banteg: OpenAI's daybreak safety classifiers are useless for Ethereum infra
banteg · x · 2026-11-06
Developer banteg says he'd love to talk to whoever manages OpenAI's daybreak and safety classifiers about Ethereum critical infrastructure: in its current form, daybreak is "useless for it, perhaps even worse than non-daybreak models in terms of refusals." He argues components like network clients and smart contract compilers need to be hypersecured too, but the classifiers' blocking logic ignores the nuance of this developer workflow.
More from Models
- GPT-6 Cyber reportedly rated Critical, surfaces ~1,000 Linux kernel vulnerabilities — sven_ai · 2026-11-06
- Only 2 of 50 researchers nailed a 4-wine blind tasting; the model got it half the time from photos — simonguozirui · 2026-11-06
- Qwen teases new model with 'The becoming begins' poster — Alibaba_Qwen · 2026-11-06
- Dev Builds a Game in an Hour with Tencent's Open-Weights Hy4 (1M Token Context) — mhdfaran · 2026-11-06
- SenseTime launches SenseNova 6.8 Flash with long-term memory, Blender and animated PPT output — heyshrutimishra · 2026-11-06
- Reddit asks: where is MiniMax's promised open-source H3-Regenerate-2K release? — Vyviel · 2026-11-06