LessWrong Post Says US and China Need Compute Treaty to Slow AI
Summary
A LessWrong post by josefely argues the US and China can only durably slow AI through a treaty governing compute, because a unilateral pause would let China erase a roughly 10-point Epoch Capabilities Index lead. Verification would depend on satellite monitoring, on-site inspections and tamper-resistant chips that meter and license compute, the post says.
Key Points
- Josefely writes that no AI chips today are designed to comply with a bilateral US-China treaty.
- The post cites FlexHEG, a retrofittable device whose ruleset updates only after separate nations holding cryptographic keys agree.
- Josefely notes Anthropic employs 3,950 people as of March 2026, against 130,000 on the Manhattan Project.