OpenAI leaks mysterious model codename, Altman slams doomsday marketing

新智元 · wechat · 2026-08-24

OpenAI employee accidentally leaked model codename 'gpt-nathree' on GitHub, likely a checkpoint of GPT-6 Astra; Anthropic's new models 'claude-marshmallow-eap' and 'claude-melon-eap' exposed. Meanwhile, Altman admitted misjudging AI disruption timeline, criticized doomsday marketing, and emphasized human-centric values.

Original post →

More from Companies & People

Companies & People channel →