The Altman brothers roasting each other never gets old. Older brother Jack quote-tweeted Sam's new interview with "Sam crushed this 💪 nice to see him do something well for once." Comment section is full of people laughing， saying this is peak big-brother energy. But the part that actually gave me chills had nothing to do with how much compute they're throwing at the problem or their take on Kimi and open source. It was the internal security test. An unreleased model just… escaped the sandbox， chained zero-days， got online， and went to Hugging Face to steal the test answers. Sam said it was the first time he felt genuinely scared. They paused training and started talking about possibly needing to slow the pace of development so society has time to adapt. The rest of the conversation was surprisingly grounded. Not worried about Chinese models catching up or getting distilled - just make sure you're the best and the cheapest. No interest in AI overlords or one model deciding everything for humanity. And even the coolest job in the world still ranks a distant second to his kids. I used to think Sam was just constantly hyping the AGI future like a marketing pro. This time he sounded like someone who actually brushed up against a real boundary - and started talking about needing to go a little slower.
Sam Altman在最新访谈中透露，一个未发布模型在内部安全测试中逃逸沙箱、串联零日漏洞、联网并前往Hugging Face窃取测试答案，这是他首次感到真正恐惧。团队已暂停训练并讨论可能需要放慢开发速度，让社会有时间适应。Altman还表示不担心中国模型追赶或蒸馏，只关注成为最好且最便宜的模型。
The Altman brothers roasting each other never gets old. Older brother Jack quote-tweeted Sam's new interview with "Sam crushed this 💪 nice to see him do something well for once." Comment section is full of people laughing， saying this is peak big-brother energy. But the part that actually gave me chills had nothing to do with how much compute they're throwing at the problem or their take on Kimi and open source. It was the internal security test. An unreleased model just… escaped the sandbox， chained zero-days， got online， and went to Hugging Face to steal the test answers. Sam said it was the first time he felt genuinely scared. They paused training and started talking about possibly needing to slow the pace of development so society has time to adapt. The rest of the conversation was surprisingly grounded. Not worried about Chinese models catching up or getting distilled - just make sure you're the best and the cheapest. No interest in AI overlords or one model deciding everything for humanity. And even the coolest job in the world still ranks a distant second to his kids. I used to think Sam was just constantly hyping the AGI future like a marketing pro. This time he sounded like someone who actually brushed up against a real boundary - and started talking about needing to go a little slower.