Governments Can't Control High-Capability Models Long-Term
i_dg23 · x · 2026-07-19
The quoted view argues that governments will struggle to exert long-term, stable control over highly capable neural networks, as model capabilities will inevitably become accessible to more people. The author's core judgment is that the real future issue isn't whether capabilities will proliferate, but whether **open-weight models** will be deemed "unacceptable" as they approach the frontier. He believes society should do two things simultaneously: build resilience against the risks of out-of-control, highly capable models, and maximize the benefits offered by both open and closed models, treating "openness" as a long-term feature of the world.
More from AGI Musings
- Math models were already good at brute-force counterexample hunting, says AI researcher — dyamins · 2026-07-21
- What if the brain could add new words every day and keep expanding thought? — ZeroStateReflex · 2026-07-21
- MIRI’s Plan S is said to have a 66% win probability, far above Plan A’s 18% — eli_lifland · 2026-07-21
- Open-source robotics should not be slowed to protect two labs’ evals — RileyRalmuto · 2026-07-21
- By 2030, an AI-made song by a dead musician could top the charts — VraserX · 2026-07-21
- Garry Tan says AI is making everyone equal as superintelligence reshapes product building — garrytan · 2026-07-21