← sources

Has anyone else noticed the quality of LLMs as a reflective tool dropping in recent months?

source: reddit (reddit:ChatGPT)published 2026-09-29 13:47ingested 2026-09-30 02:37

retained text

I've been an avid user of ChatGPT in particular, and I've also been experimenting with other chatbots. I had very little experience with LLMs or even computer programming before I started using them approximately one year ago.

I started vibe coding around September/October of 2025. I did that on the pretense that the AI probably understood actual coding language better than it understood human language and could interpret the super messy thoughts I was feeding it. And I was correct in that assumption.

It is first and foremost a coding tool, but again, I'm not a computer programmer. That was just my assessment at the time.

I've since shipped several projects using GitHub, Vercel, and Supabase, as well as working with Codex, and the programming side of things has gotten objectively better over the last 12 months.

However, the communication gap has widened significantly.

I don't think the actual AI is less smart than it was. I just think there have been more regulations placed upon it. With that being said, I think something was lost along the way.

It could be regulatory, it could be for whatever reason, but I cannot communicate as well with the AI as I could before. I mean, I was using it as a self-reflection tool, and it's really not that at all anymore.

So that's kind of why I'm writing here. I just wanted to know if anybody else is experiencing the same thing, or if I'm kind of alone in that.

Yeah, let me know.

document fetched 2026-09-30 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1wtaksz/has_anyone_else_noticed_the_quality_of_llms_as_a