Imbue raises $200 Million for AI systems that can reason and code

One of the most intriguing Generative A.I. Startups in the world right now.

Hey Everyone,

I’m obsessed with startups and venture capital related to Generative A.I. and the new sorts of funding rounds that are reflecting the general optimism of what LLMs will be able to do soon.

Imbue used to be called Generally Intelligent, which is a San Francisco based AI research company, raised $200m in a Series B funding round. This is a pretty significant sum relative to its peers.

Generally Intelligent would have been a funny name since AGI is the new watchword of Generative A.I., thanks to the marketing of OpenAI. However Imbue is a bit more vague. They want to imbue real intelligence into code and LLMs that can reason.

Now we are talking! They came out of stealth in October of 2022, just one year ago, so that they managed $200 million in a Series B truly is astounding. The CEO is Kanjun Qiu, maybe you have heard of her?

Her X profile reads:

understanding human & machine minds to build a creative abundant future.

Well the company now is sitting at a 1B+ valuation to develop AI systems that reason! That’s terribly exciting to be honest. They are hiring, they say, both in SF and remotely: https://imbue.com/careers/. They are looking for people who care deeply about reinventing computing, and who enjoy questioning assumptions, building from first principles, and operating with high agency.

This sounds like a Generative A.I. startup about now in 2023.

The previous quote from 2022 is also telling:

Is this another AGI company?

“We believe that generally intelligent computers will someday unlock extraordinary potential for human creativity and insight,” CEO Kanjun Qiu. “However, today’s AI models are missing several key elements of human intelligence, which inhibits the development of general-purpose AI systems that can be deployed safely … Generally Intelligent’s work aims to understand the fundamentals of human intelligence in order to engineer safe AI systems that can learn and understand the way humans do.”

I won’t even tell you how important it is that there is such a company, a startup that is woman-led.

Building the Future of AI

Who Invested in the Series B?

Among those participating are the Astera Institute, Nvidia, Cruise CEO Kyle Vogt and Notion co-founder Simon Last.

Including some unusual investors it was noted on X. They don’t have a commercial product yet, so how do they even raise $200 million? It’s very impressive. Wilson Sonsini Goodrich & Rosati advised Imbue on the transaction.

“Current AI systems are very limited in their ability to complete even simple tasks on their users’ behalf. While we expect capabilities in this area to advance rapidly over the next few years, much remains to be done before AI agents are capable of achieving larger goals in a way that is really robust, safe and useful.”

I actually like startups with big promises, and Imbue is a curious mix of low-key and working on tough really challenging problems.

Companies across the world, from banks to big tech, have doubled down on investments in artificial intelligence, and many of these startups won’t make it but those that are tackling the biggest problems will have an absurd advantage, and I actually consider Imbue fits into that category.

The High-Road Narrative

Instead of pretending we are near AGI like OpenAI, Imbue has taken a more considered human-centric approach to their branding and marketing narrative.

“As part of this, we’re renaming from Generally Intelligent to Imbue to better reflect our focus on imbuing computers with intelligence and human values. We aim to rekindle the dream of the personal computer—for computers to be truly intelligent tools that empower us, giving us freedom, dignity, and agency to do the things we love.”

Generative A.I. badly needs this feminine visionary vibe, as the ecosystem is full up on bro-culture vibes that eventually of course, go absolutely nowhere.

Can LLMs Learn to Reason Better?

Imbue says: “We believe reasoning is the primary blocker to effective AI agents. Robust reasoning is necessary for effective action. It involves the ability to deal with uncertainty, to know when to change our approach, to ask questions and gather new information, to play out scenarios and make decisions, to make and discard hypotheses, and generally to deal with the complicated, hard-to-predict nature of the real world.”

TechCrunch on Founder Backgrounds

Qiu, the former chief of staff at Dropbox and the co-founder of Ember Hardware, which designed laser displays for VR headsets, co-founded Generally Intelligent in 2021 after shutting down her previous startup, Sourceress, a recruiting company that used AI to scour the web. Generally Intelligent’s second co-founder is Josh Albrecht, who co-launched a number of companies, including BitBlinder (a privacy-preserving torrenting tool) and CloudFab (a 3D-printing services company).

I still can’t get over how such a young startup without a commercial product yet could be such a investor darling already. They must be well-connected and then some.

Imbue launched out of stealth last October with an ambitious goal: to research the fundamentals of human intelligence that machines currently lack. It almost sounds like an R&D based startup working on some aspects of AGI. Imbue intends to use the funds to accelerate the development of AI systems.

Rather than unleash AI on 3D worlds, Imbue says that it’s developing models it finds “internally useful” to start, including models that can code. As if we need more Generative A.I. of coding applications, but you have to start somewhere. If only Imbue focused on imbuing game development, or virtual worlds with more responsive immersive environments - and it might every well end up taking that route. If you look at the background of the founders, Imbue could actually go in any number of directions for product-market-fit (PMF). Without a live product, it’s hard to say.

A lot of us have been impressed with the reasoning capabilities of GPT-4 and seemingly the vast improvements in LLMs in relation to coding. Imbue also believes that code is an important use case beyond enabling its team to build AI apps at scale. In the blog post, according to TechCrunch, the company makes the case that code can improve reasoning and is one of the more effective ways for models to take actions on a machine.

But a world full of chatbots or copilots likely isn’t the answer to creating better human experiences. We have to imbue the product of tomorrow with AIs that are able to reason, guide and empower people. Given its humanistic branding, Imbue also seems to be involved or want to be involved in A.I. policy.

• Can we develop theoretical explanations for today’s AI systems?

Imbue appear to be an idealistic group, but on the commercial side it’s still not clear how they will drive revenue. The new breed has that Adept sort of vibe, which aims to build AI that can automate any software process. New ways to use computers? New teammates? It’s hard to tell if this will be just another copilot company or something more significant.

Boosting Productivity with AI’s that have Improved Reasoning Capabilities

Imbue says that its models are “tailor-made” for reasoning in the sense that they’re trained on data to “reinforce good reasoning patterns,” and using techniques that spend “far more compute during inference time” to arrive at “robust conclusions and actions.”

I think they are talking about productivity tools here. You are constantly having to read between the lines here: (What are they actually talking about!)

At Imbue, we create foundation models that are tailor-made for reasoning. This means taking advantage of the powerful capabilities afforded by very large language models, while understanding in a detailed, practical way how those models are trained, and where they fail.

• A Foundational model LLM company (really expensive for compute reasons)

• Towards productivity tools, like in coding

• AI agents that can accomplish larger goals (like?)

Back when they came out of stealth what was apparent was that they are very persuasive founders. While Imbue (formerly called Generally Intelligent’s co-founders) might not have traditional AI research backgrounds — Qiu was an algorithmic trader for two years — they had managed to secure support from several luminaries in the field. Among those contributing to the company’s $20 million in initial funding (plus over $100 million in options) is Tom Brown, former engineering lead for OpenAI’s GPT-3; former OpenAI robotics lead Jonas Schneider; Dropbox co-founders Drew Houston and Arash Ferdowsi; and the Astera Institute.

Astera really stands out as a backer. The round was led by Astera! They were advised by Latham and Watkins. Astera Institute is an investment firm that supports science and technology projects. Berkeley, California, United States. astera.org. Imbue is certainly a project. Latham & Watkins LLP advised Astera Institute in the transaction with an emerging companies team led by Boston partner Dan Hoffman, with associates Naomi Smith, Misola Leeman, and Jake Hummer.

Astera seems to have a particular interest in AGI like technologies.

AGI at Astera

Artificial General Intelligence will be a transformative technology in human history, one that can improve all other areas of human investigation and endeavor. At Astera, we believe AGI can bring good to humanity and that multiple approaches are needed to ensure its success and full alignment with society’s values.

The founder of Astera is Jed McCaleb. There seems to be shared values that builds trust between entities like Astera and a team like Imbue.

The Future, Faster

Astera Institute was created to bring humanity the greatest imaginable good in the most efficient possible way. 

They seem very San Francisco-ish about climate-tech and A.I.

Back when they came out of stealth, it appears that the key value proposition that resonated with investors was also the unique approach to big problems. Qiu said at the time, it’s Imbue’s (Generally Intelligent’s) approach to the problem of AI systems that struggle to learn from others, extrapolate safely, or learn continuously from small amounts of data.

• struggle to learn from others

• extrapolate safely

• learn continuously from small amounts of data

It seems their holy-grail is thus self-learning AI. In Microsoft trying to promote GPT-4 you read a lot about ChatGPT’s so-called ability to reason, but not much about its ability to self-learn or become sentient. Self-learning and sentience do seem way more interesting and important problems to solve than mere common sense and reasoning.

What is Latent Potential?

Latent potential refers to solution ideas positioned at the root of big problems that are hidden or underestimated because of inherent bottlenecks.

So will all this idealism of Generative A.I. startups and their investors actually amount to anything? That is far less clear.

According to the Imbue blog post announcing the financing, “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.” Augmented coding partners or software development buddies are manifold from GitHub Copilot to what Google DeepMind’s Gemini will be capable of doing, to ChatGPT and other more specialized coding partners like Amazon, Meta and so many others are offering.

The official founding of Imbue seems to be around April, 2021. So that they don’t have a live product or any real PMF is fascinating. $200 million isn’t so much when you factor in compute required and talent required for such a serious topic.

How will Imbue liberate people? How do we put any of our tasks on auto-pilot even with all of these Auto-GPTs? LLMs have a long way to go before they are real-world agents. What goals in the world will Imbue empower?

What’s Code Got to do with it?

Working on agents that code is a very preliminary step in all of this but this is what appears will be their first commercial product.

That this area is also saturated with other tools is fairly interesting. How do you compete vs. a Microsoft (GitHub/OpenAI) or differentiate vs. a Meta or a Google?

Adept that’s purposely working specifically on AI to use existing software and APIs feels a bit more utilitarian. I think we can put Adept and Imbue like sister startup that are likely to maybe have some overlap. Adept most recently was able to raise $350 million.

Nvidia now Backs Imbue

One of the more fascinating things about the Generative A.I. movement is how companies like Nvidia, Salesforce and others have become really aggressive in funding A.I. startups and other bets. Chalk another one up for Nvidia here:

You want eclectic backers, how about a crypto Billionaire, people involved in GM’s Cruise, and Nvidia?

Imbue is a fascinating use case of both a Generative A.I. startup at a unique moment in Silicon Valley and a very persuasive female Founder. This is extremely rare in A.I.

“Imbue is one of a growing number of AI startups raising large rounds this year in a reflection of both the frenzy around the technology and the high costs associated with training AI models by feeding them vast troves of online data.” - Bloomberg (via BNN)

But will the rise so-called AI agents capable of reasoning and what will they be able to do besides just generating seemingly mediocre content? What is the ceiling of copilot in an era of chatbots, and auto-gpt hype that usually leads - well nowhere. I want to dream but in studying many hype cycles out of Silicon Valley, unfortunately I know better.

As Imbue rightly states as of 2023: Current AI systems are very limited in their ability to complete even simple tasks on their users’ behalf.

Even LLMs are generally not very intelligent and not imbued with much common sense, reasoning, ability to synthesize, improve or actively research in a chain of steps. Generative A.I. doesn’t seem like tools today that are very helpful at all.

In today’s world I’d put more faith in RPA than GAI accomplishing real-world tasks for the most part.

As the exodus of Adept’s own founding team illustrates, the jumping around of talent into splinter projects is really pretty tremendous in the 2021 to 2024 period. It’s not just compute but talent that the market vs. demand seems to be out of sync. A.I. chips and talent are a real bottleneck for solving and working on these complex problems.

When your biggest investor is a Crypto Billionaire it begs more questions than answers. Is a coding assistant involved in reasoning and improving the ability of LLMs to be more useful? I guess potentially!

“We believe such [agents] could empower humans across a wide range of fields, including scientific discovery, materials design, personal assistants and tutors and many other applications we can’t yet fathom,” Qiu said.

Qui could also be talking about Quantum computers in the above quote. Which also make similar claims. KQ was a big vague (1:27 in the video) with Bloomberg in their interview about the product timeline. With a large Series B, the pressure will be on. When cofounders of Adept are already working on another (also Niki) Stealth startup, you wonder if it’s a bit like OpenAI with Anthropic, with a sequence of startups that results from an original idea or team.

Good reasoning is core to effective AI agents

Imbue’s pitch:

In order to create reasoning models that provide a robust foundation for AI agents, we take a “full stack” approach:

1. Training foundation models

2. Prototyping experimental agents and interfaces

3. Building robust tools and infrastructure, and

4. Understanding the theoretical underpinnings of how models learn.

A Full-Stack LLM Agency Company?

I see the forest with Imbue’s general direction, but I cannot tell what kind of tree they will plant. In the real world, speed to product market fit, is actually a thing.

Source: Lenny’s Newsletter.

So Imbue is essentially still pre launch.

Imbue’s (formerly Generally Intelligent’s) work brings to mind earlier work from Alphabet’s DeepMind and OpenAI, which sought to study the interactions of AI agents in gamelike 3D environments. I’m just not grasping how the R&D translates into products yet with the information that is publically available.

I’m all for 🦄 AI Unicorns with a female CEO, but after covering so many dozens of startups my venture capital analyst brain turns on to various signs and symptoms in the marketing and PR that demonstrate the relative potential for success of the startup.

“Our north star: truly personal computers that give us freedom, dignity, and agency to do what we love”

That it took them this long to figure out that code was their starting point for agency, means they have pivoted already likely more than once and the new round of investors bought the story.

Towards Reliable General Purpose Agents

Imbue is working on RGPAs.