Imbue Raises $200M to Build Foundation Models that Reason
The new unicorn valuation is driven by the promise of an autonomous agents
Imbue announced that it has raised $200 million from Astera Institute, NVIDIA, and other investors. The new funding brings Imbue’s valuation to over $1 billion. The company is also taking the opportunity to rename itself from Generally Intelligent to Imbue. According to Imbue CEO Kanjun Qui:
The modern-day is not that different from the 1900s. Back then, we didn't have machines to do physical labor, and today, we don't really have good machines for doing mental labor.
We have computers, but they need to be micromanaged. It's hard to even imagine our computers accomplishing larger goals like planning a family vacation or making a PowerPoint deck without us — basically doing all the work ourselves.
Reasoning Applied to Agents
Imbue raised a $20 million Series A round that was announced in October 2022 and included OpenAI as an investor. The company was originally focused on, “testing different systems’ ability to learn to solve them in highly complex 3D worlds.,” according to a TechCrunch article. However, the new announcement clarified the company’s current focus on developing AI-based agents that can reason and complete tasks:
Our goal remains the same: to build practical AI agents that can accomplish larger goals and safely work for us in the real world. To do this, we train foundation models optimized for reasoning. Today, we apply our models to develop agents that we can find useful internally, starting with agents that code. Ultimately, we hope to release systems that enable anyone to build robust, custom AI agents that put the productive power of AI at everyone’s fingertips.
Reasoning is a tough problem for generative AI models. As capable as they often seem, multi-step decision-making, even with limited variables, is unreliable. In addition, agents that can learn objectives and apply them in various situations by employing reasoning are another gap we have in creating digital assistants that mirror or exceed human capabilities.
Do you remember Google Duplex? That is an example of an agent that needed to manage both natural language communication and reasoning. And it didn’t always work as planned. You don’t hear about Google Duplex anymore precisely because this is a hard problem—likely harder than Google anticipated.
A Path to Autonomous Agents
OpenAI is conducting its own research into fine-tuning large language models (LLM) to reason more effectively and reliably. We have also seen projects such as AutoGPT and Adept that are attempting to mix intent recognition, reasoning, and agency. These have generated broad interest because the problems are hard and the benefits significant.
Imbue is focused on a similar target, but it is starting with a different focus on building agents that code. This sounds somewhat similar to Viv Labs's approach in its virtual assistant platform that became Samsung Bixby. That solution writes real-time custom programs to invoke applications based on user intent. While it did not employ foundation models, you can see some alignment in objectives.
Generative AI enabled copilots are already delivering significant benefits. However, reasoning-capable agents will provide even more value to users in business and consumer settings. In addition, the application surface area is very large for these agents, so don’t be surprised to see many companies emerge in this space, each with different specializations. Whomever can make this work will be headed for a valuation far in excess of $1 billion. Maybe it will be Imbue.