Generative AI Startup Imbue Raises $200M for ‘Reasoning’ Models
Generative AI startup Imbue has raised $200 million in a Series B funding round from Astera Institute, Nvidia, Cruise CEO Kyle Vogt, and Notion co-founder Simon Last. The new funding also marks the large language model research group’s rebranding from Generally Intelligent and elevates its valuation to $1 billion and unicorn status as it works on creating and embedding reason and other basic intelligence functions in AI.
Imbuing Reason
Imbue’s interest in the core building blocks of intelligence evolved from the company’s initial work evaluating how AI performs in three-dimensional game environments. Imbue now centers research on the biggest LLMS, those exceeding 100 billion parameters. The company believes models optimized for reasoning tasks can drive breakthroughs in general artificial intelligence. Using a 10,000 GPU cluster from Nvidia, Imbue is training and experimenting with massive models on proprietary benchmarks measuring logical deduction and other feats of reason.
Imbue’s main focus right now is on teaching AI to code. The company pointed to programming as a rigorous testbed for improving reasoning, one that allows models to take useful actions on machines. Researchers also build custom debugging tools and visual interfaces for interacting with AI. Coding has already proven a valuable application of generative AI as demonstrated by Microsoft GitHub Copilot, Google’s Codey, Amazon’s CodeWhisperer, and Hugging Face’s Starcoder, among others. But, Imbue aims to push beyond task-specific applications to more broadly applicable techniques.
“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,” Imbue explained in a blog post. “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.”
Cash Foundation
Imbue believes focusing on rigorous training regimes and optimized neural architectures can yield models that avoid pitfalls like hallucination while operating safely under a wider range of conditions. The company is also studying how large language models learn from data and human feedback. By investigating the inner workings of systems like GPT-3, it seeks foundational insights to inform next-generation designs. Imbue views its current work as laying the groundwork for customizable AI that users can fine-tune with limited data to handle specific tasks. It aims to eventually enable a platform where people can build their own models from the bottom up.
The new capital is ten times what Imbue raised before bringing investment to $220 million. Imbue fits among the generative AI companies looking at the bigger picture for the technology and what it’s capable of. For instance, Context.ai recently raised 3.5 million to analyze how enterprises embed LLMs, while Modular recently raised $100 million for its efforts at streamlining generative AI infrastructure.
“When we build AI agents, we’re actually building computers that can understand our goals, communicate proactively, and work for us in the background,” Imbue’s team wrote. “Today, we’re glued to our screens because our computers need to be micromanaged—if we’re not in front of the screen, very little gets done. Truly useful AI agents will fundamentally change this, and free us up to focus on the things we really care about.”
Follow @voicebotaiFollow @erichschwartz
