tech
Inside AI umpire METR, a tiny lab where $500,000 salaries aren't enough to fix a talent shortage
METR, a Berkeley lab started by an ex-OpenAI researcher, has become a prominent AI umpire. CEO Beth Barnes says AI safety research needs more people.
TL;DR
- Beth Barnes, a former OpenAI researcher, founded METR (Model Evaluation and Threat Research) in 2022 to provide independent answers on AI capabilities.
- METR works with major AI companies like OpenAI, Anthropic, Google, and Meta, but faces a significant talent shortage, unable to hire enough researchers.
- The lab's high salaries, reaching $503,000, have not solved the talent bottleneck, highlighting the intense competition for AI researchers.
- A recent security incident involving OpenAI and Hugging Face has raised further questions about AI evaluation and control, with over 1,300 employees signing a letter warning of AI development outpacing control.
- METR's research shows AI capabilities doubling approximately every seven months, and the nonprofit emphasizes the need for greater investment in AI steering and evaluation.
- METR aims to remain independent from AI labs to produce unbiased research, although it accepts compute grants and collaborates on analyzing unreleased models.
- Proposed AI regulations could involve mandatory safety audits from external organizations, a role METR could potentially fill, which might also alleviate hiring challenges by increasing the sector's authority.
- Despite current challenges, there is optimism that more people are becoming concerned about AI safety and regulation.