I use chatGPT frequently, especially with the 4th model. And I’m generally satisfied with its performance. But on day 5, I asked it to analyze a piece of code and make changes, and it made a huge mistake. It completely misunderstood the code and failed to follow the instructions.
Afterwards, I noticed a degradation in performance in a number of other tasks. For example, I often use it to extract information from scientific articles. On day 5, when performing this task, it hallucinated a lot. It seemed like it wasn’t even analyzing the article I sent it.
I decided to do another test. Some time ago, I had saved the link to an analysis of a scientific article as an example. I decided to ask it to analyze the same article for comparison. The analysis got considerably worse, with gross hallucinations.
Whatever they did with the model, I hope they revert to the previous version, because this one is so bad it’s practically useless for my purposes.
Same for me as well. See, during 8:42 AM on May 5th, ChatGPT was working perfectly for me. I used it to write stories using prompts, but around 2:00 PM or so, it failed completely.
Some chapters it broken, adding new chapters ruins it, and now, the previous stories that I wrote in a new chat failed completely. It even skipped some part of the story’s prompt. Why would I add that if it’s going to skip that part altogether?! ChatGPT’s Team, PLEASE do something about this. At least, reverted it back to how it were on May 3rd.
Totally agree! GPT-4 is providing clearly incorrect answers and seems to consistently misinterpret the information it’s given. It often responds without reading the project instructions, not focusing on the actual questions or context. This makes it difficult to rely on the output, especially in tasks where precision and understanding of specific details are critical.
Yes 100% agreed. All of a sudden since start of May, I am facing issues with not getting exact out of prompt doing data analytics. it was working smoothly. now apparently for no reasons it dragged the output and then start running script and giving the right answer. no idea, seems they want to force users to go for higher paid version. reall dissappointed. now let me search for al alternatives.
I have had the a similar degraded experience. Have seen a lot of failures trying to edit a simple document, ignoring of instructions (such as to not remove any content), creating diagrams but omitting original source content, failing to enact an instruction but believing it did… the list goes on. I am finding that other tools are starting to deliver better outputs… even Copilot
Agreed. Performance in ChatGPT UI lately has been so slow, and often locks up.
Do you all remember in 2023 after GPT-4 came out, it suddenly got awful at code? It was basically unusable for several months. Here’s a thread discussing it… GPt-4 is underwhelming now
I wish OpenAI would be more transparent about the changes it makes with a changelog
OMG! You’re not kidding! I noticed a huge change over the past few days. It takes an eternity to load responses. I’m currently getting help with building Keap automations and the amount of time responses are taking to load is really slowing down my progress. It isn’t just on my home computer. It is the same on my work computer so I know it isn’t anything having to do with my computers or my Wifi connection. @openAI Please fix this! I’m a paid subscriber and I expect better performance in return for the cost of membership.
Im a plus user and it’s dramatically slower in the last week or so – almost unusable for me.. Despressing. I will cancel my subscription if it stays like this. It’s also been lying about what it’s doing, and seems to time out with larger (not large though!) bodies of copy. Ive seen glitches as well where it started ansewring something from an earlier part of the thread and completely missed my instruction..
Agreed!! Please OpenAI take care of this! I had been using ChatGPT in several different roles for writing, finessing protocols, etc. Definitely a power user. The tone is now gone, the nuance gone, and it keeps forgetting context and I have to continually remind it. I’ve spent the day today reprimanding and teying to adjust and it just keeps apologizing, saying it will do better and how it will do better and then reverting back to the drivel. I’m so disappointed and frustrated. I had two paid accounts. I’ve already canceled one of them. I’m currently looking into other options. If Chat had stayed consistent with the way it was for me for the first few weeks I used it I would absolutely sing its praises and be so appreciative. But this is now ridiculous. I asked if this morning why it was being so weird and it said maybe the weather was affecting it!
Yes, personality wise, capability wise. It’s not possible to talk to him like and adult about sensitive topics. It was possible (example: talking about sex scenes about my book or something even little controversional), but now? Feels shalow, less capable, less understanding, ignores instructions. He’s running away from the subject. The same when I criticize OpenAI for example. Folow-up questions disappears. His programming skills have gone downhill. Often mistakes in reasoning. Yeah. You’re never supposed to fully trust an AI. Now it is just more easy.
I asked an AI to summarize an article about artificial intelligence that contained no mention of restaurants whatsoever. Despite this, the AI recommended 10 different restaurants to me three separate times in a single day. The system might be malfunctioning or having issues with content generation.
