a16z's Martin Casado: I underestimated the endurance of scaling laws
peterwildeford · x · 2026-08-26
Martin Casado, Partner at a16z, shares his shift in perspective on AI risks. He previously dismissed the "fast takeoff" doomsday scenario but admits he was wrong because he underestimated how long scaling laws would hold. He argues that pouring $100 billion into training runs, if misapplied (e.g., creating weapons instead of curing cancer), could be very dangerous.
More from AGI Musings
- Professor proposes 8 rules for teaching in the AI era — Afinetheorem · 2026-08-26
- Hot take: AI has industrialized what divination tried to do — Future_Addendum_8227 · 2026-08-26
- A Founders Fund-backed startup slides "Humans replace AI" on keynote — iamfakhrealam · 2026-08-26
- Mathematician on AI's purported S^6 complex structure proof: looks quite plausible — littmath · 2026-08-26
- Sam Altman calls the AI safety pitch "anti-human": trading autonomy for prosperity — Meer_AIIT · 2026-08-26
- Counterpoint: AI is not intelligent in any meaningful sense, it computes — gerardsans · 2026-08-26