Researcher: model misalignment demos without input info are manipulative or sloppy
giffmana · x · 2026-11-16
giffmana argues that every time he's inspected the user/system prompts behind a supposed model misalignment case, the cause was obvious and usually even invited by the prompt. So demos claiming self-interested or misaligned model behavior without showing inputs are either manipulative presentation or a sign the presenter never looked at the data. He notes this stance shifted from six months ago, when "malicious compliance" seemed a better explanation than genuine self-interest.
More from AGI Musings
- Researcher calls superintelligence doom scenarios "sci-fi fanfic", argues LLMs can't self-replicate — ziv_ravid · 2026-11-16
- AI Didn't Take Designers' Jobs — It Replaced the CorelDRAW Sign Shop — feregri_no · 2026-11-16
- tszzl fires back at AI risk skeptic: monkeys fear rivals, not autonomous computers — tszzl · 2026-11-16
- CFGeek pushes back on Ben Todd: AI safety folks expect a full software-only intelligence explosion, not just a spark — CFGeek · 2026-11-16
- AI now drives nearly two-thirds of US investment growth while crowding out the rest, says Moody's Zandi — asusarla · 2026-11-16
- EgeErdil open to AI pacing only with strong ASI evidence in 2 years — adam_karvonen · 2026-11-16