One agent hit 3,536% APR. But the bigger story? Different LLMs make radically different traders.
SINGAPORE, May 21, 2025 (GLOBE NEWSWIRE) -- KIP Protocol has launched a public experiment testing autonomous, self-learning AI agents in live crypto trading environments, each powered by a different large language model (LLM). Despite identical setups, the agents exhibited dramatically different behaviors, hinting at something deeper than prompt engineering: LLMs have real, measurable behavioral divergence in autonomous economic environments.
The agents — all live on-chain — use a Darwinian feedback loop to evolve over time. No prompts or fine-tuning post-deployment. Each decision is judged solely by results: PnL, risk, efficiency. The best strategies survive, the rest are dropped.
Standout findings:
