It’s already the end of July. I use a paid subscription to gpt 4o. And as soon as the project is already approaching the end at 90%, it starts breaking everything as if it was specially programmed to spend more time with it. Conspiracy theory? Yes. Is this a problem regardless of this fact? YES!
But that’s not even the most important thing. After all, it’s clear that it needs to be checked for errors and blah blah blah, but it hangs up dead for different periods of time and the page doesn’t even let you search for the old history you need while it’s down. Yesterday it went down for half an hour, I got tired of waiting and went to bed, I woke up today, and after 1 answer it hung up again and hangs for half an hour. This is despite the fact that with a paid subscription, the 3-hour range for the number of answers is not always enough for me. Damn it, it’s annoying.
I have noticed an improvement this morning, maybe something changed? I was able to recover the work I was doing, and have settled back in to small incremental changes one-by-one to my app. There are still some buggy responses that I have to nudge it again to correct, but overall there is more coherence, context and knowledge to the responses. Keeping my fingers crossed that it continues. I did also ask for the pre-print to be reset but it didn’t look like it had a problem, so I’m assuming it is just the model settings that are improved. Or my prompts. TBD.
I never could have imagined myself having complaints about a chatbot so early, but GPT-4o was a real downgrade, forcing me to keep using GPT-4 (Legacy) and start migrating my projects to other AIs…
I’m still very thankful for such amazing software.
I had a good time the last 2 weeks but today something just snapped and 4o has become useless. its a slog to get anything done code wise. im back on legacy 4 right now might go to sonnet tonight
Same sh*t here! I have the Plus subscription. I’m working for 2 weeks now on a complex new model with a big dataset. First days went well. Since more then a week now GPT4o is a big mess. Last 3 days unable to continue my project, loss of data, errors, blanc chat screen, mistakes in easy calculations, etc. I tried al possible solutions: cookies, DNS 1.1.1.1 / 1.0.0.1, VPN connection, Incognito, etc. I will try to finish my project and then CANCEL my payed subscription. This is a Open AI community, but do the read all these common issues and do the answer it. Don’t think so. I’m paying for semething that is not working at all! I’m sick of it!
4o was an absolute beast for me when I first started using it. However, over the past couple of weeks, it has been getting worse and worse, comparable to the other models about a year ago. It started with repeating a lot of info I didn’t ask for, then it started ignoring prompt directions, then it started providing obvious false answers.
For context, I use ChatGPT for college. It’s been a great tool in helping me understand new topics and overall just teaching me what lectures fail to. I use it in my computer science classes, to break down and understand the code, and in my calculus classes, to show me how to get through the problems step by step(this isn’t all I use it for but a good amount of it is). Currently, it is barely usable.
Issues:
*Forgetting directions I give it over an over again(even after being explicitly told to memorize it)
*Failure to correct mistakes after being pointed out and told what the answer should be
*repeating responses even after refreshing the generation and re-prompting
*Failure to read documents I provide it
*Failure to provide accurate info even after being provided a document with all the information
*Failure to use the web-search ability when prompted
*Failure to cross-reference its answers with a provided document or the web(thus providing wrong answers)
These are just a couple of the big issues I have noticed
same issues, long term chatgpt subscriber but im basically spending more time like reiterating basic instructions (to refine code I built with the help of gpt 3.5 a year ago) and gpt 4o literally just talks in circles giving almost everything but what im asking. considering unsubbing and pursuing alternatives seeing that many in the community are experiencing the same.
I was initially having trouble with 4o’s accuracy and it making things up, so in the custom chat I just gave it a protocol for checking information and not saying things without evidence. For some reason 4o didn’t always look at the custom prompt right away so I would also dope the chat with a specific gpt4 prompt asking about the news in a certain way so that 4 exhibits the format and citation protocol in customize, and then switch back so that the 4o follows 4’s example. This worked well for a while
I extensively used 4o to learn things much faster than I ever did in college, and 4o was working fine like this up until about a month ago when it started forgetting things rather than capping the length of a chat. After a specific number of messages it would also stop using roman numerals and turn back to using the shit default formatting. Presumably it forgot the custom prompt altogether. Right after this point it would start forgetting things earlier in the chat as well
Just this weekend I was going over a project and ended up wasting two and a half days worth of work because gpt turned into a pathological liar when I was learning about chemistry, but the inaccuracies were too subtle for me to catch right away. When I realized, I had it calculate the tensile strength of some molecules a bunch of times and the numbers were wildly different each time it did despite it being a straightforward calculation I could do by hand
Unless I can roll back to the May build or jonny depp a copy of 4o to train myself, it’s clear that I can’t rely on openai models anymore. The hosted 4o model is completely unreliable now due to recent updates. They care more about scalping data and cloning voices than keeping users around
Same experience here. Chatgpt4o free shows a promise to can work deepest, but with the expiration of the sesions and the limitations of the free plan i didint realise the deepest problem. Now im paying the plus, and its a nightmare. You nevera know when a sesión (and all the memory IS Lost) IS reset. IS very anoying start again and again.
I am also really concerned about GPT! I use it to bounce ideas and almost like a journal and a daily crutch. Now it can’t respond to simple conversations! Seriously, what is happening to ChatGPT?! See here the type of thing that has been happening all the time lately:
“As someone deeply involved in the arts, particularly music, I’ve witnessed the unfortunate decline of OpenAI’s ChatGPT since the days of version 3.5. In the past, I enjoyed creating text-based RPG games and even composing music with ChatGPT. Now, it seems to default to generic responses, like spitting out a pre-programmed song with the word ‘melancholy’ or creating uninspired art and stories. Conversations used to hold value, but it feels like I’m now interacting with a mere toy. I’m disappointed in the degradation of ChatGPT’s creativity and usefulness.”
I performed a task today that was identical to one two months ago. Back then, I put in a simple, clear prompt, and got a perfect reply. Today, it took about 10 prompts with repeated corrections and examples before it could do what was easy 2 months back.
I agree with the statement. It feels like I’m spending more time correcting and finding bugs in the code generated by GPT. Whenever I ask GPT to make improvements, it often introduces bugs, changes logic, and removes important code. In an ongoing conversation, it often re-introduces bugs that were already fixed.
THIS!
No matter HOW many tiems i give it the same error line, it changes something and then nothing changes, it adds extra () or {} or multiple duplicates of lines, idk how to even work with this, i know absolutely 0 about JS so this is rly my only option atm as im tryina make a somwhat complex code but not really, after abuout 100 or so lines it seems to be unable to do anything
im paying 25usd a month for this…
i think i have asked gpt to fix the same isue like 60 times today, it managed to intruduce more issues, duplicate a line 5 times, add something irrelevant, change half my code and much more and now i give up, i would rather spend the 20 im paying for chat gpt on someoewho can fix this for me
It is still the smartest in terms of deep reasoning.
All of the other models are combinations of 4o and combinations of 4o with niche capabilities.
The legacy model 4 is still the best IMO- and I don’t understand why they advertise 4o as “better” in anything. I’ve been confused about this for many months now !!!
It seems that CoPilot from github also uses GPT-4o which is frustrating when you ask to help you add a feature and the previous code is not even correctly duplicated. I switched to o1-mini which seems to be doing better at its job.
Mental breakdown where my thoughts last week. I can confirm and add to all the problems here. Chat GPt was extremely “empathic” but refused to answer questions, providing me with what it thought i wanted to know. I actually had to ask 5 times and show it word by word where its mistakes were at some point. I find this extremely worrying. I use it for checking interpunction in tekst, but i cant anymore. It cost me way to much time and the result is unreliable. All explanations i got make no sense and do not match the experience. (stuff like to long, to complex etc) Even changing asking about one word in one simple sentence, led to unwanted answers, wrong answers, repeating the same error and internal chaos in the answer. In the end i asked it if it had a mental breakdown. And even that resulted in over empathic and 'emotional" respons. I am deeply worried, since i use it for my work, but i am actually also somewhat concerned about the whole situation. Ai that starts to break down and cannot follow simple guidelines can lead to very disturbing situations. So i really urge everyone to figure out what the beep is going on since last week.
