Advanced Minds: Exploring Imbue's $200M Investment in AI Reasoning

0:00So the big headline here is that MBU, which is an AI research lab that was formerly known as generally intelligent, has recently secured$200 million in Series B funding. So this is pushing the company's valuation beyond the billion-dollar mark, you know, the coveted unicorn status. It's kind of interesting because right now we're seeing overall in the tech and the software, you know, the SaaS landscape, a ton of companies that once were considered unicorns or billion-dollar valued companies and above falling below that billion-dollar valuation because they raised money at, you know, sky high valuations back in 2021.

0:35In any case, Imbue has a bunch of prominent investors in this round that include, um, Astera Institute, Nvidia, Crew CEO, Kyle Vogt, and also Notion co-founder Simon Last. So with the latest infusion, Imbue's total capital raise now stands at$220 million. This is pretty impressive, right? This is a 10X all the money, but like previous they'd raised 20 million. So 10X everything they'd previously raised. And this is really positioning among the better funded AI startups like AI21 Labs, Cohere, Adept, Anthropic, so a lot of these bigger ones. In a recently published blog post, Imbue said, quote, this latest funding will accelerate our development of AI systems that can reason and code so they can help us accomplish larger goals in the world.

1:21The company maintains its aim to kind of develop AI agents that are not only capable of complex tasks, but are also practical and safe for real world applications. So actually last October, imbue emerged from stealth mode with a fairly bold agenda, which was to explore the underpinnings of human intelligence that AI has yet to replicate. And initially, the idea was to construct an array of tasks grounded in these, you know, quote unquote, fundamentals, and then design AI models to tackle these tasks in intricate 3D environments crafted by the team. However, the company strategy seems to have evolved a bit.

2:00So instead of focusing solely on AI in 3D worlds, Imbue is now working on models that are internally useful, such as those capable of coding in a fashion similar to, you know, like pretty much GitHub Copilot or Amazon Code Whisper, those kind of things. This is really interesting. I mean, you got to give them credit. They were doing this, you know, they came out of Stealth last October, pre-ChatGPT launch. So they're a pre-ChatGPT hype kind of engine. So what really sets Imbue's models apart? According to the company, it's their ability to reason robustly. And in their own words, they said, quote, we believe reasoning is the primary blocker to effective AI agents.

2:39Robust reasoning is necessary for essentially effective action. And this involves an array of skills such as grappling with uncertainty, adapting approaches and making informed decisions, all which are essential for navigating the intricacies and often unpredictable nature of the real world. So I think what's cool here is they're taking a completely different approach than a lot of AI models, right? They're not just another LLM that's able to generate the text. And I think this is kind of a good move on their part because there are a lot of even open source or free LLMs that are coming out. And I think that that area will be a little bit, it's gonna be a little bit, like there's gonna be a lot of players in that space in any case.

3:24So I think what they're doing by taking a new approach to this is really smart. They're essentially looking at things that AI models typically struggle with that humans are able to do. So Code, Imbue argues, serves as more than just a tool for building scalable AI applications. They should believe that code can actually enhance an AI model's reasoning ability. So this is a really interesting concept. And this kind of echoes a couple similar sentiments. We have people at Adept and Google DeepMind that also say that. And companies also invested in creating AI systems that can autonomously control software processes or execute tasks based on human interaction.

4:05So in technical terms, Imbue is investing in, quote, very large models with over 100 billion parameters. And these models are fine tuned to perform exceptionally well on the company's internal benchmarks for reasoning. Notably, the models are being trained on a compute cluster that NVIDIA co-designed, housing, you know, 10 ,000 GPUs from NVIDIA's H100 series. And the thing to note here, right, is they're partnering with NVIDIA, so they're definitely going to be able to get their hands on some of these GPUs that are incredibly powerful and are going to help their model. So I think, you know, from their team's perspective, I think they're doing a fantastic job partnering with the right people, taking the right approach and building something really novel and useful in this space that a lot of people I think will really appreciate when this launches.

4:53So besides working on these models, though, imbue is also, you know, directing efforts towards creating its own suite of AI and machine learning tools. So this includes some a some AI prototypes for debugging and also visual interfaces for AI models. And the aim is not just to, you know, simply bring these developments to market, but to also use them as a stepping stone for more generalized future ready AI systems. So that's kind of interesting. It's not kind of the end goal. It's just a stepping stone. I think looking ahead and view aims to lay the foundation for a platform where users can construct their own custom models.

5:28They said, quote, ultimately, we hope to release systems that enable anyone to build robust custom AI agents that put the productive power of AI everyone's fingertips, end quote. So the company further noted in the blog post, kind of emphasizing that and focusing on that specifically, but I think with the latest funding round, Imbue aims to really expedite the realization of these ambitious objectives, and I think it's going to be a very interesting company to follow, considering the amount of money they've raised, what they're doing, and some of the really incredible partners that they currently have.