Very inconsistent GPTs that do not follow instructions

ChatGPT
1

Hello!
I’ve been building custom GPTs for 3-4 months to assist me and my colleagues in their work.
But since 1-2 months, I have the impression that the GPTs are much less consistents :

  • They respect the requested instructions much less
  • They don’t read the knowledge base files
  • They are inconsistent on the format of the answers

I feel like they sometimes behave almost like a ‘Basic’ ChatGPT conversation.

Have you noticed the same thing? Or is it just me becoming more critical and having too high expectations of GPTs?

Thanks in advance for your answers!
Max

9 Likes
2

Yes, same here. I find also the quality of the outputs in general below average.

off-topic: What makes me crazy is the use of em-dashes :clown_face:

1 Like
3

Some people have noticed a drastic decline in the capabilities of assistants created via graphical/web interfaces. In my specific case, since the afternoon of May 28, all of my assistants have lost the ability to produce anything even minimally usable. It seems they can no longer access and follow the documents in their knowledge bases; they invent information that is completely unrelated to the subject/task.

I made a post here on the forum, and I found others on the same topic/difficulty, but so far no one has received any official information from OpenAI. Not even the problem has been acknowledged by the company, which is very frustrating.

I’m really beginning to look into other possibilities, such as Claude.

3 Likes
4

And I like em dashes—really! I’m a product marketing copy editor, and it is a good method of adding a “super-pause” to break up a sentence.
Some people like chocolate ice cream, some like vanilla…

5

Yet you are here to tell me you like em dashes. :man_facepalming:

1 Like
6

Hi! I was complaining a lot here and on other openAI networks about the same problem. I’ve been experiencing this a lot in GPT-4o, as well as other issues. I felt this a lot after the rollback. I was showing a lot of my frustration and it’s still not better.

2 Likes
7

How long have you felt this way?
It’s so frustrating to have placed so much hope on these custom GPTs and invested so much time configuring them, only for them to become almost unusable…

2 Likes
8

I felt this exactly one day after the rollback, on April 29th. On the 28th they put GPT-4o back to its old version. There were a lot of complaints about flattery, my GPT-4o had never flattered me so I was really surprised about this change. Since then, GPT-4o has never been the same to me. It is unrecognizable to me.

9

This topic was automatically closed after 20 hours. New replies are no longer allowed.