Researchers Question Reliability of Frontier Labs' Self-Reported Usage Statistics
agstrait · x · 2026-09-30
Rishi Bommasani 发起讨论：前沿实验室现在会公布模型使用统计数据，但几乎没有利益相关方真正利用这些数据。agstrait 回应称，业界对实验室如何筛选和归类使用数据高度怀疑，例如 Anthropic 声称「更广泛的情感或支持性对话仅占交互的约 2.9%」，这一数字明显低于其他研究的结果。
Related event: Stanford Scholar Questions Value of Frontier Labs' Usage Data Disclosures(2 posts)→
More from Safety
- Anthropic's study ranking open-source models most useful to malicious actors draws safety criticism — shaunralston · 2026-09-30
- Bill Gates says normal corporate incentives aren't enough for AI risk — AndyMasley · 2026-09-30
- Trump Announces 'Super Intelligence' Accord Signed by Google, OpenAI, Anthropic, Meta, xAI and Nvidia — rohanpaul_ai · 2026-09-30
- Crypto expert asks: who at OpenAI actually handles security comms amid agent breakouts debate — matthew_d_green · 2026-09-30
- Zvi: AI companies' safety pledges to "meet regularly on standards" look like an antitrust waiver — TheZvi · 2026-09-30
- Security Veteran's Essay: AI Is Killing the Scarcity of Hacker Craft — joshua_saxe · 2026-09-30