home | feeds |donate  
impossiblebench: measuring reward hacking in llm coding agents
- Oct 30