Debates
Mercor says AI benchmarks miss 90% of American working hours
Developing 3 sources
Mercor's provisional map of how 29 selected AI benchmarks cover tasks across the US economy.
What happened lately
The Benchmark Map connects prompts from a selected index of 29 benchmarks to O*NET occupations and tasks. Mercor uses model-generated task mappings and reports human review for a subset of its APEX benchmarks. Coverage is a measure of representation in this index, not an assessment of whether an AI system can do an entire job.