Thought experiment: Pause pretraining to focus on controlling inner drives
louisvarge · x · 2026-08-28
The author proposes a thought experiment: imagine the world coordinates to halt significant improvements in pretraining or RL, focusing instead on controlling the inner drives arising from training.
Key questions raised:
- How long until we have a settled science of inner drives?
- What specific questions must be answered before resuming scaling?
Related event: Radical Proposal: Pause Pretraining to Solve AI Control First(5 posts)→
More from AGI Musings
- Viewpoint: Biology is nanomachinery, tech has hard limits — sebkrier · 2026-08-29
- Analyzing all 1,329 OpenAI and Anthropic job listings: Personal AGI and robotics are next — imjustnewatai · 2026-08-29
- Princeton Prof estimates 80% of scientific literature is 'flat-out wrong', sees AI opportunity — ashajadeja · 2026-08-29
- The Minimum You Need to Do Before AI Changes Your Life — Least-Measurement350 · 2026-08-29
- Widespread Automation Causes Price Collapse; High-Cost Lawyers Replaced by Compute Costs — zacharynado · 2026-08-29
- Adam Grant: AI Struggles with Creative Leaps Due to Historical Training Data — blaizedsouza · 2026-08-29