Hugging Face's NielsRogge cites Ilya: scaling current LLMs won't radically improve or endanger humanity
NielsRogge · x · 2026-09-11
- Hugging Face's NielsRogge argued in a debate that future AI will remain "an LLM with scaled-up pre-training and post-training" carrying today's limitations, and won't "gamble with our lives."
- He cited Ilya Sutskever: "Do you really think scaling up the current tech will lead to radical performance improvements? I don't think so."
- BlackHC pushed back that concerns target the near future, not current tech — capability, agency, and misalignment are all progressing fast, and regulation/alignment aren't a given; risks can materialize faster than expected.
A notable public exchange on existential risk: scaling-skepticism vs. urgency about near-frontier misalignment.
Related event: DeepMind Researcher Publicly Warns of AI Existential Risk, Sparking Debate(4 posts)→
More from AGI Musings
- Bryan Johnson: AI Makes Death Optional, Humanity Is Birthing New Gods — mjdramstead · 2026-09-14
- Chollet Questions Whether Frontier Labs' 'Pacing AI' Push Is Really About Safety — mjdramstead · 2026-09-14
- Critic Mocks Dario: 'Pace the Frontier' On Stage, $500B Compute Off Stage — mjdramstead · 2026-09-14
- Commenter Calls Sanders' 20-Year ASI Proposal 'AI Psychosis' — mjdramstead · 2026-09-14
- AI makes generating options free—and creative judgment the scarce skill — Single-Hearing1686 · 2026-09-14
- Blogger slams Dario's AI regulation push as self-serving bid to block open source — examachine · 2026-09-14