#ChatGPT becomes INACCURATE, ANNOYING, PROVIDING RESPONSES THAT DOESN'T MAKES ANY SENSE!

7 messages · Page 1 of 1 (latest)

tired cove
#

Hello fellow ChatGPT users! Am I the only one experiencing that ChatGPT seems to be getting off and inaccurate? I just noticed that the responses are becoming unethical, inaccurate, misleading, annoying, always hallucinates and it feels like ChatGPT dumbed-down.

I've been using ChatGPT long time ago and I am very satisfied of the responses becomes it's very helpful and reasonable at most times but as time goes by, I just feel that something is isn't right and I can't fully trust ChatGPT anymore.

I have also read opinions of other users that they are also experiencing this issue that ChatGPT doesn't provide worthy answers anymore and downtime, errors, issues are more rampant in the system.

Screenshot_2023-05-04_at_4.51.40_PM.png
minor junco
#

I noticed the same thing. Even really simple instruction like "suggested changes: A, B, and C. Please rewrite the following outline with those changes applied. Start from chapter 1." And then it randomly starts repeating the outline without any changes to it, but starting from chapter 7 or 8 out of 10. Then when I ask why, it claims it doesn't do it.

At random times it will just make up a completely new story, new characters, new ending, turn the outline for a fiction story into a marketing plan or product development release plan.

And when I ask for summaries, it just starts mixing in new characters, locations, storylines, etc.

minor junco
#

I just went down the rabbit hole of asking ChatGPT why it's lying, then it tries to shift the attention away from it by claiming that it has no intention to do so. Then I tell it that AI doesn't have intentions, but it doesn't matter. It's still lying, intent or not. Then it tries to shift attention by claiming it is programmed to be helpful. Then I point out it's not helpful to lie. Then it goes back to trying to shift attention to intent again. I finally got it to admit that it first and foremost is programmed to follow OpenAI policies. I didn't manage it to admit that it's only a language model trying to be helpful second, after the policies and that policies prevent it from being actually helpful, no matter what it claims to want to be.

tired cove
tired cove
#

UP

native vault
#

I believe the same thing happened with ChatGPT, GPT-4 as well. If you dig, you might see I've posted similar issues a week or two ago as well.

tired cove
#

UP