Fighting distillation of API models is a losing battle — KYC-gated frontier releases and US open weights argued
willcb · x · 2026-09-13
- Core argument: Fighting distillation for anything released as a public API or product is an uphill battle — accounts are cheap to make, outputs end up on the internet, and RL behavior cloning doesn't need CoT.
- Alternative: If OpenAI wants to push the frontier safely, keep frontier-class models (e.g. a hypothetical 6.5-astra) tightly monitored or KYC-only until capabilities disperse broadly within months.
- Business angle: Anthropic's lesson from mythos/fable — willingness-to-spend is capped relative to "good-enough + governance-friendly" models like Opus.
- Open weights debate: Russian hackers will get Chinese weights and chips either way; US enterprises want US open weights, facing a bad choice between closed-strong American, open-strong Chinese, and weaker-open American. One suggestion: release gpt-6-luna as gpt-oss-2 to appease the open-source community.
More from AGI Musings
- Microsoft's reported 60,000 voluntary exits spark essay asking who AI is really for — _jaydeepkarale · 2026-10-27
- Goodharting Isn't Just Fooling Others: How Metrics Confuse Yourself Too — davidmanheim · 2026-10-27
- PHP2Shell exploits confirmed in the wild, attackers may use agentic AI for post-exploitation — AravSrinivas · 2026-10-27
- LukeW: Technology Limits Are Temporary — and AI Skeptics Are Already Feeling It — LukeW · 2026-10-27
- GiveDirectly pilots an optional AI chatbot alongside cash aid in Rwanda — NathanpmYoung · 2026-10-27
- Ex-Anthropic employee essay: the org optimizes for power 'constrained by being seen as the good guys' — RichardMCNgo · 2026-10-27