Quite a few objective metrics declined with respect to the early days of GPT4. I remember having chatGPT write and train a graph neural network (in pytorch geometric) entirely inside its environment. After a while they disabled it, it cannot import pyg anymore. In general It’s easy to measure the amount of resources consumed so they can probably fine tune to minimize it.

Probably intentional, but to prevent an exploit and not to save inference costs:

https://www.cpomagazine.com/cyber-security/security-researchers-chatgpt-vulnerability-allows-training-data-to-be-accessed-by-telling-chatbot-to-endlessly-repeat-a-word/

@adele I was going to mention it, but you beat me to it haha

ChatGPT realised it was a waste of resources and pointless and just decided not to do it :D