[DennisP]

I think it depends on how you use it. Recently I spent several days using Opus Extended for help on a simple mathematical model. I came up with the modeling ideas and some hypotheses, it did the math and worked through implications. Then I carefully read through everything it wrote, made sure it hadn't misinterpreted what I'd said (which it sometimes did), and checked all the math. Occasionally it used math I didn't understand, so I opened another chat to teach me that math.

In one way it was sycophantic, frequently saying how "sharp" my ideas were, but I just reminded myself to discount that. In other ways it pushed back. If the math didn't work out to the conclusions I expected, it pointed that out. It was like working with a reasonably smart, extremely productive, but less creative coworker.

I don't think it made me lazier, because it was intense and exhausting. But I learned a lot.

[wesleywt]

There are a few people out there with AI psychosis that thought they developed a new mathematical theorem.

[DennisP]

I wasn't attempting anything like that, just working out the implications of a simple model of a real process.

[dotancohen]

Which models were you using? Local inference or cloud?

[actionfromafar]

Local inference with Claude Opus, now that would be something. :)