View: Framing Pretraining as Instructive Process
Sauers_ · x · 2026-08-30
The post discusses a technical viewpoint suggesting that if post-training privileges the assistant persona to ground mind-like behavior, adopting an instructive process in pretraining from token zero could yield much more mind-like models. The quote references a paper arguing for framing pretraining as an explicit learning process for better stability and alignment.
More from AGI Musings
- When Will an AI Agent Autonomously Gain 1M Followers? — bilawalsidhu · 2026-08-30
- Humans are "the cheapest dates in the universe" for seeing minds in machines — artBoffin · 2026-08-30
- Gary Marcus warns that anthropomorphizing AI misleads our understanding — GaryMarcus · 2026-08-30
- MIT warns AI can now credibly complete most undergraduate assignments — Hubbardia · 2026-08-30
- Overcoming status quo bias: The need for visible AI breakthroughs — Afinetheorem · 2026-08-30
- Industry fears liability: Drunk driving vs AI cyberattacks — iamtrask · 2026-08-30