在纽约市议会 10 月 5 日的听证会上，只有 Anthropic 向议员确认其模型会编写代码用于训练更聪明的后继模型。OpenAI 和 Google 未作是或否的回答，而是质疑议员对递归自我改进的描述是否符合其技术定义；Meta 的代表则无法说明其模型是否会自行生成代码。信息来自 Forbes Breaking News 的 YouTube 频道。
Looks like, only Anthropic confirmed to NYC lawmakers that its models write code for smarter successors.
At the Council’s October 5 hearing, a New York City Council member was questioning them about a loop where a model’s own code runs in the next training round and produces a smarter model.
OpenAI and Google avoided a yes or no by questioning whether the council member’s description matched their own technical definition of recursive self-improvement
Meta’s representative couldn’t say whether its model self-generates code.
---
From 'Forbes Breaking News' YT channel (link in comment)
来源：Rohan Paul · x.com