The Allen Institute’s BenchMIRT method reveals that LLM benchmark scores are influenced by safety and reasoning capabilities, challenging traditional evaluations.
How BenchMIRT Reveals The True Capabilities Of Large Language Models In AI
