willcb: Stop reasoning about AI alignment from first principles, adapt human mechanisms
willcb · x · 2026-10-17
In a discussion thread, willcb argues the AI community often mistakenly reasons about model alignment from underspecified first principles because models are treated as "fundamentally new." Instead, he says, we should directly adapt the mechanisms that work best for human alignment. His closing quip: language models are made of humans and computers — just as societies are made of humans and computers, and are usually, if not always, aligned.
Related event: Researcher says AI alignment should borrow from proven human mechanisms(2 posts)→
More from AGI Musings
- OpenAI's mass math-solutions dump slammed for not warning mathematicians first — RileyRalmuto · 2026-10-17
- willcb: Extreme Capital Concentration and Agent Legal Personhood Are Humanity's Biggest Risks — willcb · 2026-10-17
- Jensen Huang on Lex Fridman: build the flywheel, don't just block chip exports — rohanpaul_ai · 2026-10-17
- Knowing what to ask and how to judge answers is the intelligence that matters — vasuman · 2026-10-17
- iMessage May Be the Most Sensitive Social Graph, and AI Agents Are Moving In — signulll · 2026-10-17
- The Tech-Wise Family's 10 principles for keeping tech from replacing family life — nwilliams030 · 2026-10-17