Benchmark矛盾基准 / Contradiction Benchmark
Contradiction Benchmark
An academic paper automated review benchmark dataset constructed by Sakana AI, comprising 1,164 carefully designed error cases. It is specifically intended to evaluate the capability of automated systems in detecting factual contradictions and logical conflicts within academic papers.
Timeline (last 90 days)
Oct 11
The research team constructed a test set called the Contradiction Benchmark, containing 1,164 deliberately designed error cases.
Unverified50%