Parv Mahajan discusses a METR pilot assessment of AI research and development accelerationAI-assisted scientific discovery uses AI to support hypothesis generation, experiment design, analysis, simulation, literature work, and interpretation while researchers retain responsibility. at Anthropic. He describes evidence of faster progress on well-specified tasks and says the study’s evidence about causes and broader future acceleration is uncertain.
The interview explores whether measured gains generalize to open-ended researchEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria., how limited and redacted data constrain conclusions, and why self-reported productivity is difficult to interpret. Mahajan says the available methods cannot confidently establish or rule out very rapid acceleration.
Mahajan also discusses METR’s view of Anthropic’s acceleration warning threshold and calls for more public evidence. These judgments are attributed to him and the report rather than presented as settled findings.
Watch on YouTube
