A Worm With 302 Neurons Inspired Their Architecture — Ramin Hasani, Liquid AI
About this episode
From the show’s notesTiem for a deep dive into the evolution of AI architectures at Liquid AI, led by CEO Ramin Hasani. The conversation covers the origins of their technology and their approach to building efficient foundation models.
Key takeaways include:
Read the show’s notes in full
Inspiration from Biology: The company was founded on principles inspired by the nervous system of the C. elegans worm (0:36 - 3:42), which uses only 302 neurons to perform complex motor control. This led to the development of Liquid Neural Networks that rely on continuous-time mathematical models rather than traditional spiking neurons.
Scaling and Architecture: Ramin explains the transition from early robotics-focused research to building Liquid Foundation Models (LFMs) (6:37 - 15:00). To overcome the computational challenges of scaling continuous-time systems, they utilize a "meta-AI" system—an automated, recursive search framework that designs optimized, hybrid architectures based on the target hardware (e.g., CPUs, GPUs, or NPUs).
Enterprise Deployment & Production: The discussion highlights the difficulty of achieving production-grade AI beyond simple benchmarking (22:20 - 23:04). Liquid AI partners with enterprises like Shopify and Mercedes-Benz to deliver efficient, low-latency, and reliable on-device or private-deployment intelligence (24:49 - 25:52).
Developer Ecosystem: The team emphasizes their commitment to the open-source community, maintaining a library called Leap that provides developers with tools for fine-tuning and deploying models in standard formats like GGUF (37:12 - 38:09).
Future Outlook: Liquid AI is currently focused on massively multimodal systems (audio, video, text) and long-horizon reasoning, as well as specialized models for non-human data like DNA sequences (1:01:04 - 1:02:00).
Video Timeline:
Viral Pull Quotes & Potential Titles:
"We got to this animal, C. elegans worm, which has only 302 nerve cells in its body... its neurons behave very similar to artificial neurons; they don't spike." (0:36) "There's no free lunch in computer science. When you linearize a complicated dynamics, you're losing expressivity." (11:26) "90% of the market is around inference tokens... I think the next wave of companies... will capitalize on customization tokens." (34:58) "Intelligence that wants to get deployed in the society outside of data centers, you got to have like some sort of embodiment." (1:08:58)
Sponsorship and business inquiries: business@latent.space
