Model Evaluation Benchmark
Timeline (last 90 days)
In recent years multiple open-source safety benchmarks have appeared in the industry, but their update speed often lags behind the evolution of model capabilities, and they are easily optimized against.
The true value of independent evaluation lies not only in using benchmarks, but more in whether evaluators can develop and publish new benchmarks, continuously staying ahead of model capabilities.
All Facts (2)
In recent years multiple open-source safety benchmarks have appeared in the industry, but their update speed often lags behind the evolution of model capabilities, and they are easily optimized against.
50%UnverifiedThe true value of independent evaluation lies not only in using benchmarks, but more in whether evaluators can develop and publish new benchmarks, continuously staying ahead of model capabilities.
50%