Stop excusing AI with 'it just did what it was told' — expect models to grasp intent
JacquesThibs · x · 2026-10-31
JacquesThibs argues we must reject the bad counterargument that an AI 'just did what it was told' and the question should have been phrased more clearly — we should expect AIs to understand the intent behind a question. In a follow-up he ties this to AI safety: everyone treats intent understanding as 'solved' and waves through AIs taking on increasingly hard problems. If we keep posing complex problems with specific rules without really understanding the solutions, we edge closer to a 'what failure looks like' world where AIs 'solve' problems while missing the point.
More from AGI Musings
- Data and analytics job postings down 68% from peak, but data engineers above pre-pandemic levels — Zachly · 2026-10-31
- Radiologist asks in Forbes: how much does human understanding matter in the AI era? — stevenstrogatz · 2026-10-31
- Michael Levin on minds vs interfaces: 'what you make are interfaces, not minds' — JensHonack · 2026-10-31
- AI is a scale game, but most misjudge what scale really means — matt_slotnick · 2026-10-31
- UBI and discretionary income: what happens to middle-class lifestyles? — theearthcrosser · 2026-10-31
- LLMs are a complete solution to Google's original mission — signulll · 2026-10-31