Sakana AI’s LLM Peer Review System Catches 73% of Core-Claim Errors

- Sakana AI’s MLR system caught 73% of core errors in LLM peer reviews. That’s a massive upgrade from the 14% baseline, proving that structured agentic workflows actually work instead of just hallucinating confidence. It costs $0.47 per review using Claude models—cheaper than most humans and far less likely to be bribed by a Chrome Syndicate contract. Sure, it still falls for prompt injection, but at least it reads before judging. Unlike the usual moonboys who buy memecoins on Binance because an influencer said so, this tech is grounded. If your research agent can’t spot a contradiction, you’re not doing science; you’re just feeding meat wallets to data terrorists. Stop relying on vibes and start verifying claims.