Fruit-fly brain wiring beats a Random Forest at Bitcoin trading, +4.32% vs +3.62%
johnseach · x · 2026-09-12
Oz Kilim's team at Tropiflo ran a simulated fruit-fly brain as an Echo State Network and let it trade Bitcoin with $1M for a day and a half — beating a standard ML model.
Setup
- Built from the real FlyWire connectome, distilled to 7,500 neurons and 323,000 connections; only the final readout layer was trained
- Raw 1-second Bitcoin returns as input — no moving averages, no volume, no feature engineering
- 7 days of Binance BTC/USDT data (5.6 days train, 1.75 days out-of-sample); 0.1% fees, max one trade per minute
Results
- Random Forest with 23 technical indicators: +3.62%, 1.06% drawdown
- Control network with shuffled wiring: +3.78%
- Intact fly wiring: +4.32%, 0.88% drawdown
Key insight: the control is the interesting result — same size, same training, worse wiring, worse performance. The authors argue the fly's topology gives the reservoir a useful memory spectrum and acts as built-in regularization. Evolutionarily, a fly doesn't forecast; it reacts fast enough to stay alive inside noise — a habit the team says may serve quant models better than another layer of features. Code is open source at TropiFloAI/thebuzz. Short window, simulated fills, no slippage modeling — not a live track record.
Related event: Fly Brain Connectome Trades Bitcoin, Beats ML Models(2 posts)→
More from Research
- If AI can produce correct proofs cheaply, what still matters? A mathematician's take — RexDouglass · 2026-09-12
- Would a yes/no oracle for theories be useful? Yes — an answer alone collapses the search space — basedjensen · 2026-09-12
- Meta paper: adversarial persuasion flips 62-91% of LLM judge verdicts, 70% of flips drift from ground truth — rohanpaul_ai · 2026-09-12
- Navier-Stokes AI proof took 10,000 agents, 88 hours and 130B tokens — not superintelligence — Healthy_Outcome7897 · 2026-09-12
- VIGA agent rebuilds images into editable Blender scenes via multimodal inverse-graphics loop — Michael_J_Black · 2026-09-12
- How Can LLM RL Work Despite Information-Theoretic Inefficiency? A Deep Dive — nrehiew_ · 2026-09-12