How do you differentiate AGI from ASI?
热度趋势
趋势数据积累中
百分比基于当前可用热度信号，而非评论数或独立用户人数。
关于通用人工智能（AGI）和超级人工智能（ASI）之间的区别，人们一直在讨论，有人认为随着人工智能能力的进步，AGI的定义会不断变化。例如，图灵测试在没有引起太多关注的情况下就被超越了，但它并不是一个很好的基准。有人预测，在AGI出现之前，可能会先出现领域特定的ASI，即模型在编码和数学等领域达到“超人”水平。…
It seems like the goalposts will just constantly nudge forward any time an achievement closes in on AGI until we end up at ASI. The Turing test flew by without much fanfare for example, but that wasn’t exactly a great benchmark.
The current models have very spikey domain excellence, and it’s probably going to continue in the path for being great at tasks with verifiable rewards unless/until we get another breakthrough.
(This already feels true but seems worth talking about) I think we will have Domain specific ASI before AGI, to the point of models being “superhuman” at coding and math. So, it seems like AGI will constantly be goalpost nudged and be achieved shortly before full blown ASI, but where exactly are the lines?