and some change

#ai

2025

• LLM benchmarks like SWE-bench are not trustworthy Jan 8