San Francisco-based AI startup Imbue has raised $200 million in Series B funding led by nonprofit Astera Institute, valuing the company at over $1 billion. Imbue, formerly known as Generally Intelligent, aims to develop AI “agents” that can reason, communicate, and accomplish complex goals autonomously.
The sizable round gives Imbue access to 10,000 NVIDIA H100 GPUs to power its research. While rivals like OpenAI and Anthropic focus on foundation models like GPT-4, Imbue is taking a different approach, aiming to build virtual assistants that act more like autonomous human researchers.
Imbue is adopting a comprehensive approach to train and build AI systems that are optimized for reasoning. The initiative comprises several layers:
- Models: Imbue has pretrained its own LLM (>100B parameter) models that performs well on internal reasoning benchmarks.
- Agents: These models underpin their experimental agents that are intended for serious contexts of use, and focused on coding tasks at this stage.
- Interfaces: The company is exploring novel user interfaces and experiences aimed at enhancing robustness, trust, and collaboration between humans and intelligent machines.
- Tools: Imbue invests in building its own array of tools, from debugging and visualization interfaces, to systems that automate much of hyperparameter tuning and network architecture search.
- Theory: A commitment to understanding the theoretical basis of deep learning, aimed at establishing models that are not only powerful but safe.
By employing a "full-stack" approach, Imbue aims to create a virtuous cycle where improvements in agents and tools lead to better models, thereby unlocking even more useful agents and further model improvements.
Imbue's agents could potentially help with a range of tasks from biological research to travel planning and coding, executing analysis and follow-up steps without supervision. According to the company, coding offers an ideal test-bed for reasoning capabilities because it is highly objective: code either works or it doesn't. This focus also offers strategic advantages. As agents improve and assume more of the work, they contribute to a faster pace of research and engineering, eventually offering a glimpse into what organizations might look like when augmented by functional AI agents. But the startup remains in the early research stage, with no public demos available yet and just 20 employees, according to Forbes.
Imbue's founders, Kanjun Qiu and Josh Albrecht, aim to build the equivalent of what Xerox PARC was for personal computers. PARC made PCs accessible to everyday users; Imbue wants to do the same for AI agents. “We believe AI has the potential to thin the barrier between ideas and execution,” said Qiu in a statement.
The founders acknowledge commercial applications could be years away. Imbue backer Jed McCaleb said the startup's valuation stems from long-term opportunity. He saw an internal demo persuasive enough to lead his nonprofit Astera Institute to spearhead the round rather than traditional VC firms.
With massive funding and resources, Imbue is making an ambitious bet to realize its vision. But some critics have questioned whether Imbue has the expertise to operate a serious AI research lab, given the co-founders' relatively unconventional paths. Qiu and Albrecht dismiss these concerns, pointing to their team’s diverse academic backgrounds in AI research, neuroscience, and plasma physics as a strength. The founders say they are comfortable with their decision to bypass traditional venture capital avenues for funding, acknowledging that their groundbreaking work could take years to commercialize.
Imbue's massive funding round and valuation reflect the red-hot interest and sky-high expectations around AI. The startup is making a risky, long-term bet that its unconventional approach will pay off where others have failed. With its patient backers and commitment to a comprehensive research strategy, Imbue believes it can overcome skepticism and pioneer the next generation of AI agents. But the startup still has everything to prove. Whether Imbue's ambitions manifest or go down as speculation will likely take years to assess. For now, the AI community will be watching closely as this dark horse tries to make good on its billion-dollar promise.