Position paper says pluralistic alignment matters only if deployed models adopt it
dhadfieldmenell · x · 2026-07-24
A position paper argues that pluralistic alignment has become popular at conferences but has not translated into deployed systems.
The authors say the field should treat adoption in real-world models as the main goal, not just theoretical pluralism. They lay out a research agenda aimed at making deployed AI systems more pluralistic and better aligned with the diversity of people who actually use them.
Related event: Position Paper: Pluralistic Alignment Fails to Impact Deployed AI(4 posts)→
More from Research
- Interpretability’s local-to-global guarantees may break, a Jacobian analogy argues — forestmars · 2026-07-25
- Anthropic says Opus 5 is more token-efficient and smoother for coding tasks — alexalbert__ · 2026-07-25
- BPBench finds Chinese open-weight models dominate text compression scores — ctnzr · 2026-07-25
- AI treaty verification may need three layers: pragmatic checks, enclaves, and math — geoffreyirving · 2026-07-25
- Secure enclaves may still leak keys and weights through side channels or microscopy — geoffreyirving · 2026-07-25
- Pure cryptographic obfuscation for AI verification still costs multiple orders of magnitude — geoffreyirving · 2026-07-25