AI safety researchers debate whether evidence-based iteration implicitly relies on prediction checking
davidmanheim · x · 2026-10-06
davidmanheim challenges shlevy and Kelsey Tuoc: if you iterate on ideas grounded in evidence without relying on where evidence contradicts your expectations, isn't that just an informal version of making and checking predictions?
In the quoted tweet, shlevy argues that while iterative idea development certainly needs to be grounded in evidence, the means of doing so is rarely explicit prediction-making and checking — a niche but substantive epistemics debate within the AI safety community.
More from AGI Musings
- Gary Marcus alarmed: Musk sees humanity as a 'biological bootloader' for superintelligence — GaryMarcus · 2026-10-06
- Carl Feynman lists hypothetical 'cheap ultradestructive' AI weapons, from engineered plagues to U-235 extraction — zetalyrae · 2026-10-06
- The Verge: Brains aren't computers, and AI is junk food for the mind — benjaminjriley · 2026-10-06
- Critic: Big Tech is systematically dismantling institutions for human learning via AI — benjaminjriley · 2026-10-06
- Nobel's 3-person rule outdated: scientists rethink prizes for AI-assisted research — JMateosGarcia · 2026-10-06
- X users clash over whether lobbying the Pope on AI consciousness is 'dangerous and insane' — ctjlewis · 2026-10-06