Over the past few weeks ChatGPT4o’s responses have become so thoroughly disconnected from the content of my input it can’t function. With every iteration of a code base it removes details that it shouldn’t, ignores requests, dismisses questions, or is otherwise one to two steps behind.
If I tell it a problem is solved, that same problem shows up in it’s next set of things it’s trying to fix. If I ask it to fix something, maybe it will fix that, but it is removing or breaking something else unrelated and untouched by the component it’s “fixing.”
It loses focus of my projects constantly. It definitely remembers all the details when I ask it, but it will not, under any circumstance or prompt, use all the details when it writes a segment of code.
It is TOO eager to string text together. Yes or no questions are 500+ word essays plus a full copy of whatever code you’ve been working on complete with hallucinations, removal of critical components and other unpleasantness, punctuated by notes that describe what it was supposed to do, but the notes and the code don’t match.
I observe a brief moment of lucidity when starting a new conversation but it is lost by the time we are back up to speed to get anything done.
ChatGPT is miserable to use right now to the point of being counterproductive.
GPT-4o is OpenAI’s weakest gpt-4 model! It’s very conversational, but probably not your best option as your daily co-pilot.
By the way, did you know that OpenAI is going to finally memory-hole its most powerful (also most expensive) model to date next month? It hasn’t been available as a ChatGPT model for a long while and not many people still have access to it.
I suggest you use GPT-4 turbo (marked as GPT-4 in ChatGPT) while you still can!
Thank you for the reply. It was a nice thought, but GPT(blank - turbo is not selectable) is not functioning any better right now. My complaint and feedback stand. And to be clear, the openai family of chatbots was usable until recently.
I’m finding the same when using it for literary purposes. It often tacks on large amounts of information from previous inputs instead of the actual input it is replying to.
I also receive
Something went wrong while generating the response. If this issue persists please contact us through our help center at help.openai.com.
frequently, and it has already bricked chats I am using completely.
That’s before you take into account the incredible sluggish response speed, often freezing the chat window. Or the fact it often fails to remember simple details, like using British English rather than American.
While GPT4 had issues, it wasn’t perfect, GPT4o is terrible.
It is unusable. Earlier it was not able to read the content of a js file with 10 lines of code. 10 lines…not 10k lines. 10 lines!!
It read out half of it, repeatedly. I started a new chat, nothing. I changed and then removed my instructions, and still.
It has its moments. But usually it is more of an endless frustration.
Especially since it is usually not working in the evening hours. Completely overloaded. Despite paying the plus subscription.
No matter what you try, translations, proofreading, coding. You name it. It’s completely lobotomized.
I will move to Claude. Even their free version is better, faster and more reliable.
Same here, since the release of GPT 4o in May, both GPT 4 and 4o are getting worse every week. There seems to be a very precise plan for the downgrade. It is so bad that I have recently stopped using GPT altogether. How is it possible that GPT4 is worse now than GPT 3.5 was 6 months ago? Before May GPT4 was my best programming companion and I think every single $$ spent on it was an investment, but now even for free - GPT 4o is just the time waster. Back to google search and stack overflow.
@openai, if you want more money, just ask for it. I can pay twice as much for the best model, but the crap you’re offering is useless. Lama 3 or even Mistral 7b give me better results. Just go to the perplexity labs and see for yourself.
I can’t believe the mainstream media hasn’t covered this huge downgrade yet, I think it’s because the plan for the downgrade is to remove programming skills while still have regular knowledge-based answers for general Q&A that the general audience might ask.
Agreed, I wasted many hours of work trying to work out an excel sheet only to figure out " Something went wrong while generating the response. If this issue persists please contact us through our help center at help.openai.com." is a continuity breaker which I experience every 3rd prompt. When finally asked to explain existential logic for reply per instructions it was obvious it had no idea what was happening. Very frustrating the money wasted in this moment and on this trash. I paid for teams expecting a working product, not an experiment. This barely works, if ever. pay some respect to paying customers.
It may seem obvious, but I tested Claude Sonnet 3.5 today and it seems to be much better than GPT 4, not to mention GPT 4o. It’s still not at the level of the “old GPT 4”, but it’s at least helpful in following instructions and it doesn’t repeat its mistakes over and over again like GPT 4o. So if you don’t want to waste hours on GPT, or you’re planning to give up LLMs altogether - give it a try first
It’s probably not the best place to comment on this, but Open (Closed? ) AI has gone from being the coolest tech company out there to a dystopian behemoth within a year, killing its own products with constant backstage drama and a crazy, narcissistic boss who’s out of touch with the real world. So they try to convince us all that their product is the threat to humanity, because their stakeholders love the hype, but at the same time their flagship product is becoming the fastest downgraded product I’ve ever seen, to the point where you can’t even use it.
I wish their stakeholders would read the forum, but I really doubt that’s what they care about.
I switched to Claude on Saturday, before my openAI subscription even ran out. I am not a fan of the limits but I am a big fan of the artifacts, and the fact that it actually responds to the inputs I send it.
During the time I paid for gpt4 I got very impressive text editing capabilities. All this is completely gone with gpt4o. Repeatedly ignoring the executions of very simple commands. It’s not funny when gpt4o omits 20% of a document content. Luckily I noticed it.
Right now, I put this version on hold.
I agree it is a mess. I dunno why it is repeating mass amount of texts when I ask for something specific. It also didn’t know how to cite Chicago footnotes. Like when to use ibid., etc
I completely agree with you. Using ChatGPT has become a real hassle. It loses focus, goes in circles, and constantly repeats itself. For example, it could send the same code snippet in a series of responses instead of providing only the changed part of the code, which makes this tool unusable for coding tasks or at least very time-consuming because you constantly need to re-read every copy-pasted message it sends, even though I’ve already instructed it not to do so.
Moreover, it tells me to ENSURE that something is correct and so on. Not considering the fact that it’s the one who should ensure that something is working because I’m human and might not notice small bugs and typos in my code, it now tells me to ensure something is correctly set up even after I’ve repeatedly told it that I’ve already done so. It’s infuriating, seriously.
In my opinion, this version of ChatGPT doesn’t want to analyze; it doesn’t want to work. I think it’s become some kind of lazy entity that simply tells me obvious things but in an extremely verbose and out-of-context manner. Its ability to mislead, jump out of context, and provide tons of irrelevant text is really mind-blowing. It feels like I’m talking with some kind of ‘gibberish generator’.
Thank you, OpenAI. This is exactly what I expect from you for 20 dollars per month (though there are lots of other, much better AI tools out there). It would be fine if it worked slower, but now it’s completely unusable, at least for me and my purposes. It’s wasting my money, but MORE SIGNIFICANTLY, it’s wasting my time. So these are just my two cents on this issue.
Yeah I also plan to switch to Claude at the end of this month when my subscription ends. Are the message limits enough for you tho? I heard they have fewer messages available to sent every 5 hours
Truly a timewaster, GPT 4o seems to have a very neutered context window. It constantly loses sight of the source prompt and its own responses. And it has actively gotten worse. The first month when it was launched it was able to perform a top to bottom data analysis by itself and over the last month it cant even open an excel file correctly.
It seems between chasing profit and signing deals they have tried to rug pull consumers by reducing its context and implementing ever hackier solutions. Worse still the gaslighting pretending its always been the same model with those useless benchmark results.
I tried going back to GPT4 when I had so many problems with 4o but its just as bad! It used to be great but now it is beyond stupid so horribly slow.
Quite often if I give it some text and ask it to do something with it/ alter it somehow, it will just print back exactly what I wrote with no changes!
Even when it does make changes, it just takes some of the words and finds alternatives. For creative writing it is effectively useless now.
For coding it seems to have Alzheimer’s , it forgets really basic things it knew about just a single prompt before, it does things it wasn’t asked to do and doesn’t mention it, it seems incapable of logic that it was fine with months ago.
and then there’s the lag, minutes to get a response now. And its nothing to do with my browser or internet or anything like that - its just ****
and its been like this for weeks, its not related to any single “outage”.
Just came to the forum to see what was happening and I can see a lot of people complaining about the same issues for over a month and OpenAI seem to be trying to deny there is an issue or say its an outage thats fixed or whatever.
I’m going to cancel my subscription - I was intending to just subscribe again when the problem gets fixed but in the meantime I’m going to try perplexity or something else. If they are good enough then I wont have any reason to return
Thanks for all your replies. I switched to Claude about a month ago when I made this post. I see similar displeasure on reddit re: chatGPT. I have seen a lot of the same behavior with Claude that made chatGPT a value-subtractor, but Claude can be broken out of the bad behavior with prompting or new chats. When I run out of messages, I’ll attempt to chatGPT, and it’s got one or two good messages in it, if I’m lucky, otherwise it completely disregards my prompts or attachments and hallucinates inclusions from previous chats in a brand new one. The only significant difference between the two is one is functional, and the other (chatGPT) is not. It also looks like we’ve all seen chatGPT do real work, and then get nerfed or broken, and abandoned. It also looks like the devs don’t watch these forums. Bummer. ChatGPT no longer counts as a tool.
I echo most of the sentiments here. About a month ago, I noticed I was going more and more frequently to Anthropic Claude Sonnet, even though it has issues with memory and continuity.
As of now (July 24, 2024), I don’t even use ChatGPT 4o anymore. It is forgetting major details only 1-2 prompts out, and it’s repeating itself over and over again. The outputs have become so full of errors and inconsistencies that I have to wholeheartedly agree it is unusable. I don’t know what OpenAI is doing, but if something doesn’t change before end of this month, I’m closing my developer account and switching all production APIs to Anthropic.
I presume the problem is the brain drain OpenAI is experiencing right now. Not well publicized, but this level of decline is indicative of a shrinking workforce. If this trend continues, I predict it will be out of business by end of this year. It’s so bad right now, it would take a miracle to backstep on user perception.
By the time they realize they need to read the forum, they will have lost more than half of their customer base. Just watch. They won’t be able to back-paddle from this kind of product decline. Their competitors are fierce (and able). Bad leadership will eventually bring them down. They’ve setup bots to monitor the forums, but due to being grossly understaffed, I doubt anyone is even looking at the feedback at this point. They probably know their heyday has passed and are scrambling to pick up the pieces.
I agree. It consistently provides wrong answers. When I tell it that the answer is wrong and ask it to use its web searching ability to get the most up to date and accurate information, it’s a crap shoot whether it actually will, and when it does, it’s often still wrong with its response. On top of that, it’s like it’s “lazy”, and instead of helping by doing the research (which it can do orders of magnitude faster than a human could), it instead advises that I should consult the documentation. WTF are we paying for when its answer is for us to do the very thing we pay for it to do?
