TLDR Tech / Crypto
Why Public AI Models Won't Give Retail an Edge
Research replicating the NeurIPS best paper "Artificial Hivemind" in a crypto trading context shows that GPT-5.4, Opus 4.6, Sonnet 4.6, Gemini 3.1 Pro, and Qwen3 Max all recommend identical trades (buy SOL) across 150 experiments when given the same 30-second orderbook data – even at high temperature settings. The homogeneity implies that unsophisticated retail traders using basic AI prompts will herd into identical positions, with first-movers profiting and latecomers serving as exit liquidity. Expert traders with richer context may maintain an edge, but public models with basic inputs are structurally unlikely to generate sustained alpha.
Read the full story