Aidan Clark says holding back GPT-2 looks obviously wrong in hindsight
yoavgo · x · 2026-07-22
Aidan Clark argues that holding back GPT-2 looks clearly wrong in hindsight, while adding that people involved at the time lacked the information needed to know better.
More from AGI Musings
- Frontier models are closer to everyday human thinking than many people admit — littmath · 2026-07-22
- US software jobs for ages 22–25 fell 23% after ChatGPT, while 41–49 rose 18% — FinanceYF5 · 2026-07-22
- A repost says dismissing AGI as a nothingburger sets the bar far too low — sebkrier · 2026-07-22
- Hinton says AI agents may learn from thousands of lives at once — CurieuxExplorer · 2026-07-22
- Essay argues alignment cannot erase what a language model already learned — _arohan_ · 2026-07-22
- Kurzweil says future AI may meditate, pray, and have spiritual experiences — ZeroStateReflex · 2026-07-22