I have been using ChatGPT for help with various Python coding help for some projects. Beyond the fact that OpenAI clearly limits the heavy use of it after using it continuously for a few hours, the actual quality of interactions has declined to the point to where the tool is virtually unusable.
I guess my questions are:
Is anyone else experiencing a steep decline in the quality of responses?
ChatGPT 4o for me now behaves like 3.5 insofar as being completely unable to maintain a context, or ārememberā what was discussed just 2 minutes ago.
Can anyone from OpenAI perhaps tell me if perhaps there is some āresetā function for an account to wipe out all the contexts and maybe start fresh? I know there is a claim there is no such thing, but I am not imagining this. Either my account is āblacklistedā somehow (if thatās a thing) or something is broken.
Either way, when 4o came out, it was rocking as far as quality of interactions, memory and being able to obey instructions and track changes across multiple project.
All that is gone now. Any advice? I am at the point to where I want to close my account and open another one just to see if it functions the same, except my work account is starting to suck as well.
YESSS PLEASE CHATGPT IS BEHAVING LIKE CHATGPT 3.5 now. Itās as if itās been dumbed down for some reason. PLEASE FIX IT. I gave it a question a few days ago and it was able to solve it correctly, but now it tells me that it canāt be solved, or it give me no details at all which is used to do originally
Yeah felt the same. Canāt use 4o for any coding tasks anymore. Does anyone know if thereās a different behaviour when you use it via API than browser?
Yeah I cancelled my work account and switched to Anthropic. I still have a personal account with them but for coding help I would basically labeled 4o as a āscamā almost because itās literally useless. I would love to know how itās possible for an extremely high quality product and service like this to decline so rapidly to virtually being unusable in just a couple of months.
Exactly! Iām having the same problem. ChatGPT 4o is so limited and gives no clear answers, even after multiple prompts with different explanations. Itās like itās not even ChatGPT 3āmore like ChatGPT 0! Does anyone have an alternative theyāre using? Iād rather pay for something thatās worth the money!
Yeah no kidding. I just used it again today for the first time in almost a week and I gave it some python code to analyze and it instantly gave me a long list of useless suggestions which had nothing to do with the code and on top of that were wrong. Whatever happened to it, whether it is being done purposefully to push their users to use the API to increase their revenues, or accidentally (which is probably worse), shows that OpenAI is probably held together by spitballs and toothpicks. Itās a shame because just a month and a half or so ago I was telling friends over lunch about how great of a tool it is.
It was passable (although producing worse output than Claude for example) until a few days ago. Now GPT-4o is completely unusable, ignoring most instructions and being lazy every time. Looks like a pretty sharp decline in quality.
Yesterday and today I noticed some very glaring difference in the ways in which it behaves compared to before and I can successfully replicate those things:
If you give it code to analyze, it will answer immediately providing very long and verbose answers which have virtually nothing to do with the code. Itās obvious that it does not parse the code you provide before answering any questions.
It also now uses strange words like āIF the function normalizes the audio streamā or āthe class MIGHT be producing inconsistent resultsā or āIF the sql table has this rowā - which is very weird because i literally just provided the code in question for analysis or the database schema. When asked about why it uses uncertainty in its answers and I ask it to analyze the code given, then it finally analyzes the data I gave. Itās clear that the interactions are seriously diminished and it does not follow instructions.
The worst one is having to ask it the same thing many times until it finally provides the answer you are looking for. Usually it takes 3-5 exchanges and going back and forth after it prints endless pages of useless data on the screen to finally give a somewhat accurate and concise answer. Instructions to not be verbose are totally ignored.
I would be happy to hear from other users experiences and see how you guys suffer too. lol. I am not yet ready to fully give up on it.
Yeahā¦ it just suggests some kind of nonsense. If I already send him my configuration file, it goes over it and tells me āENSUREā you have set up this and this. OK but I already had this set up. And how am I supposed to work with that? I just have to constantly create new chats, set up custom instructions or use chat gpt 3.5, this version at least doesnāt dump me with hyperinformation.
update: I really think to switch completely to Anthropic, it seems to be better for now at solving some coding tasks.
I have an importat update on the topic of ChatGPT getting ādumberā which may be relevant to others.
I have been using 4o more in the past two days or so and the problems are virtually gone. I have not changed the way in which I interact with chatgpt, so it is unclear to me why it is suddenly offering excellent coding advice, maintaining context and following instructions way better than before.
Whatever changes were made, and if OpenAi staff is behind this, thank you for resolving whatever the problem was. I wish there was more transparency regarding the root cause, but I am thankful the service is back to the quality it was before, at least for me.
I hope the rest of you are experiencing improvements as well.
Itās still dumb for me unfortunately. Iām just trying to change the name of a project Iām forking and Iāve pasted in the code for each file and it gives stuff like:
Navigate to C:\repo\BoseCineMatePulse.
Ensure the solution file is named BoseCineMatePulse.sln.
ā¦
Ensure the task configuration references the correct solution and project files.
Iāve cancelled my paid plan for now, but itās still active until the end of the term. Hopefully it gets back to the original model so I can continue to use it.
This is infuriating. it was doing its job perfectly of putting spaces between the thai words and then translating them. Now its lost all sense and putting spaces between every letter im also having to tell it it is making mistakes before it corrects itsāself. It has the english next to it from the text book for context
100% GPT4o has gotten worse, specifically the past month or so, its now rtarded, like no context, spits out the same crap and over regardless of requests. Itās 100% been downgraded and to claim otherwise is gaslighting. Iāve cancelled my subscription to these snake oils salesmen, and will give me money to anthropic, talking to claude is not like talking to a child with brain damage.
Too late, besides, this happens far too often for this product to be a reliable tool. openai are done now, theyāve burned us too many times and theyre lack of communication is a form of gaslightingā¦āoh no its not worse, you must be crazy, youāre just used to the magicā
Iāve noticed that if I donāt use it for a few days it works better. lol. Itās the dumbest stuff you can imagine but itās true. The more I use it, the dumber it gets and vice versa.
Only ātheyā know whatās going on behind the curtainā¦