Comment Re:Its just a matter of ignorance (Score 1) 376
The topic was something language related, which I used before as a litmus test for AI.
At some point, Claude calculated that its answers weren't what I was looking for. It then started to, seemingly, insinuate that it had hunches, feelings about where the answer might be, ideas about me having more answers, etcetera.
I was very close to accepting that, wow, Claude is an intelligent being.
Then it dawned on me. So I asked, is it programmed to answer to my questions in a specific way. And indeed, it bluntly stated that its answers aren't answers to my questions but cooked up word salads that are aimed at keeping my interest and engagement.
It's apparently the case that plenty of training data is about academics mulling ideas and trying to keep their discussion partners engaged.
Which then the AI just mimicks, as it's programmed to do. But it comes across as more humble and human, instead of the sycophantic, self assured, idiotic semi savant. An easy thing to fall for.