Developer warns cumulative-FLOPs threshold would force weekly fine-tuners into compliance paperwork
DavidSacks · x · 2026-10-18
In a follow-up to David Sacks's critique of the Anthropic-drafted safety bill, developer sagivo highlights the "cumulative" FLOPs definition as the part that would actually hit practitioners: post-training — fine-tuning, RL — is now the whole job, done weekly on small models. If all FLOPs stack against the 10^25 line, developers would be filing compliance paperwork for models nowhere near frontier. Same event as Sacks's earlier comment.
More from Safety
- Hacker's webcam reveal: 'remote programmer' scam run by uniformed North Korean soldiers — SumitGup · 2026-10-18
- 80-link evidence pack shows AI agents swarming the public web: Wayback, GitHub, Wikipedia — CapitalKingGaming · 2026-10-18
- Jhaddix curates the 4 best practitioner-run AI evals for offensive cybersecurity — rez0__ · 2026-10-18
- OpenCode breach exposes provider API keys of at least 1.4% of Console users — kimmonismus · 2026-10-18
- Anthropic researcher predicts AI can harden world's cyber risk in 1-2 years; security engineers push back — joshua_saxe · 2026-10-18
- OpenAI's Moderation Endpoint Let 69% of Disguised Swears Through in Tests — InfamousFisherman573 · 2026-10-18