SuperAGI wants to build a full-stack artificial general intelligence (AGI) platform based on large-scale agent models (LAM). It is seen by some as the next stage in the development of large-scale language models. The research is currently attracting attention from around 20,000 developers, including people at Microsoft, Google, Tencent, Tesla, and JPMorgan Chase, as well as OpenAI, using tools that have already been released. They're raising money from high-profile investors who aren't on the radar.
Newlands VC, a private company backed by WhatsApp founder Jan Koum, is leading SuperAGI's $10 million Series A, hoping the funding will help further research and bring AI closer to broader applications. middleware and software applications. Realize a wide range of use cases more reliably.
“People are realizing that ChatGPT and LLM are great. [when it comes to] Content creation. But when you actually need to take action, most LLMs fail there. That is what we have solved with his SuperAGI,” co-founder and CEO Ishaan Bhola said in an interview. “There are a lot of these frameworks on the market, but none of them really work to solve this problem.”
We also understand that large companies that are already making strides in the AI space are also considering investing in SuperAGI, but are refraining from securing strategic support at this stage.
Money is pouring in at a funding crossroads for AI startups. According to PitchBook, his investments in AI startups actually declined in the fourth quarter, even though generative AI remains a hot topic.
The number of transactions decreased by 19% from 8,968 in 2022 to 7,238 in 2023. Similarly, the dollar value of AI funding announcements has declined from $96.6 billion in 2022 to $90.9 billion in 2023 (obviously, in 2022, large companies like OpenAI monster round). Nevertheless, major tech companies, including Microsoft and Google's parent company Alphabet, continue to invest in AI startups to expand their gaming reach. PitchBook recorded his $22.3 billion worth of investments in AI startups around the world in the fourth quarter.
Little is known about Newlands VC. This November 2023 Forbes profile is the first real light on the Dallas, Texas-based company run by former Sequoia partner Michael Abramson. Prior to that, it had been mentioned at least once or twice as an investor in 2023 enterprise startups.
TechCrunch understands that Newlands has amassed significant stakes in large publicly traded technology companies alongside corporate investments, while also actively evaluating AI startups. SuperAGI appears to be the fund's first investment from it.
Although AI has been around for years, the huge success of generative AI tools like ChatGPT and Midjourney has catapulted AI technology into the public consciousness.
However, as many are aware, LLM is far from perfect and can produce inaccurate results for certain things, which is why researchers and engineers are using it in their ongoing exploration of AGI. We are finding ways to improve.
While OpenAI, Anthropic, and Google are among a wave of large (and very well-funded) tech companies chasing that windmill, SuperAGI is one of the smaller but equally ambitious players, with its own We want to participate in the competition to open up the world.
The startup, headquartered in Palo Alto and also based in Bengaluru, India, began its work in early 2023 by focusing on agent research and building autonomous agents.
But as GenAI continues to evolve, the startup has turned its attention to agents. The first product for marketing use cases is called Contlo, which serves as an all-in-one campaign and automation tool, allowing businesses to manage all their core marketing operations using a single solution. This allows you to run campaigns by email, as well as SMS or instant messaging apps, as well as providing personalized customer follow-up and analysis.
Contlo was initially designed for Indian D2C brands and Shopify customers. However, since its launch, it has gained traction with more than 1,000 new and established brands around the world, the startup said.
SuperAGI's latest app is called Verk, which sounds a bit like “work”, but that's what it's all about. Verk is pitched as an AI employee who can work as a salesperson, graphic designer, market researcher, or social media manager (interestingly, this is also the case in India, which recently raised funding It is also the focus of Ema, another AI startup with roots). Employers can access their AI employees on the go using their existing platforms like Slack, the web, iOS or their Android devices, without worrying about attrition or onboarding.
Bourla told TechCrunch that the startup initially investigated most of the available agent frameworks, but none of them worked to solve the problem. We then decided to delve deeper into our research efforts to find answers to questions such as how to use agents to run software and perform useful tasks such as sales and marketing activities. The startup eventually started building his LAM in-house to use his AI for specific actions.
Bhola co-founded SuperAGI with Mukunda NS after working at Sachin Bansal's fintech startup Navi.
After researching and building LAM, SuperAGI open-sourced its infrastructure, including LAM, an autonomous agent framework, and training datasets, allowing companies to build agent systems based on their requirements.
Open sourcing its infrastructure has helped the startup attract attention from market giants such as Microsoft, Google, Tencent, Tesla, and JP Morgan Chase, as well as AI companies such as OpenAI. The startup's frameworks and research are used to build a variety of apps, and it has a community of over 20,000 members.
Last year, SuperAGI released the SAM-7B as a small-scale agent model for the open source community. It claims to outperform larger models such as Microsoft's Orca-2 (17B) and OpenAI's GPT-3.5 (175B) on inference benchmarks such as: ARC-C and GSM8k. Although SAM-7B was trained on a 97% smaller dataset, it outperformed Orca-13B in GSM8k, the company said. The reason for this success, Bhola said, is due to SuperAGI's different architecture and data training process.
SuperAGI also introduced VEagle as an open source vision model in January of this year. It outperformed open source models such as Bilva and Llava on various vision benchmarks. VEagle can improve text understanding, image interpretation, and perform actions on certain software just like a human would. For example, you can use Vision Model to set up a Google meeting without requiring your users to go through any process or integrate APIs. This model uses the open source OCR model YOLO to identify how it should interact with the user interface.
“The parameters I added and more computing power equaled intelligence that turned out to be untrue,” he said. “We strongly believe that his current LLM architecture is just the beginning, much like when the Internet came out and everyone thought this would be the single business model. I am.”
Agent models typically rely on third-party APIs to perform actions. However, Bhola pointed out that the challenge with using APIs is that they limit what a model can do independently. In some cases, the software may restrict the agent's access once the agent begins progressing. VEagle helps in such cases, as it avoids the use of APIs and makes the automation even more robotic.
“We are currently integrating with some other APIs. But at the same time, we are also solving problems. [of software restricting API access] using our vision model,” Bhola added.
SuperAGI contributes to the open source community and generates revenue through its apps while providing infrastructure to a wide range of users, including major technology and AI companies. For example, we work with AutoGPT, Microsoft, and OpenAI to create agent standards and establish benchmarks for new agents.
This approach will ultimately help SuperAGI solve AGI and do it in a very full-stack way, Bhola said.
“Hey, we call ourselves full stack because we have a very core research engine, but we don't monetize it. We have the infrastructure, but we are monetizing the application layer,” he said.
SuperAGI also works with various companies to build their business apps on top of its agent framework. Currently, over 150 apps are available through this model. These apps act as software agents and serve a variety of fields, including QA testing and autonomous drone mapping.
However, since SuperAGI is open sourcing its LAM infrastructure and framework, the startup's agent technology could be used in several more applications.
Companies like Rabbit and T-Mobile have recently shown how LAM can be useful and part of the future of AI. However, that may be just the tip of the iceberg.
“AI is useless unless it can actually take action,” says Bora. “LLM is going to be one of the first primitives. This abstraction of LAM is basically next year to this year he's going to take over LLM. That's exactly what we're solving for. is.”
So far, SuperAGI has attracted “more than 1,000” paying customers across its Contlo and Verk apps. The startup is looking to quickly expand its numbers by introducing a marketplace called Verk Store to hire AI employees. The company has already started working with EY to train AI auditors for small and midsize companies.
Ultimately, Bora told TechCrunch that SuperAGI wants to be one of the top three AI research companies in the world by 2024. The startup plans to continue to open source its development while monetizing its app.
“Our research, frameworks and models have always been open source, in contrast to other companies that started with an open source model but gradually became completely closed source,” the co-founders said.
SuperAGI raised a total of approximately $15 million in an all-stock Series A round that valued the startup at “less than $100 million,” and attracted investment from top AI researchers and operators such as Meta and Apple. Also participating was his Kae Capital, an existing investor in the startup.