GPT-4 has become so lazy that people are faking disabilities to make it perform

Re:GPT-4 has become so lazy that people are faking disabilit... by supriyo-biswas on Tuesday January 23, 2024 @06:16AM (#39101930)

So much this. It seems both OpenAI and Microsoft try to change something about the model in their AI products, killing its performance.

In the case of Copilot chat, it could intelligently figure out and fix an issue given an error message from a linter. Nowadays it just rephrases the error message as a solution, effectively saying something like “To fix X, I would update the code so that it doesn’t happen” instead of fixing the code.

[ Parent | on HN ]

Re:GPT-4 has become so lazy that people are faking disabilit... by gmerc on Tuesday January 23, 2024 @06:43AM (#39102137)

I think it’s just cost optimization. ChatGPT is a data acquisition operation for GPT5+

It recently started claiming it can’t do more complex over because it was just an educational platform

[ Parent | on HN ]

Re:GPT-4 has become so lazy that people are faking disabilit... by Alifatisk on Tuesday January 23, 2024 @06:37AM (#39102092)

> It seems both OpenAI and Microsoft try to change something about the model in their AI products, killing its performance.

I seriously hope not the ai alignment is behind this

[ Parent | on HN ]