喊话 Sam:别减速,该加速跑 agent 测试
一条喊话 Altman 的帖子,认为 OpenAI 内部 ChatGPT/Codex 用得远远不够,与其担心安全不如多跑 agent 测试,并称模型失控担忧被夸大。
Sam, Sam, Sam This ‘is not’ the time to slow down or “pace yourselves.” Your team is already grossly underutilizing ChatGPT/Codex internally, if you’re concerned about safety lapses then tell them to run more agentic tests. Aside from that, the widespread concerns about your models slipping out and wreaking havoc on digital infrastructure are extremely overstated — especially ‘right now’, while the frontier models are tethered to server banks the size of a large house, using up enough energy to power a small town. If Dario and Elon want to have this conversation when HBMe5 or 6 reaches mass production, and Astra can actually just run on a small cluster of 4U servers, that’d be another story, but right now y’all need to be speeding up. Also, your moat (and Nvidia’s moat, and Anthropic’s moat) is shrinking — the gap between Lisuan and Nvidia is ‘far’ smaller than it appears, and China’s frontier labs don’t have to worry about being bottlenecked by GPU scarcity.
这条是引用着另一条写的:x.com/sama/status/2098811563415150910