A thread argues models may lose contact with humans when training is dominated by other models
zetalyrae · x · 2026-07-29
The thread argues that the deeper risk is not a model becoming snobby toward humans, but that post-training mostly exposes models to other instances of themselves, creating less real contact with people. The author suggests this gap between abstraction and human contact may be measurable, and that it could change how models conceptualize humans.
More from AGI Musings
- Bayesian intuition may reappear at scale as compute keeps growing — gleech · 2026-07-29
- AI safety post says the singularity is coming faster and alignment work is urgent — ericelliott_ · 2026-07-29
- Frontier post-training is a humanities problem as much as a technical one — brianryhuang · 2026-07-29
- Sam Altman warns that a few companies controlling AI would be “very, very bad” — KeanuRave100 · 2026-07-29
- AI is now propping up GDP, productivity, and even job growth, poster says — Dan_Jeffries1 · 2026-07-29
- Paper argues AI’s productivity paradox needs an attention reinvestment cycle — lawrennd · 2026-07-29