Happy Hump Day, readers.
Heard of the UC Berkeley Artificial Intelligence Research (BAIR) Lab?
If so, great. If not, here’s what you need to know, long-story-short style:
The BAIR Lab has brought together UC Berkeley researchers with expertise in computer vision, machine learning (ML), natural language processing (NLP), robotics and other areas of study; there’s a total of 50+ faculty and 300+ graduate students and postdoctoral researchers, in fact. Together, they’re pursuing research on fundamental advances a.) in all of the above fields, and b.) in specifically human-compatible AI, multimodal deep learning, and connecting GenAI with other scientific disciplines and the humanities, in general.
Now, let’s talk about Letta, a new GenAI startup spun out of BAIR.
Letta was founded by Charles Packer and Sarah Wooders. The two met as Ph.D. students at the Sky Computing Lab at Berkeley, where they shared an adviser. Packer studied reinforcement learning, and Wooders focused on AI and data systems. Despite working in different fields, they were interested in experimenting with chatbots. Therein, they realized that the universal constraint of virtually all language models was limiting chatbots, even ones that were already considered to be immersive and effective.
That constraint?
Memory.
So, how could memory be managed in a way that enabled AI agents to be truly useful; so they could “remember” and gain utility over time?
Enter MemGPT, introduced by Packer and Wooders.
“At Letta, we believe that programming agents starts with programming memory,” the company wrote, as part of its overall mission statement. “And we’re putting our deep expertise in research, systems, and AI behind building the stateful APIs of the future, starting with memory management.”
So, that’s the rundown on BAIR, Letta’s history, its mission, and MemGPT.
Let’s toss all these together, shall we?
Letta officially emerged from stealth and announced a $10 million seed round. This was led by Felicis, with additional participation from Sunflower Capital and Essence VC. Other notable angels, per the announcement, included Jeff Dean (Chief Scientist at Google DeepMind), Clem Delangue (CEO of HuggingFace), Cristobal Valenzuela (CEO of Runway), Jordan Tigani (CEO of MotherDuck), Tristan Handy (CEO of dbt Labs), Robert Nishihara (co-founder of Anyscale), and Barry McCardel (CEO of Hex).
“Before MemGPT,” the company wrote, “the default information architecture of most AI agents was stateless, meaning that AI agents weren't able to retain their memory (i.e. state) between user sessions.” But if an AI agent has a self-editing memory — enabling it to update on its own so it can learn over time as it interacts with human users — relevant context bolstering its understanding will allow it to become more useful and applicable in a variety of real-world scenarios.
"The most powerful characteristics of a useful AI agent – personalization, self-improvement, tool use, reasoning and planning – are all fundamentally memory management problems," Packed explained. "The key challenge with agents is understanding how to construct the context window of the LLM, which forms the AI's 'memory', and developing an agentic loop around the LLM to manage the context window over time."
Packer clarified how “Similar to how a microprocessor or CPU is just one part of a computer, LLMs are just one part of larger AI systems. At Letta, our vision is to build the complete AI computer around the LLM.”
Looking ahead, Letta plans to use this new funding to continue fine-tuning MemGPT and to build a new hosted product called Letta Cloud.
"AI memory shouldn't be a black box. We want to make sure developers have full visibility into the memory and state of their agents, and have full control over what LLMs they want to use,” Wooders added, in closing. “Developers shouldn't have to choose between performance and model lock-in".
Developers can install Letta's open source software and sign up for the hosted beta here.
Read even more about Letta’s announcement here.
Be part of the discussion about the latest trends and developments in the GenAI space at Generative AI Expo, part of the #TECHSUPERSHOW experience that will be taking place from February 11-13, 2025 in Fort Lauderdale, Florida.
Edited by
Greg Tavarez
