FlyGPT trains a language model on a real fruit-fly brain connectome topology
QuixiAI · x · 2026-09-14
QuixiAI released FlyGPT, a character-level language model whose recurrent architecture is a real subgraph of the fruit-fly central-brain connectome (MaleCNS v1.0): each biological neuron is a scalar hidden state, each real synapse a trainable weight.
Key points:
- The fly's wiring fixes the topology; gradient descent learns only the connection strengths
- Trained from scratch on Tiny Shakespeare and compared against degree-preserving scrambled-connection controls across paired seeds, asking whether real brain wiring learns better
- Unlike the earlier frozen-reservoir approach (ngxson/fly-llm-hf), it trains every synaptic connection
- Open source on GitHub with weights on Hugging Face (cc-by-4.0), deterministic graph extraction
Related event: QuixiAI Open-Sources FlyGPT, a Language Model Built on Fly Brain Connectome(2 posts)→
More from Research
- Proxy Policy Steering adapts frozen VLA models to new tasks at inference time — weichiuma · 2026-09-14
- New research: standard SGD matches AdamW for LLM RL training, with far less memory overhead — zhaoran_wang · 2026-09-14
- RSI work separates practical harness self-improvement from unproven intelligence explosion — arthurcolle · 2026-09-14
- Amazon Proposes Query-Aware Index Pruning to Optimize Retrieval Under Budget Constraints — _reachsumit · 2026-09-14
- New Paper Finds Retrieval Signals Give No Reliable Routing Gain in Adaptive Multimodal RAG — _reachsumit · 2026-09-14
- Google: Graph RAG Cuts API Hallucination Rate from 56.4% to 16.2% in Java-to-Python Migration — _reachsumit · 2026-09-14