ChatGPT 4 reduction in capability


Just wondering if there is some information on why ChatGPT 4 seems to have lost some of the capability it had when it was initially released? Is this due to performance issues and outages and trying to fix them?

Really seems easier to get useful output out of version 3.5 at the moment with a little extra prompting, but recently version 4 is lacking the magic it seemed to have when I started initially using it and the output is a lot less useful.


1 Like

This situation seems to have worsened since my original post, anyone have any info?

I have only seen it improve. Do you have examples of what you’re encountering?

Increased speed of response, but more frequent hallucination the longer your prompt is, needing to be put back on task more frequently.

Basic example, if you feed it info containing a known problem seemed to spot it 100% of the time, now maybe 50% of the time, so seems to be ‘skimming’ input tokens rather than using them all etc

Honestly wish there was more info being released about model changes and performance tuning as things change quite noticably quite often.

This isn’t an example, this is barely an anecdote.