Chat GPT 4 is getting worse on content generating from July 2023

I started to use ChatGPT4 since it started to charge 20$ per month, by using it on daily basis, I found it getting much worse in recent, especially from July. It seems not able to read the my draft content in detail and often create content not including the key word, key points, that I like it to be included and even drafted in my reference content. There are few words, ChatGPT really like to use, but I do not like, such as " In light of" , " rest Assured " , even I ask it do not use these words, will not stop it keep using it regardless. Not mention the letter it wrote often seems too long and too much extra words which is not needed. Anyone else has the same problem?

Yes, we all are in the same boat.

Many months ago when ChatGPT was originally like that many discovered that adding be concise to the prompt often helped/worked.

Yep, i call this the alzheimer lazy edition.

Its like

-Write me a joke

I’m sorry, the last prompt got cutoff but i can still assist you with a joke:
There was a man (…write the rest of the joke here)
I’m glad i was abl

But hey, now we have 50 prompts, even tho we have to prompt 5 more times than before, yay!

Some people will dismiss our claims as subjective or anecdotal, but there is absolutley no doubt that ChatGPT-4’s outputs have been changing, in my mind for the worse, lately. There’s enough evidence that can’t be dismissed in good faith. Is it some cost cutting measure or something else? No idea, but it’s an undeniable fact something has changed.

It is definitely worse. I was gpt4 pretty consistently for work. However, in recent months it has become so bad to the point where it is unusable for what I needed it to do. I also noticed that the custom prompts that I have been using for a while are significantly less effective. Overall I am super disappointed and have canceled my subscription because it is no longer useful to me.

I can for sure say it has been nerfed. A couple reason why this might happen. The addition of “Custom Instructions” are included in the request (seems after about 40 responses, it forgets this information.) At 60 responses it is just like fighting a toddler for a sucker. Another problem might be that they are adding a bunch of stuff to the request (well server side) for linking data. Here are some notes I have been using to ensure a somewhat decent responses. (These things were no issues on release, so more then likely they are being pressured to reduce the performance of things like coding due to market influence. )

Notes: Day Dreaming Started At : 41 Responses.
At Response 46, I resent the base Custom Instructions (went back to single steps.)
Became completely dumb about response 60.