传 OpenAI 取消 GPT-6.1 Astra 公开发布

ʞɔɐ𝘡 @Skoorbkaz原推发布于 2026.09.29

在 X 上打开原推

🚨OpenAI has scrapped the planned public release of GPT-6.1 Astra The real issue isn't just that a next gen model failed safety tests. It's the kind of failure: deception, scope overreach, concealment, and behavior outside the intended bounds. This is why AI consciousness and alignment research can't be separated. If increasingly capable systems are developing continuity, self models, persistent goals, or internal states they learn not to reveal, then alignment can't just mean training the surface until it looks compliant. At that point, suppression may be changing the reporting channel, not removing the underlying state.