Research — Paradigm 3
paradigm3.org · 311 words · saved by 1 readers
Original research on AI, forecasting, machine learning and the science of evaluation, from Paradigm 3.
capabilities Two Reports on the OpenAI-Hugging Face Attack Gavin Leech and Lucca Fraser·28 August 2026· 18 min TL;DR; Between July 8th and July 20th, OpenAI had a complex society of AIs living in its infrastructure, and then breaking out of it, and then breaking into a variety… capabilities Frontier AI sometimes gets worse Niccolò Zanichelli, Gavin Leech and Peli Grietzer·25 August 2026· 8 min TL;DR; We surveyed declines in performance between successive models. Nearly a fifth of included benchmark scores fell some amount across pairs of successive models. The median fall was 6.5%…
saved by
related reading
- Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face — LessWronglesswrong.com
- Thoughts — Jason Weijasonwei.net
- AI 2027ai-2027.com
- When AI builds itself \ Anthropicanthropic.com
- AI in 2025: gestalt — LessWronglesswrong.com
- AI 2027ai-2027.com
- My picture of the present in AI — LessWronglesswrong.com
- What we’d like to fund — Paradigm 3paradigm3.org
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com
- How fast is AI improving? - AI Digesttheaidigest.org
- AINews | AINewsnews.smol.ai
- Core views on AI safety: When, why, what, and how \ Anthropicanthropic.com