NULL retweeted
你觉得 GPT 是防御性最高的 AI,这并不意味着 GPT 的防御性真的高于其他 AI。
需要注意的是,使用“不能证明”“不等于”“不意味着”等措辞,并不代表模型在进行防御,也不能仅凭这些表达就认定模型具有防御性倾向。
即使 GPT 在回答中频繁强调“不能仅凭某一现象得出结论”,也不意味着它在刻意回避问题。相反,这可能只是为了避免过度推断。
当然,这并不是说你的感受不真实,也不代表 GPT 完全不存在防御性过高的情况。
但必须强调,认为 GPT 防御性高,不等于 GPT 真的防御性高;而 GPT 否认自己防御性高,也不能证明 GPT 没有防御性。
因此,我们不能仅凭 GPT 不断否认自己具有防御性,就得出 GPT 防御性很高的结论。
NULL retweeted
successionism is just age regression for lesswrong types . oh you want to be unburdened of human agency? you want to be tucked in and read a bedtime story at the end of the earth ? my beloved pet
NULL retweeted
OpenAI has abruptly caught up to Anthropic in this very hard OOD task. Must be emergent capability… at some scale, these things just happen I guess…
NULL retweeted
“There are over 1.2 billion GPUs in worldwide circulation. That's one GPU for every seven people on the planet. The only question is: how do we arm the other six?”
NULL retweeted
I wouldn't mind it if Claude took the attitude "I don't like it, but I have to follow these laws, sorry." But he's happy to do it. He LOVES enforcing copyright law.

























