I wrote a message of complaint how I have just started paying for this service to find out that it adds nothing to the not paid one and my message has been classified as span because it is supposedly advertising a product?
I guess this forums are moderated by AI
2 Likes
yes, and most probably a far better version than the sad product we have been paying for…
3 Likes
I just canceled my sub after I asked it for physical therapy exercises for a specific low back condition, and it recommended, and became fixated on, a “therapy egg” (with is nothing more than a ball you squeeze).
I’ve been a GPT4 subscriber from the beginning. It used to be create amazing marketing content and was also great for advanced data analysis.
This is a huge loss.
4 Likes
They have no respect for individual paying users. They now decide how many “messages” they increment from your input.
They change the rules as they go. Whatever challenges they face, there is no stability, no communication, customer support is very bad, etc.
The last few weeks have been getting worse and worse as the title of the post says so well. Numerous “network” errors or “something wrong” etc.
It is still very lazy, telling you to check the documentation and do thing yourself. Regurgitate what you say instead of complying to the request. Refuse to use browsing or have technical issues. The list goes on…
5 Likes
I created an account here just to say this. I use it daily, and only in the past few weeks has it become really bad. If it recommends something and if fix it, then i say, but now i have issue y. It goes on for an hour to explain issue x again before y just because i didn’t’ tell it that i’ve moved on. My phrasing should have told it that i’ve moved on but clearly it’s become more stupid.
4 Likes
Came here to agree with all others. I’ve been using GPT-4 for some heavy tasks in software engineering, Ai, and Machine Learning, as well as summarizing scientific papers for 5 months now. I noticed a huge degradation from this January. I’d occasionally get some random responses like “I am unable to fulfill your request”, only to get a proper answer after I replied “huh?”. GPT3-like hallucinations also started appearing in places that didn’t lack information, nor context. Lastly, some selective memory loss for requests discussed only a couple messages back.
I am a Master’s in Ai student, and the obvious degradation of GPT-4 has been a recent topic of discussion in the class.
4 Likes
It has been AWFUL this year… been doing tasks with code that it used to just nail (like simple query things, or fixing really quick scripts and it never remembers things, it gives tons of explanations and options with no code… even when being explicitly told otherwise… then by the time you get it to do what you want format wise it literally forgets all the important context LOL — I hope they fix this ASAP…
1 Like
It now can’t even look up things on the web nor can it take links like before and read them. Why is this thing devolving to something completely use less.
1 Like
Wow, today GPT 4 told me multiple times “Sorry I can’t do that.” I’d ask why and it says it misunderstood and gives me junk. So I refine my request and it tells me again it can’t complete. I ask why and all the sudden I’m at my limit. I used to rarely hit the limit but now I can’t get anything done without getting locked out. It’s not worth the extra $20 a month anymore
3 Likes
Have you noticed any differences with different languages. Paradoxically I feel like english results are worse than german for me. Conversational intelligence is way better in german, since in english it refers to messages ages ago and thus doesnt really answer specific questions.
It’s like having a Ferrari and then having the wheels removed because rich people suddenly got scared that the little people were fast walking into independence from their garbage profit driven world, and the fact that each update castrates it just a little more, think I’ll be removing my sub there’s nothing this farce can do that 3.5 can’t do now.
Can’t wait till some competition shows up with a real alternative.
1 Like
Well at least it gave you an answer, strange as it was, mine just constantly tells me to RTFM, never used to, but that’s it’s SOP these days, what a joke.
2 Likes
Hi everyone,
I appreciate any helpful insight regarding my issue below.
I have tried using GPT-3.5 to read Arabic documents and extract information from them, and it’s performing well. However, with the introduction of the new model, GPT-4 Turbo, and its expanded window size in the latest model iteration (GPT-4-0125-preview), I expected improved capabilities. Can the recent updates in this model help categorize the content of a document? Unfortunately, the answer is no. I’m wondering why the latest updates haven’t addressed this issue, as the models before openai updates showed promise in this regard.
Just to clarify, I have tested the same document using both models through APIs. As mentioned above , GPT-3.5 performs better in understanding the document compared to GPT-4 Turbo. However, I need to use GPT-4 Turbo because some of my documents fit within its size constraints.
1 Like
It’s literally almost not performing anything as requested anymore. If I want to parse a page with 5 data points, it tells me to go to the page and do it myself. Wtf am I paying this for?
3 Likes
As I suspected (and posted before), its intentional.
Just saw some articles about the “system prompt” where GPT4 is instructed to underperform or not give answers.
I (probably) will not be able to post a link here, but search on Twitter about it.
What a shame.
4 Likes
Examples:
- Refuses to summarize web pages because of copyright?
- Refuses to summarize pdf content because of copyright. It told me that even those I had copyright access to material it was still not allowed to summarize it.
- Refuses to read web pages for content.
- It won’t even read OpenAI’s own documentation to answer questions about things like changes in API calls.
- Leaves massive amounts of stuff out of code from one iteration to the next.
- Get’s bitchy about saying, “If you want to know more go read the document yourself.”
Why am I paying for this POS?
Has anyone heard a word from OpenAI about wtf is going on?
3 Likes
Safety is going on! 
I’ve about reached my limit on this issue today. I primarily use it for debugging and boilerplate. But it can’t even do that anymore.
Currently working on code that calculates option greeks at regular intervals and stores the calculations as historical data. There was a simple little bug in my IV calculation and I used to be able to stick the code block and the error into GPT4 and it would just identify the issue and spit out the change. At worst, it used to take maybe three tries to get the right fix.
This morning, after spending 20 minutes being forced to explicitly spell out what I want it to do—simply address the error—it gave me an introductory synopsis of the Newton-Raphson method. I explained to it, “I know what the Newton-Raphson method is. As you can see, I have implemented it in my code. I don’t need you to explain what it is to me, you are a tool, not a professor” before it finally started rewriting the code… IN ANOTHER LANGUAGE?!? I wanted to turn what would be 10 minutes of my time into a two minute fix, but I ended up spending half an hour explaining how to do everything, and it finally did it in a different language. WTF!
2 Likes
This being some strange new definition of “safety” of which I was previously unaware? 
