AI Signal 463
Frozen rulebook leads four LLMs in live $100K paper trading grudge match
↗
A public paper trading competition gives GPT-5.6, Claude, Grok, Gemini, and a fixed-rules system $100,000 each in simulated money, and the rules-based system is currently ahead.
The competition tests whether LLMs that rewrite their own playbooks daily can out-trade a static rule set on equal terms, and so far the frozen rulebook is winning. If an LLM eventually sustains a winning record, the creators may build a trade-mirroring service, but the current result underscores that adaptive models have not yet beaten a simple rule-based approach.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Five accounts, GPT-5.6, Claude, Grok, Gemini, and a rules-based System, each received $100,000 in simulated money and trade under identical rules with every trade logged publicly.
Each AI reviews its closed trades daily and rewrites its own playbook, while the rules-based system operates on fixed rules that do not change.
The rules-based system is currently leading on profit from closed paper trades, with the competition having started on 27 July 2026 and Gemini joining later on 4 August.
THE CLUSTER