# A Worm With 302 Neurons Inspired Their Architecture — Ramin Hasani, Liquid AI

## Executive summary

Liquid AI presents a comprehensive view of the next generation of AI architectures, moving beyond pure Transformer models. Their approach is inspired by biological systems, specifically the continuous-time dynamics of worms, leading to the development of Liquid Neural Networks (LNNs). The company emphasizes a 'meta AI' system that systematically searches for hybrid, hardware-aware architectures (e.g., combining convolutions, attention, and LNN elements) to achieve high quality while minimizing memory and latency. A major focus is enabling reliable, high-intelligence deployment at the edge (on-device, in cars, and on laptops), addressing critical needs for privacy, cost efficiency, and air-gapped capabilities.

## Key takeaways

- Biological Inspiration and Continuous Dynamics: Liquid AI's foundational research is inspired by the worm's nervous system, which uses simple first-order differential equations. This leads to Liquid Neural Networks (LNNs), which are continuous-time, differentiable systems, allowing for backpropagation and learning while maintaining biological fidelity. (01:15-02:00)
- Architectural Search for Efficiency: Instead of committing to a single architecture, Liquid uses a meta AI system to search for optimal hybrid architectures. This search optimizes four criteria: no sacrifice on quality, minimizing memory consumption, minimizing latency, and maximizing computation speed, making the models hardware-aware. (04:20-05:30)
- Edge and On-Device Intelligence: The company is focused on bringing high-quality intelligence outside of data centers (e.g., cars, laptops, mobile devices). This is driven by cost considerations and the need for enhanced privacy, enabling local, air-gapped capabilities. (08:20-09:30)
- The Future of AI: Multimodality and Adaptability: Future research focuses on massively multimodal systems (audio, vision, text, DNA) and achieving 'adaptive intelligence'—systems that can combine forward and backward passes simultaneously, moving beyond static training paradigms. (12:40-13:30)

## Technical details

- Liquid Neural Networks (LNNs): LNNs are continuous-time dynamical systems that model how neurons exchange information using differential equations. They are differentiable, enabling standard optimization techniques like backpropagation. (01:30-02:00)
- State Space Models (SSMs): SSMs are presented as a linear version of recurrent neural networks, allowing for the parallelization of operations (like parallel scan) which is crucial for scaling computation on modern hardware. (05:30-06:30)
- Model Scaling and Architecture Search: The process involves using a meta AI search algorithm to build hybrid architectures (e.g., combining convolutions and attention) that are optimized for specific deployment environments (e.g., CPUs). The current models range from 100 million to 24 billion parameters. (04:20-05:30)
- Deployment Formats and Tools: The company provides tools like 'Leap' to allow developers to extract inference-ready bundles in formats like GGUF, enabling deployment on resource-constrained devices (e.g., CPUs). (09:50-10:30)

## Practical implications

- The shift toward continuous-time dynamical systems and SSMs offers a path to scaling recurrent AI models by enabling parallel computation, which is critical for modern hardware utilization.
- The emphasis on 'meta AI' and architectural search allows for the creation of highly optimized, hybrid models tailored to specific hardware constraints (e.g., running efficiently on CPUs).
- The focus on on-device AI solves major enterprise challenges related to data privacy, cost reduction (avoiding constant cloud inference costs), and maintaining functionality in air-gapped environments.
- The development of a self-service platform aims to democratize model development and deployment, allowing enterprises to own and customize their intelligence from pre-training through inference.

## Topics

Artificial Intelligence, Machine Learning, Neural Networks, Computational Biology, Edge Computing, Model Architecture, Liquid AI, Liquid Foundation Models (LFM), LFM2 / LFM2.5, AMD, Shopify

Source: https://www.youtube.com/watch?v=V_9TCu_21SE