This is a fun thread because 10% of replies answer the question while 90% instead answer the question “did you predict the bubble during the dot com boom”
For people who were not kids during the dotcom bubble:
Did absolutely everyone went around telling the world that it was a bubble ready to burst at any moment like everyone is doing now for AI?
Older models did this too but Opus 4.5 feels like the first that recognizes what I actually want, yet still can't bring itself to write a failing test.
Even when I explicitly ask for it, Opus 4.5 will never write a failing test. It would rather
- assert the incorrect behaviour
- re-write half the codebase
- actually write the test you want, then try to get it to pass, give up and delete everything
GPT5 router took my quick estimate Q and decided it needed thinking mode for some reason.
"Estimate how many people I could feed with a 6" diameter tiramisu cake"
I wanted a vibe check but got a fermi estimate with way too many decimal points. RLHF math-maxxing too hard?
@sailaunderscore 'Get the bag at any cost' seems born of nihilism with a global society playing endless status games. Moral virtue appears to go unrewarded while depravity looks consistently elevated in the public sphere. So they hop on.