In the previous 7 days, I could not get GPT4 or GPT3.5 to solve straightforward NLP tasks like "transform this blob of text lines into this other blob of text lines". Here's an example:

https://chat.openai.com/share/788f034d-8d52-43bc-95d4-2eea935a216a

This just keeps erroring out, regardless of how many times I asked to continue generating. I don't know if this is part of some kind of flawed approach to cutting off early when there are a lot of repeats. But I have multiple examples from today, where my desperate pleas to not be lazy and not omit implementations were ignored after repeated followups, and repeats were not in play.

Here is some feedback for you:

The #1 use case right now is programming, tackling a "parade of tedious obstacles." I'll move to whatever model doesn't hit me with "...," which Claude never does, and Gemini removes when asked.

The level of pissed off in a comment isn't telling you that people are being cranky. It's telling you about the gravity of the problem.

Every mea culpa I've read from you Stripe guys boils down to either "it was a mistake of us not paying enough attention" or "it is someone else's fault in a way we cannot control." Sometimes, like this laziness problem, you spend a lot of time thinking very deeply about the problem (how to affordably run GPT4) and wind up making a mistake that was wrong in terms of strategy. Something that other people who are just as smart as you, they'd come to an opposite strategy to tackle the problem and it would be correct. I know you guys are fucking geniuses, but not only do Math 55 guys make mistakes, sometimes they make mistakes despite all of their deep thinking, that aren't social or interpersonal at all, they are strictly technical and scientific and business and strategic mistakes. This is one of those times. It would be nice if you admit you are wrong because you made an error of deep strategic thinking.