GPT-4 getting lazy again, prompted it for detailed respones, getting short responses now

It’s clear that they keep trying to prompt it for concision. I assume that’s great for people who hate to read, but for me, it breaks the most useful tool I’ve ever found. I want long, detailed, creative, verbose, informative answers that completely address my questions as well as related topics I might not have thought to ask about.

I had it prompted to just my liking, and it worked well for a couple of months. Every few months or so, though, it seems like OpenAI prompts my prompt work out from under me. The difference is stark. As of yesterday, I just can’t coerce a good answer out of this thing anymore, it offers nothing but as short of an answer as possible now.

1 Like

It may be speculative to say, but sometimes the initial responses resemble those of GPT-3.5.

It is possible that GPT-3.5 is being used for initial responses as part of conserving computational resources.

Regenerating seems to yield responses from GPT-4.


Thanks for pointing that out. I’ll start giving it a try. Now that you mention it, these new responses do seem kind of 3.5ish, don’t they?

Edit: is it a resource issue? I’m already a paying customer, I want to pay for a quality product. If the cheeseburger costs a bit more I can live with that.


It may indeed be a resource issue. They might be strategically using GPT-3.5 and GPT-4 to deliver ChatGPT services to a larger audience, optimizing resource allocation.
This is just speculation, though, so it’s unclear if this is the case now or will continue to be.

Even if the initial responses are powered by GPT-3.5, they might be sufficient in some cases. Therefore, regenerating the response when the initial one isn’t adequate could be a viable approach.

If you’re looking for higher quality, using the API might be one option to consider.

Additionally, the Teams plan, which has a more relaxed limit on the number of interactions compared to the Plus plan, might also be worth considering.

1 Like

Interesting, lately I have been experiencing paid ChatGPT 4 consistently printing out entirety of code for even single line minor changes. I have actually been prompting it to not print out entire code, to be concise, and just show modifications because response takes so long as it prints out entire long code blocks.

I use the Teams plan, and I like it. You’ve still got a great observation in the initial response.

@curious_learner There are a bunch of different ways to address this, which you can see discussed in that thread; and, @dignity_for_all is correct, the underlying engine for the ChatGPT interface is up-for-grabs:

  • If you always want to ensure the model you’re talking to, look at Assistants.
  • If you’re just using ChatGPT4 Plus, make sure your Custom Instructions are up to date with your desired format.
  • Your needs sound more robust than the average person’s, so making a CustomGPT for yourself with your exact specifications is probably the rout for you.

I don’t think I’ve started a conversation with the standard ChatGPT interface since CustomGPTs were unveiled. I had a conversation go on for something like 130 double-spaced pages the other day. We go on deep-dives for sure.

It’s because of that that the Assistants API might be more expensive, or at least equal to, a teams subscription… which I admit doesn’t seem as beset by difficulty. My biggest challenge is getting them to change the name on my Builder Profile. :man_shrugging:

1 Like

Thanks @thinktank , I’ll look into using Assistants and go through that thread. I really like the verbose version of ChatGPT, and if it means I don’t have to deal with this every 2-3 months, it’s worth the extra effort and expense.

1 Like

If you want the verbose ChatGPT then you want to use CustomGPTs via Plus or Teams.

“Assistants” is the developer solution, and while you can have it be verbose, price-wise, you’re better option is the ChatGPT UI using Plus or Teams. With Assistants, there’s actually good reason to make your prompts as efficient as possible.

But, alas friend, as with all of this brand-spanking-new future technology, we’re all pretty much committed to updating it at least every three months. Word on the street is that OpenAI is releasing stuff today (monday, 5/13) and that’s going to change everything again. It’s the name of the game.

I’m already a Plus subscriber. this post’s complaint is as a plus user.