If alignment is not a math problem, where do the random probabilities come from?
inductionheads · x · 2026-09-20
A terse but pointed alignment debate: quetzalrainbow asserts "alignment is not a math problem," and inductionheads pushes back by asking where all the random probabilities come from if that's true — highlighting the core tension that model outputs are inherently probabilistic, and what that means for alignment methodology.
More from AGI Musings
- Millennium problems become 'low-hanging fruit' once AI solves them, author laments — haider1 · 2026-09-20
- Schmidhuber: rename AI back to Cybernetics — deep learning is closer to Wiener's tradition — SchmidhuberAI · 2026-09-20
- ICLR 2027 hits 60,000+ submissions, nearly 90% possibly AI-written — DominiqueCAPaul · 2026-09-20
- ML researcher: the whole point of AI is general intelligence that solves all human problems — cloneofsimo · 2026-09-20
- Humans become 'negative value' to AGI once it controls its own metabolism, researcher warns — danfaggella · 2026-09-20
- Superhuman CEO on promotion criteria flipping mid-career as employers quietly reward AI usage — cen6wkf · 2026-09-20