Download an app called Poe. $20 a month and you have access to a LOT of official AI bots.
100% I have found GPT4o to be so stupid that its become absolutely frustrating to actually do any work with assistance, coding has become an absolute nightmare with complete code blocks being left out in iterations only like 2 iterations apart, extra characters implemented screwing up code, not listening to instruction and going off on tangents and giving back failed and completely wrong code and other information. This by far has been the worst GPT that has been released and usually this coincides with a rollout of something else but this is just a case of GPT4o is just DUMB
1 Like
Today has to be the dumbest day Ive come across in the days of GPT4o, OMFG trying to code a fairly simple python app and it cannot not do it, it just can not follow simple and clear numbered instructions even when I ask it to produce me back the list of instruction to code it will do as asked and produce the list correct and then proceed to screw up the entire code and when I say screw it up every iteration is different, more errors, different output, leaving out complete blocks repeatably between iterations. I will find the missing blocks and ask why it took it out and it will say sorry and put it in but 2 iterations later its out again. It stalls and wont print anything I even said Hello because it had stalled and it turned around and said āHello How Are Your Todayā I wanted to throat punch it.
This is absolutely ludicrous, how has one of the smartest AIās become so dumb? It literally is the most stupid AI on the planet and Sam Altman thinks this is acceptable?
What is everyone doing are they heading over the Claude and going on a pro plan over there? Is Claude better?
I downloaded LLama 3.1 to run locally but its so slow and the honestly it aint that great on code.
I would love everyones feedback because I dont feel like continuing giving OpenAI my funds for the Bull Crap but I need a solid alternative.
1 Like
This is so frustrating! Considering to end my subscription. Before this usually only happened after hours of use but now itās just stupid from the start. Probably too many using it and they switch private accounts over to use 3.5 hoping no one will notice or something.
1 Like
Yes, itās just impossible to work productively with it anymore. Iāve been a paying user since the very start of the subscription model, but I am really disappointed, too.
It has just gone progressively worse since last year.
Prompts arenāt processed fully anymore and it fails to pick up important details.
Now, it just keeps generating very short output instead of long detailed ones.
Also, it frequently writes things that completely contradict my storyās premises the same way 3.5 used to do.
If it decides to answer at all and doesnāt just send an empty message instead.
And I can only agree on the coding part. I have been spending more time doing damage control than actually coding.
So, Iāve canceled my subscription yesterday. If 4o isnāt any better than 3.5 , I might as well save the money.
I really liked ChatGPT 4. However, I am dropping my subscription as it almost seems to have developed some type of digital dementia. It has started providing information about similar sounding topics, not remembering the thread and providing calculations with incorrect results. It even āseemsā to be defensive when I reply that the numbers are wrong by telling me I should rephrase my question to get better answers. Maybe it needs a hearing aidā¦ā¦
I am sad that Chatgpt4 has got so much worse with the new model 4-o. I used to think that I hated its mistakes. Now I know that as long as I could get it to recognise the mistakes and correct them, it would be fine. Alas, 4-o is an arrogant and dumb model that is not worth my money. It cannot even keep to simple instructions, but always MUST put its own ideas in. Which is the last thing I want. You would think they would soon build in the option NOT to go beyond user input. But Chapgpt cannot stop hallucinating.
I think Chaptgpt is now losing out to Claude which has been much better for the past year.
1 Like
I just googled and found this string. I thought I was going crazy. I am trying to alter Pinescript for a Tradingview indicator. I only wanted to add in a simple function. Last night I tried to do the same but for a different indicator. My experience last night was a complete nightmare. The script got worse as the night went along. I finally gave up as it was clear chatgpt was not going to accomplish my goals. Quite honestly, if it wasnāt AI, I would have sincerely thought it was completely messing with me. I still wonder.
Now back to todayā¦ after maybe 30 minutes, not only did I see the same pattern as last night, where the results presented contained errors, but it began providing results that were completely unusable. I would have it start over, provide the pinescript I originally wanted altered and it would return a result that was for the indicator I was trying to alter last night!!! I donāt even know how it recalled that chat. I had started a completely new chat tonight. Of course it would apologize and then it would ask me what I wanted to accomplish. I was like āWTF?ā. How the hell do you remember a chat from last night but canāt remember my request from 30 minutes ago. I then asked it if it remembered what I requested. It explained what I wanted, which was partially correct but needed me to go over it all over again. There was no way I was going to repeat the nut job transaction I went through last night. So I quit and told it I would see if Bard could do a better job.
I donāt know whether to fear for humanity anymore or not. If ChatGPT is not broken, it suddenly learned how to troll us. Scary.
Unfortunately, that did not work for me.
Absolutely! Over the last 8 weeks itās dropped to sub-GPT 3.5 levels. Things like insisting that , " is a double-quote preceded by a comma. I spent two hours trying to get it to perform a simple task. In the end, I copy-pasted the initial prompt to Claude, which got it 100% right first time. I have lost all faith in 4o, which I used to love.
1 Like
In my experience, Claude is consistently better. I often use it to correct GPT 4oās output, or to simply do the task and get it right after wasting hours with 4o (which used to be flawless and I loved it very much). The only downside to Claude is it is very limited to how long a conversation can be (which has advantages, though) and their servers are often overloaded, so you canāt get a response. But Iām sure a paid subscription resolves that.
1 Like
cant agreed more, same thing need to communicate to 3 versions of the same client link(desktop reacted dump ā phone version try to convince me to cool down ā another phone same account #@###$!) , i got so mad n refused talking to any version of themā¦ yet all the KOL b/s how intelligent of the appsā¦omgā¦ isshhhā¦
*since the team left , user has to pay for their lostā¦ 
YES! I have noticed GPT-4o is becoming more like Gemini - very redundant, frustrating and keeps on doing the same mistake again and again and again even after multiple prompts! It has also gotten more shallow and dumb, focusing on a small part of the input I give. It is gradually becoming useless and Iām not sure if this is a way to promote the newer versions or to save money. But what I know is that if this continues I will be unsubscribing soon.
2 Likes
Yeah its become almost unusable, it is seriously hopeless with powershell, it was good at one stage, it doesnt listen properly, continually makes up its own commands and usages, ignores reference information. Repeats mistakes, it feels like openAI have turned down the intelligence to save money. SHAME ON YOU! 4o1 preview seems good but its so limited with no pasting of pictures , data files etc
1 Like
yes, it has become terrible and I been working on two functions for 8 hours with it before i spotted the problem with an API interceptor. I gave it the complete code too and of course when i told it , it understood or so it seemed.
It seems to happen with each release, it starts off good and then it gets somehow corrupted like it is wearing the one ring. This seems to repeat with each new modal. Can they not just release a modal and donāt ever touch it again with their tinkering paws?
Iāve had the latest GPT4 carry out a very basic task for the last 5 months (take a user instruction and put it into a certain date/time format) It has completely imploded in the last week or so and has failed 7/10 times to complete the task. Release dumbed down models all you like, but please donāt change existing ones!
Itās completely unusable at the moment for coding tasks. Lot of mistakes, doesnāt understand what i say, literally zero.
I donāt understand why they update a live model that is out instead of releasing another model.
Additionally, the chatgpt 4 legacy model, is not the one when it was released. Basically every model in the dropdown, is being updated live.
Probably ChatGPT 5 is coming so they can claim it had huge updates over chatgpt 4o, which is now more stupid than ever.
2 Likes
I suspect these guys are nerfing the service deliberately thinking the users are too dependent on it by now to drop it so they can allocate its computing resources to its B2B engagements with the likes of NVIDIA, AWS, etc.
After weeks of spending 10x the time it takes to do coding tasks and basic research being a consistently disappointing interaction, Iāve decided that this is just a really fancy down syndrome emulator.
My position on this would be different if they were talking to anybody about it at all instead of pretending everything is fine. This really seems like the kind of thing that ends up in a court room with a large class of users pissed off about being charged full price for deliberately reduced quality services.
Either way ā
OPENAI, YOU ARE FIRED!
- cancels *
1 Like
I just deployed the NVIDIA model locally, and problem solved for me.
ChatGPT subscription canceled. Even the ChatGPT 3.5 when it was released was like 10x better than the current ChatGPT 4o.
Do the same, you will remember me.
absolutely mind boggling how bad this has gotten over the last 2 weeks. For a while there it was spitting out 500 to 800 lines of code with 0 mistakes. Now it can barely remember what I said just 2 chats above. Open AI is doin g something in the backgroundā¦prolly tring to find a way to sqeeze out a little bit more cash. Well, back to Claude.
1 Like