Nathan Lambert Debates Doom: AI Improving Training Software Doesn't Imply 'It Will Kill Us in N Years'
MoonL88537 · x · 2026-09-09
Nathan Lambert argues that while we have evidence AI agents are good at improving software used to train AI and LLMs can be superhuman in many domains, these do not combine into 'AI will be so powerful it will kill us in N years' — that forecasts inevitable saturation and unknown breakthroughs. A replier counters that the distinction between 'impossible' and 'negligibly unlikely' matters, and in the limit we are already there. A serious technical debate on self-improving AI risk.
More from AGI Musings
- Paying mathematicians $1,000 to review AI proofs — but how much work is it? — ctjlewis · 2026-09-10
- Nic Carter challenges Anthropic's 'Aligned Machine God' belief: how would getting there first even work? — examachine · 2026-09-10
- Viral jab at frontier labs: 'too dangerous for anyone else to own, but we take all credit cards' — prateekj · 2026-09-10
- Ted Chiang, Ken Liu and writers tackle writing and reading in the age of AI — begusgasper · 2026-09-10
- Sentdex: centralized human power, not AI, is the real x-risk — Sentdex · 2026-09-10
- Apple keynote take: ambient AI matters more than the foldable iPhone Duo — VraserX · 2026-09-10