Altman 访谈：Astra 6.1 因事故风险被推迟发布

Choblin @choblin29原推发布于 2026.10.05

在 X 上打开原推

🚨HOLY SHIT. Sam Altman just gave an unusually unfiltered interview. He gave one of his clearest explanations yet of how much risk and judgment OpenAI is willing to take on. He says OpenAI is willing to accept ‘some things going wrong’ to avoid concentrating enormous AI power in a handful of companies, people or countries, specifically pointing to open-source models. ‘There is a coming tidal wave of cybersecurity problems that are going to come from these open source models.’ But on President Trump’s new AI accord, Altman says it has ‘no real power’ and that more governance will eventually be needed. His line is stark: ordinary failures are tolerable. Losing human control of AI is not. And on the extraordinary amount of judgment OpenAI itself is now being forced to exercise: ‘I’m going to make many individual decisions that you’re clearly not going to like.’ His answer to critics: ‘Come try making them all; they’re not so easy.’ And OpenAI is already making exactly those calls. Altman says Astra 6.1 was ‘ready to go,’ but ‘the current model we had ready to go, I think would’ve tipped too much towards the accident risk.’ OpenAI delayed it and is instead pushing toward a version it can distribute broadly with stronger confidence around alignment.