the next standard for LLM accuracy, is something that remains to be seen. 

The post The Need for Prover-Verifier Games for LLMs  appeared first on AIM.

On a mission to create the world’s most secure and fastest global point-to-point low-latency orbital network, Rivada Space Network has been employing digital twins to create a simulation environment to build the same. 

“Digital twin is fantastic,” said Declan Ganley, the founder and CEO of Rivada Space, in an exclusive interaction with AIM at the recently concluded AWS Summit in Washington DC. 

“When you are working with your enterprise and government customers, you can show them all sorts of options like geo-fencing and routing. You can mesh every single vessel,” he added.

The Vital Digital Twin 

Rivada’s digital twin enables them to accurately model the behaviour of their satellite constellation. This modelling capability is crucial for testing various operational scenarios before actual deployment. 

By simulating different conditions, such as orbital dynamics and network traffic patterns, Rivada can predict how their satellite network will perform in real-world scenarios. This pre-testing reduces the risk of errors and enhances the efficiency of network deployment and operation.

“What we can do is, in AWS, build a digital twin of this entire constellation to be able to not only visualise the constellation and calculate visibility windows and contacts in seconds. This would usually take hours, but we get it down to seconds. 

“That enables Rivada to run what-if scenarios, capacity planning, and after-support and using the serverless power of the cloud,” said Alistair McLean, principal tech BDM for aerospace and satellite, AWS. 

Speaking about the tech configuration for the digital twin, McLean explains that they use a queue-based system, Amazon’s SQS (simple queue service) which creates Lambdas and Batches. “If you flip any good simulator on its head and actually feed it data, it really becomes a digital twin,” said McLean. 

Elaborating on the vitality of digital twins in space and satellite technologies, Ganley talks about its usage on the manufacturing side. “You can do things in one day that five years ago would have taken you 50,” he said. Simply put, satellite efforts that require 100 years are easily cut to two years. 

With this accelerated pace, Rivada Space, a relatively new entrant setup in 2022, is expected to bring low-earth-orbit satellites in 2028. 

Interestingly, Ganley believes cost is a better deal than the time advantage. “Digital twins give you savings that are better than that,” he said. 

Rivada’s approach is not the first time a satellite network company has adopted the concept of digital twin. Elon Musk’s satellite internet constellation Starlink has already adopted the same. 

Space and Simulation 

The use of digital twin technology in this sector not only helps create virtual models of satellites but also enables better security measures that facilitates smoother operations. They even help with troubleshooting and quicker resolution for network issues and disruptions. 

The process thereby helps save time, cost and resources. 

SpaceX has been using digital twins to test their Starlink and is said to be so advanced that it can even emulate the effects of solar flares on its network. 

It was said that digital twins can reduce the time needed to deploy AI-driven capabilities by up to 60%, cutting capital and operating expenditures by up to 15%, and improving commercial efficiency by 10%. 

Further, digital twins facilitate virtual testing within a secure and regulated environment, leading to substantial reductions in the time and resources needed for validation and verification.

Not Challenge Free

Though promising, digital twin also come with its challenges. It can encounter limitations such as integrating complex data sources, ensuring data accuracy and availability, scaling for large systems, addressing security concerns, and achieving interoperability in diverse environments. 

However, a recent research paper named ‘Plotinus: A Satellite Internet Digital Twin System’, talks about a digital twin based on microservices and designed for satellite internet emulation. 

Plotinus aims to tackle SAGIN (space-air-ground networks) development challenges with modular design for aerial vehicle emulation, flexible path computation methods, real-time emulation with live network traffic, and dynamic satellite network modelling.

Big Tech Leads the Way

GPU giant NVIDIA has been advancing its digital twin capabilities through various platforms and collaborations. Its Omniverse platform has been widely adopted across robotics and autonomous industries. Considering how simulated training is one of the key methods for training robots, digital twin is the perfect fit. 

“Eventually we’ll have sophisticated autonomous robots working alongside humans in settings like kitchens — manipulating knives and other dangerous tools. We need digital twins of the worlds they are going to be operating in, so we can teach them safely in the virtual world before transferring their intelligence into the real world,” said Rev Lebaredian, the VP of omniverse and simulation technology at NVIDIA.  

NVIDIA has also partnered with Foxconn to build AI factories or factory digital twins. “Our digital twin will guide us to new levels of automation and industrial efficiency, saving time, cost and energy,” said Foxconn chairman Young Liu. 

Similarly, NVIDIA’s Earth is a digital twin cloud platform for simulating and visualising weather and climate conditions. The company has a similar project to monitor global environmental conditions via digital twin for the National Oceanic and Atmospheric Administration (NOAA).  

Other industries are following suit. Semiconductor companies have already seen the need to employ digital twins to replicate their semicon factory settings. While this is in motion, it is possible that digital twins can help India catch up on the semiconductor race against incumbent players. 

The post Digital Twin in Space Research Cuts 100 Years of Work Down to 2 Years appeared first on AIM.

During a recent podcast at Johns Hopkins University, Mira Murati, the chief technology officer of OpenAI, acknowledged the criticism that ChatGPT has received for being overly liberal and emphasised that this bias was unintentional. 

“We’ve been very focused on reducing political bias in the model behaviour. ChatGPT was criticised for being overly liberal, and we’re working really hard to reduce those biases,” said Murati. 

However, no specific details or measures on the redressal efforts have been provided yet. This is all part of their ongoing effort to improve the model and make it more balanced and fair.

However, in an interview back in March, Murati was asked where the video data that was used to train Sora came from. The CTO feigned ignorance, claiming to not know the answer, making her the talk of the town on social media.

Netizens were quick to create memes highlighting her as “an absolute PR disaster”.

OpenAI Needs No Safety Lessons

OpenAI has a history of bad PR, but it knows how to turn a crisis into an opportunity. In a previous discussion moderated at Dartmouth, Murati focused on safety, usability, and reducing biases to democratise creativity and free up humans for higher-level tasks.

In a recent post on X, she said that to make sure these technologies are developed and used in a way that does the most good and the least harm, they work closely with red-teaming experts from the early stages of research.

“You have to build them alongside the technology and actually in a deeply embedded way to get it right. And for capabilities and safety, they’re actually not separate domains. They go hand in hand,” she added.

Notably, her optimism on AI stems from the belief that developing smarter and more secure systems will lead to safer and more beneficial outcomes for the future. However, she is now facing questions about ChatGPT’s perceived liberal bias.

Meanwhile, OpenAI’s former chief scientist Ilya Sutskever launched Safe Superintelligence shortly after leaving the company in May 2024, allegedly due to disagreements with CEO Sam Altman over AGI safety and advancement.

In an apparent response to this and to ward off safety concerns, OpenAI formed a Safety and Security Committee led by directors Bret Taylor, Adam D’Angelo, Nicole Seligman, and Altman.

Murati to the Rescue 

In a July 2023 discussion with Microsoft CTO Kevin Scott, Murati expressed concerns about the prevailing uncertainty in the AI field, emphasising the need for clear guidance and decision-making processes. 

She highlighted the challenge of determining which aspects of AI to prioritise, develop, release, and position effectively. “When we began building GPT more than five years ago, our primary focus was the safety of AI systems,” said Murati.

Highlighting the risks of letting humans directly set goals for AI systems—due to the potential for complex, opaque processes to cause serious errors or unintended consequences—Murati and her team shifted their focus to using RLHF to ensure AI’s safe and effective development.

Briefly, after GPT-3 was developed and released in the API, OpenAI was able to integrate AI safety into real-world systems for the first time.

An Accidental PR 

Murati’s acknowledgement of ChatGPT’s perceived liberal bias and her emphasis that this bias was unintentional represent a significant and positive step towards the responsible use of AI. 

Her addressing criticisms openly demonstrates a commitment to transparency and accountability, which are crucial for the ethical development of technology. 

Murati’s approach not only seeks to rectify past concerns but also underscores a proactive stance on refining AI systems to better serve diverse user needs. This openness fosters trust and shows that OpenAI is dedicated to addressing issues constructively. 

Murati’s tryst with responsible AI is not new-found. In a 2021 interview, she discussed AI’s potential for harm, emphasising that unmanaged technology could lead to serious ethical and safety concerns. Some critics argued that Murati’s comments were too alarmist or did not fully acknowledge the positive potential of AI. 

While Murati aimed to promote responsible AI, the backlash led to broader debates on the technology’s future and its societal impacts.

Not to forget the ‘OpenAI is nothing without its people’ campaign started by Murati during Sam Altman’s ousting. One thing is for sure: Murati is truly mysterious, and no one knows what she’s going to say next to the media. We are not complaining! 

The post OpenAI CTO Mira Murati is an Absolute PR Disaster appeared first on AIM.

In May, AWS announced the general availability of Amazon Q, which comes in three variations – Amazon Q for Developers, Amazon Q for Business and Amazon Q Apps.

At the ongoing AWS Summit in New York, the hyperscaler announced AWS App Studio– a generative AI-powered service that uses natural language to create enterprise-grade applications.

In an interaction with AIM prior to the announcement, Sriram Devanathan, GM of Amazon Q Apps and AWS App Studio at AWS said App Studio is going to be transformative. He said, “It’s the fastest way to build enterprise-grade applications in the sense that they have multiple UI pages, they can pull data from multiple sources and they can embed complex business logic in them.”

While Amazon Q is designed for developers and businesses, AWS App Studio will open the doors for technical folks who are not professional developers like an IT project manager, data engineer or enterprise architect.

Amazon Q is Powered by Bedrock

Devanathan also revealed that, like Amazon Q, AWS App Studio is powered by Amazon Bedrock.

Bedrock incorporates models such as the Titan Series from AWS, the Claude models from Anthropic, and models from Cohere, AI21 Labs, Mistral, and Meta.

In App Studio, you simply describe the problem you need to solve—for instance, tracking inventory and managing rentals and replacement parts. App Studio utilises a diverse array of models behind the scenes to address various aspects of the problem.

“App Studio analyse your input to define requirements, much like a human team member would. We outline the specific features needed to support your requirements. You can engage in a dialogue, adjusting features as needed through natural language interactions, again powered by different models,” Devanathan pointed out.

Furthermore, once the user has identified the necessary features, they can request, ‘Create the app for me.’ This action triggers a separate set of models to generate a fully functional application pathway.

“Bedrock features a variety of powerful models, including the Titan family from Amazon and some of the world’s most advanced general-purpose models from Anthropic. We leverage them all,” he said.

Building Applications in a Minute

AWS App Studio is transformative because one can build an application by interacting in natural language and under a minute.

Users only need to describe their desired application’s functionality and data sources to integrate. App Studio then swiftly constructs an application that might otherwise require days for a professional developer to build from scratch.

Modifying App Studio applications is straightforward through its point-and-click interface. Additionally, its generative AI-powered assistant provides real-time guidance on task completion.

“We have a customer named Campus Life and Style, overseeing 50 locations catering to 28,000 students’ diverse needs, including housing and events management. Given the variety in processes across locations, they implemented App Studio in a few areas with the help of two technical team members.”

This initiative resulted in a 20% boost in productivity and a remarkable 98% reduction in errors from manual data entry. These gains have prompted them to expand its use across multiple locations and processes, showcasing significant improvements in operational efficiency.

Currently, App Studio is available in preview in the US West (Oregon). However, when we asked Devanathan about its availability in India, he mentioned that there is no timeline as of yet.

“We are getting a lot of interest from customers in that region and we will soon figure out the region’s roadmap.”

New Updates to Bedrock

At the ongoing Summit, AWS has also announced new Amazon Bedrock innovations. Bedrock users will now be able to fine tune Anthropic’s Claude 3 Haiku, making Bedrock the only fully managed service that enables this.

AWS also announced vector search for Amazon MemoryDB, offering unparalleled speed and recall rates among leading vector databases on AWS. This feature is ideal for applications needing ultra-low latency in the single-digit milliseconds.

Moreover, “Today, we’re adding contextual grounding checks in Guardrails for Amazon Bedrock to detect hallucinations in model responses for applications using RAG and summarisation applications,” AWS said.

The post AWS Announces New App Studio Powered by Amazon Bedrock appeared first on AIM.

Baidu is clearly the darling of autonomous giants. A few months ago, Tesla chief Elon Musk visited China, the second largest market for the company, to release FSD (full-self driving) robotaxis in the region later this year. 

However, this was made possible only through a partnership with Baidu’s Apollo, which is already slaying in the autonomous segment with its fleet of robotaxis making it the largest autonomous ride-hailing service in the world. 

Baidu’s robotaxis are currently available in four cities in China, including Wuhan, Beijing, Shanghai and Chongqing. In Wuhan alone, it caters to a population of seven to eight million covering an area of 3000 sq km. 

“Here [Wuhan], we have 6000 driverless rides per day. So, all of them combined together is quite a big achievement,” said Helen K Pan, general manager and board of directors for Baidu Apollo, California, in an exclusive interaction with AIM.

Though robotaxis currently operate in only four cities, their autonomous driving service is available in 11 cities across China. 

In May, Waymo, a self-driving cab service, confirmed that they are operating more than 50,000 paid trips every week across three cities in the US. 

Level 5, Not Yet 

Baidu has been working in the driverless segment for over a decade and has developed various levels of autonomy. It currently operates at Level 4. However, their technology borders on Level 5 too. “Typically, we call it L4 and L5, with L4 [autonomy] implying in geo-fenced regions,” said Pan. 

“With the government policy, we are actually more in the L4 region.” 

As of April 2024, the cumulative test mileage of Apollo L4 has exceeded 100 million kilometres.

While Baidu has been advancing its autonomous capabilities for over a decade, the West is already competing in this space with Tesla and Waymo. However, Level 4 is what they all are at. 

“I think we are all at L4 today; and with the government regulation, it’s not possible to do L5. Another thing is that, I think, all of us providing this technology haven’t been tested in all the scenarios. We wouldn’t have this confidence to claim that we have the L5 capability.”

Baidu vs the World?

Pan believes that the training conditions for Baidu’s vehicles act as the biggest differentiator when compared to Tesla or Waymo. 

“China has a couple of unique market technology challenges compared to the US and other countries,” said Pan. “The roads are in a diversified condition in China. The driving behaviour, the population density, and the driving scenarios are much more complex.”

Pan also elaborated on how the aggressive driving behaviour of drivers in the Chinese market, especially in certain regions of the country, contributes to building rich data. “Outside China, you’ll take a longer time to see, but you will face those corner cases here more frequently. So, that’s a benefit,” she said.  

In addition to rich data, costing is another differentiator. “What I see in China is that the cost of the robot taxi is really low. We have to have the cost constraints much tighter than the rest of the world.”

Interestingly, Pan comes with a rich experience of working with autonomous technology. Before Baidu, she worked with Google and Waymo on their self-driving technology.  

Challenges Remain

Though comfortably placed in the autonomous vehicle (AV) space, Baidu is still experimenting to navigate the challenges associated with it. “If you look at it from an AI point of view, I think it’s challenging. AI algorithms are evolving,” said Pan, who explains that you need real-time and extremely high-accuracy data. 

Pan confirmed that the company is also looking at developments around large models. “We’re providing this development, an end-to-end large model and we are able to tailor all the diversified scenarios. 

“It will also help us migrate from one city to the next, once you have this large model instead of a very customised one for a particular city. So, I think that’s quite a big breakthrough for us to be able to further expand our technology into multiple regions in China,” said Pan. 

Safety Regulation

Pan discussed the government’s apprehensions associated with AV technology. 

“The government actually wants to embrace this new disruptive technology. But on the other hand, they don’t know how to really regulate it. They want to foster this technology, and make sure it’s not chaotic and that they can really control this one,” she said. 

To address this, Baidu is actively working with the Chinese transportation department to help form regulations to ensure safe deployment. 

“We run a lot of simulations, and we also do a lot of early testing with the safety driver behind the wheel, and proving our technology is working. Then, we gradually remove the safety driver, and become completely driverless. So, when we reach the driverless stage, we know our technology, and we have the confidence,” said Pan. 

Furthermore, there are 10 levels of redundancies built into the autonomous driving systems, and AI algorithms. The Chinese government also mandates specific test conditions and routes to obtain a driverless certification. 

When asked if 100% safety can be achieved, Pan was pragmatic in her answer. “Nobody can say everything is 100% safe, but I think it will evolve. If you look at it 5-10 years from now, we’ll still continue to improve our safety.” 

The post Baidu Clocks 6,000 Driverless Rides Per Day in Wuhan, China appeared first on AIM.

Are AI agents the next big thing? Bengaluru and San Francisco-based startup Floworks definitely think so. The startup, funded by Y Combinator in the Winter 23 cohort, is building AI agents that can handle end-to-end sales functions in an organisation.

In an exclusive interaction with AIM, Floworks co-founders Sudipta Biswas and Sarthak Shrivastava said that the startup is building what they call an ‘AI employee’. 

Large language models (LLMs) today can’t do certain things despite their impressive knowledge base, like sending an email. The startup, which raised $1.5 million in seed funding earlier this year, has developed Alisha, which they call the world’s first fully autonomous AI-powered sales development representative (SDR).

Alisha can automate the entire lead generation process, from prospecting and qualification to scheduling meetings and sending emails, allowing human sales representatives to focus on more concrete tasks.

Alisha Achieves 100% Reliability 

The startup claims that with Alisha, users will generate 10x more leads each day. The AI assistant, which is powered by Thorv2, can interact with external tools such as emails, CRMs, Google Search, and Calendars. It can also parse through one’s email for leads and update the CRM accordingly.

“On any given day, we probably use 10 to 15 different software or tools. But a standard language model like ChatGPT cannot use these tools, and this is where we come in. 

“Our model, which we are internally calling ThorV2, is the most accurate and the most reliable model out there in the world when it comes to using external tools right now,” Biswas said.

Alisha is a product that comes out of the box and can be set up in under five minutes. However, the startup helps fine-tune Alisha to the particular business’s use case by training Alisha with that particular company’s data. 

“Alisha, our SDR, trains itself using this data, integrating workflows and processes specific to each customer. Essentially, each instance of Alisha becomes tailored to each customer, equipped to handle all inquiries about their company and products,” Shrivastava said.

The startup claims Alisha is the most reliable tool in the market, with ThorV2 near zero hallucination and reliability for function calling at 99.5%, which is impressive. 

The startup also claims other models like Claude-3 Opus and GPT-4o have a reliability score of 59.7% and 83.9%, respectively. 

“When using ChatGPT, if you provide the same prompt twice, chances are you’ll receive different answers each time. This variability is inherent to how language models operate,” Biswas pointed out.

However, this very nature of LLMs and the fact that they hallucinate limit the use of these large models for AI agentic workflows.

“When it comes to tool usage, for instance, accessing your CRM, you want very reliable and deterministic action because you don’t want wrong information to be put in your CRM or send a wrong email to your customer. 

“Mistakes at the LLM level can propagate widely, impacting multiple areas with significant business repercussions. Our model ensures 100% reliability in these critical functions,” Biswas added.

(From left to right: Sarthak Shrivastava and Sudipta Biswas, co-founders at Floworks)

ThorV2 Powers Alisha

Thor2, which powers Alisha, is a mixture of agents (MoA) based on a proprietary LLM architecture built from scratch.

“ThorV2 comprises eight distinct AI agents or LLMs, incorporating both open-source and proprietary models. While some are pre-trained, others require us to rebuild their architecture entirely to align with our specific requirements,” Biswas said.

He further claims that Thorv2 is not only 36% more accurate than OpenAI’s GPT-4o but also 4x cheaper and almost 30% faster in terms of latency.

When asked about the cost of building ThoV2, Biswas revealed that it was around $1 million lower, which is significantly lower compared to foundational models.

Moreover, ThorV2 costs just $1.60 per 1000 queries, making it 175% cheaper than GPT-4o.

(Source: Floworks)

Voice Capabilities in the Pipeline 

The founders also revealed that voice capabilities for their Alisha is something they are working towards. Earlier this year, we saw both OpenAI and Google fascinate everyone with the human-like voice capabilities of their models.

Though OpenAI made GPT-4o available for free, it is yet to release the voice capabilities. Biswas reveals that voice is something they are working towards, but he does not see a demand for multimodal, which involves videos. 

Moreover, he claims Floworks does not want to make bold promises and then take an eternity to deliver a product. 

“It’s very easy to build a prototype and woo the audience, but then to actually build and release a production version, where everyone can use it scalably, that’s a whole different challenge,” Biswas said.

An alumni of IIT Kharagpur, Biswas points out a competitor of Floworks called Adept AI. “The company has raised around $400 million so far, and despite being in existence for over two years, still hasn’t released a product. In fact, the company is on the verge of breaking up,” he added.

Automating End-to-end Sales 

While Alisha is designed for function calling, the vision of the company is to build an AI system that handles end-to-end sales functions in an enterprise.

“We actually envision that in the near future, using Floworks, companies will not require sales teams. It will just be basically creating a good product and then actually training an AI system to sell the product,” Shrivastava said.

When asked if Alisha can make cold calls, he added that Alisha is already sending emails on behalf of humans. Soon, it will start reaching out to people on LinkedIn, WhatsApp and other mediums.

“Voice call too is just another mode of communication, and once the technology is ready, Alisha, too, will start making calls,” Shrivastava added.

While the co-founders are confident, it remains to be seen how much of these can be achieved, particularly whether AI can grasp the nuances of sales functions.

Expansion Plans 

Biswas revealed that the team size is 17 currently and they plan to hire for additional roles and expand the team to around 30-35 in the coming months. 

The company started active sales only two months back and so far has acquired around 14 customers both in the Indian and the US markets. Some of the customers include Anya, Unscript, and Qodex.

“We are growing 100% every month when it comes to customer acquisition, and we are getting a lot of referral customers who are knocking on our doors and enquiring about our product,” Shrivastava said.

Alongside Alisha, Floworks’ upcoming products include AI RevOps, AI project managers, and AI Executive Assistants. These innovations aim to optimise business operations, enabling “one-person unicorns.”

The post AIM Exclusive: YC-backed Indian Startup Claims its AI Agent is Better than OpenAI’s GPT-4o appeared first on AIM.

India boasts the world’s highest number of female STEM graduates, with women making up about 40% of these degree holders.

In 2020, a World Bank report highlighted global trends in women’s participation in STEM, revealing higher graduation rates for women compared to men. However, only 14% of these women make it into the workforce.

Women are notably underrepresented in fields such as engineering, Information and Communication Technology (ICT), and physics. 

According to an HRD ministry report for the academic year 2019-2020, the percentage of girls admitted in core engineering disciplines is as follows- mechanical engineering: 5.9%, civil engineering: 22.5%, metallurgical engineering: 22.7%, chemical engineering: 23.2% and electrical engineering: 27.3%.

These figures highlight the critical need to bridge the gap between education and employment for women in STEM. And with initiatives by Infosys Foundation and TalentSprint, things are slowly changing. 

Efforts to Close Gender Gap in Engineering

Sudha Murthy broke barriers as the first woman in India to study mechanical engineering, who then went on to become the chairperson of Infosys Foundation (now former). Today, her legacy of defying societal expectations is echoed in the journey of Ashna Kapoor, a young woman from Amritsar.

Kapoor convinced her family to support her dream of pursuing engineering, a path that led her to a career at Arcesium. She attributes it to the empowerment she gained from a Woman Engineering program. 

Kapoor’s story is just one of many that showcase how such programs can challenge and change gender stereotypes. 

Take Dharani Devi, for instance. Despite battling a medical condition and lacking a background in computer science, she secured a six-month internship followed by a permanent position at American Express, all thanks to an engineering program that helped her. 

In a significant move to further this cause, the Infosys Foundation recently signed an MoU with ICT Academy of Tamil Nadu, a non-profit organisation, to increase the employability of young learners in rural India.  

The partnership further aims to establish ‘Centers of Excellence for Women and Youth Empowerment’ in over 450 colleges in India, that will serve as the hubs for skill development and training, both online and offline, as well as enable job placements. 

The curriculum includes 80 hours of core skills training, 20 hours of soft skills development, and placement facilitation for certified students. 

The centres will also host youth empowerment summits and coding practice sessions to tackle real-world problems, ensuring that participants are well-equipped for the challenges of the tech industry. 

TalentSprint Walks the Same Path

Similarly, TalentSprint, an Indian edtech company, is creating an ecosystem of women in tech talent through its Women in Software Engineering (WISE) and the Women Engineers program (WE).

Supported by Google, the WE program aims to enable enterprising and aspiring women engineering students from diverse socio-economic backgrounds to aspire and achieve high-growth tech careers.

Rohit Agarwal, chief delivery officer, TalentSprint, mentioned, “This commitment to inclusivity is evident in the program’s makeup: 39% of participants come from low-income families, 32% are first-generation graduates, and 25% hail from rural and semi-urban areas.” 

“By offering targeted support, resources, and opportunities, WE foster a more inclusive and supportive ecosystem for women in Indian tech.”

So far, the program has received over 1,30,000 applicants from 11,000+ pin codes and 100+ tier 2 colleges in the last 6 cohorts and substantially created an impact across India.

Through financial assistance, including 100% scholarship and additional cash support, the program empowers aspiring women engineers to pursue and excel in their academic and professional endeavours. 

Selected participants not only receive financial support but also gain access to various resources. This includes mentorship by Google engineers, specialised boot camps, networking events, and avenues for potential career opportunities. 

The post India’s Women Engineers are Ready to Work, But Jobs Seem to Be on Vacation appeared first on AIM.

Declan Ganley has been travelling the world and meeting many influential people, raising billions of dollars for his project. The founder, chairman and CEO of Rivada Space Networks, Ganley, an English-born Irish entrepreneur who is also a political activist, has been busy spreading the word on ‘outernet’. 

The Outernet is Coming

“The outernet is not a branding term or a marketing thing. It is something that is distinct and separate from the internet, but is truly global,” said Ganley in an exclusive interaction with AIM, at the recently concluded AWS Summit held in Washington DC. 

Ganley’s team is working on building the world’s most secure, fastest, global point-to-point low-latency orbital network to transmit data from and to every part of the world—a private space network for enterprise and government. Based out of Munchen, Germany, Rivada Space was set up in 2022.  

Rivada has already signed a $10.3 billion MoU with many enterprise customers and even the US government.  The company has partnered with AWS to leverage their infrastructure to meet Rivada’s needs.

“What we want to do is deploy a piece of global communications infrastructure that would be the fastest at a distance of 4000 km or more, have the lowest latency, and have the smallest attack surface on the network, and that would be truly point to multi point. From any point on the surface of the planet to any other point on the surface of the planet,” said Ganley. 

‘The Goldilocks Zone’ 

The ambitious project is banking heavily on setting low-earth-orbit satellites, which are expected to be completed in 2028. Rivada is smartly choosing the optimal zone, which is neither too low nor too high for deploying these satellites. 

“We can achieve dense global coverage with 600 satellites at 1050 kilometres, because you can see more from there,” said Ganley. “The Financial Times calls it the ‘Goldilocks zone’ for low earth orbit, because you’re not too high and you’re not too low, so you don’t need too many satellites, but you’re not so high that you lose in latency.”

The minimal number of satellites facilitates faster connections, reducing the number of hops needed to travel from one point to another. With each satellite serving as a router with four lasers on it, it acts as fibre optic cables, except that it travels through the vacuum of space. “Light through a laser in the near vacuum of space is 60% faster than light through a fibre. So you’ve got a faster connection, because you don’t have 1000s of satellites,” said Ganley. 

Big Players Are Around the Corner

What Rivada is attempting to achieve is something SpaceX and Amazon have already heavily invested in via Starlink and Project Kuiper, respectively. Last year, Amazon completed the tests for an optimal mesh network in low-earth orbits with infrared lasers, similar to Ravida’s mission. 

The difference, however, lies in logistics. Amazon has launched over 3000 satellites into an orbit of 590 to 630 km. On the other hand, SpaceX is in a bigger league, with over 5000 satellites aiming to reach a constellation of 42,000. 

Early Believers of Generative AI 

While the company works with space tech, generative AI has been a prominent player at Rivada. Ganley saw generative AI as something that would be important before it became fashionable.

“If you look at Rivada’s patent portfolio, you will find a very early patent for generative AI, which has been awarded, and not a pending patent. It’s from two years ago, and has been through the whole evaluation [process],” he said. 

Rivada’s early patents pertains to utilising generative AI for managing obfuscation masking, effectively enhancing security in wireless communication networks, including satellite networks. It focuses on leveraging generative AI techniques to obscure traffic patterns, thereby adding an additional layer of security to ensure robust confidentiality and integrity in data transmission.

Treads India Carefully

While Rivada looks to capture the global market, India is on the list too. However, the company is carefully treading the route of collaboration. The company is working towards obtaining market licences in India. 

“We are not going to do anything in India without going very carefully and working with the regulator in India. There will be no surprises,” said Ganley. 

He also assured in abiding by the rules of the Indian government, and guaranteeing data sovereignty, residency and security to allow any enterprise or government entity in India to have total possession of its data at all times. 

The CEO was also optimistic about the country turning this technology around to its best use. “If I was to talk about economies, to say, who do I think would be able to exploit this in the shortest period of time to really get this and turn this to their advantage? I think the Indian economy is extremely well positioned to do that. I would say, even South Korea,” said Ganley. 

Revealing his personal side, Ganley believes in the “mission” as a crucial factor that gets him going. “Mission, the mission, the mission. I think about this, like we’re building a piece of global infrastructure that everybody thinks is a game changer. They should, and I like that,” he concluded. 

The post Killing the Internet from 1050 km Above Ground appeared first on AIM.

OnFinance, a Bengaluru-based finance startup is leveraging AI to bring solutions for the banking, financial services and insurance (BFSI) sector. NeoGPT, the company’s proprietary LLM is specifically built to cater to analysts, advisors, and BFSI companies.

“Recently, we became the first GenAI startup to have NSE [National Stock Exchange of India] as a client,” said Anuj Srivastava, co-founder and CEO of OnFinance, in an exclusive interaction with AIM.  

ChatGPT for Enterprise Finance 

OnFinance’s use cases cover an entire gamut of financial services for enterprises in the form of financial research, equity research, AI copilots for underwriting, and index research. Furthermore, there’s relationship management, compliance and AI copilots for wealth advisory. “Think of it as ChatGPT, but for enterprise finance,” said Srivastava. 

Working within the critical finance sector, where customers are wary of their data being used for AI, OnFinance ensures that the criticality is maintained. Srivastava points this out as the primary reason for enterprises to shy away from ChatGPT APIs. 

“All the financial clients in India currently want on-prem solutions because they don’t want their data to go out of their cloud or infra. That’s why they are still not able to use OpenAI, APIs and so on,” said Srivastava. 

To ensure data security, the company ensures that no customer data is used for training. “We don’t use any data for training, fine tuning or even feedback loop analysis. We actually pass the data on a real time basis to a copilot, so that, you know, it’s not used for those purposes.” 

Accelerating AI Startups

With the option of GPT Stores and similar models, where companies can train and fine-tune their own models with their proprietary data, finance startups such as OnFinance find better relevance. 

 “What happens is that they [models built on GPT Store] are not scalable, and can’t be productionised. The main reason why they are not implementable is that you can’t directly access those APIs after building a GPT. Even if you are able to access them, they have a rate limit, so you won’t be able to productionise them,” said Srivastava. 

“In GPT Stores, any data or question that your analyst would ask, will go directly to OpenAI,” quipped Srivastava. 

Furthermore, big players such as Microsoft will never be a direct threat to these companies simply because it only boosts their business. Srivastava explains how, just like them, there would be 20-40 companies building domain-specific models and would approach Azure for cloud and GPU compute. 

“So the revenue that they would be able to earn from these 40 companies would be way higher than from building their domain-specific model and selling them as copilots,” he said. 

Interestingly, the startup has cloud partnerships with AWS, Google and Microsoft Azure. NeoGPT is also available on AWS Bedrock. 

Open Source is Still The Way

While data security is a priority, OnFinance’s proprietary models are built on open source models that are publicly available, namely Code Llama by Meta AI and Mistral 7B by Mistral AI. “On top of these two models, we fine-tune a lot of financial data sets such as AGM reports, credit reports, compliance circulars, and more,” said Srivastava. 

In addition, instruction tuning is done to the Q&As formulated with the information of 25,000 news websites related to finance, including Moneycontrol, Financial Times, and others. These datasets are made on a Q&A basis.  “So, that’s why it’s very intelligent when it comes to financial data,” he said. 

Partnerships and JioGenNext

OnFinance has already partnered with 15 of the biggest banks and wealth management players in India, including ICICI and exchange platform Centrum. It recently tied up with NSE, Oister Global, and Let’s Venture. 

The startup is also part of JioGenNext, the renowned startup accelerator program that promotes GenAI startups. Through this program, OnFinance will be able to provide their services to big tech clients of JioGenNext.

Srivastava, a BITS Pilani alum, co-founded OnFinance with Priyesh Srivastava, who is also from the same college. Last November, the startup received a seed funding of $1.1 million, and both founders were named in the Forbes 30 Under 30 Asia list. 

The post This Bengaluru-Based AI Startup Offers Secure ChatGPT for BFSI Sector appeared first on AIM.