Noam Brown is the author of "Implications of Large-Scale Test-Time Compute," an X Article arguing that frontier-model capability measurement should treat inference budget as a first-class variable rather than reporting single scalar benchmark scores.source: noam-brown-test-time-compute-evaluations-2026.md
The article connects test-time-compute-evaluations to AI preparedness: if model capability increases with tokens, dollars, or wall-clock time at inference, then safety evaluations and release thresholds need to specify and vary the inference budget they test against.source: noam-brown-test-time-compute-evaluations-2026.md
Related pages: test-time-compute-evaluations, agent-loops, harness-engineering.