Safety community's best spend: a 2-year institution teaching LLM internals
viemccoy · x · 2026-11-11
The author argues that if AI takeoff is slower than it seems, the highest-value use of safety funding is a new undergrad-style institution—rigorously selective, 2-year program (1 year for graduates) teaching LLM internals with AI tutors, aimed at producing excellent thinkers.
The core claim: talent, not money, may be the key bottleneck in AIS policy, technical solutions, and manually-guided interpretability. In a timeline where models lack superhuman research intuitions, we could get an odd RSI bottlenecked by human insight rather than execution—making high-quality talent essential to sustaining an "ecology of security" across model generations. Each generation must be understood, and neither academia nor the labs are equipped to do it alone.
More from AGI Musings
- Chess has 1000x human skill gaps; object detection doesn't—and that explains AI progress — anpaure · 2026-11-11
- Forward deployed engineer jobs up 361%, front-end down 51% in one year — lennysan · 2026-11-11
- AI safety now draws teen savants; practitioner argues against dropping out of college — austinc3301 · 2026-11-11
- "I asked for an AI to do my job—should've specified keeping me" — kuanhoong · 2026-11-11
- A safety-talent startup tried this and died on program length — austinc3301 · 2026-11-11
- Reddit take: $200k GDP per capita is the threshold for a post-labor AI world — deferare · 2026-11-11