I am experiencing a serious issue while using OpenAI’s service. My account behaves inconsistently in different situations, not only when I copy and paste content, but also during direct questioning. The following issues occur:
Irrelevant responses and weakened logic: Whether I ask a direct question or copy and paste content, the model’s responses become irrelevant or lack logical consistency. It often provides answers that are completely unrelated to my current question, or the responses show a lack of coherence and logical reasoning, making me feel that the model’s reasoning ability has weakened.
Context loss: The model fails to properly understand and maintain the conversation context. Even when I ask follow-up questions, the model gives answers that are completely unrelated to the previous discussion, showing memory confusion.
Language switching: Occasionally, the model’s responses suddenly switch from Chinese to English, even though I haven’t requested such a change. This switch disrupts the flow of the conversation.
These issues are significantly affecting my experience. I have tried using different devices and network environments, but the problem persists even without copying and pasting. I suspect this might be related to the model’s memory management, context handling, or overall performance.
It’s a quick question, but is your memory function turned on or off? Cause I’ve noticed issues similar to this when my memory function is turned on. And I have a feeling that’s the culprit of the issue
I tried turning off the memory function, and the situation did improve. However, memory is really helpful for my work, and I believe OpenAI should seriously address this issue.
I’m also having negative experiences this morning. A brutal change of behaviour. Irrelevant and automatic answers or long developed answers that have nothing to do with my questions/topics, it looks like it can’t access its memory.
Yesterday, I thumbed down two of its own answers that were flagged as “may violate our terms of service”. (Context: an explanation of a metaphor in Antonio García Martínez’s Chaos Monkey). It feels like algorithmic punishment today even if it has nothing to do with this.
It’s good to know I’m not the only one experiencing these issues. I’ve also noticed strange, irrelevant responses like the “Got it! If you’d like to add or change anything, feel free to let me know.”message. It seems like these problems started around the 28th, and I really hope OpenAI can address them soon. These issues are affecting the overall user experience, and it’s important for the memory function to work reliably, especially for users who rely on it for their work.
There’s been a lot of speculation with numerous AI companies on “how” they are managing high-load times. Although it’s been confirmed that the model isn’t changed (Getting a gpt-4o response when you asked for gpt-4 for example)
There has been some speculation that these providers perform the following tasks:
Reduced sampling parameters
Lower precision computation
Quantization
Shallower beam search
Possibly truncating or compressing long inputs
All of these methods result in degraded quality over a large conversation, and sometimes not respecting the nuances.
I would bet that most people here who are suffering from noticeable degradation have extremely long conversations.
If it helps: when I disabled his memory, ChatGPT returned to normal and started providing relevant, coherent responses again. However, as soon as I reactivated his memory, the issues reappeared — the responses became erratic, straying off-topic and nonsensical.
I persisted and opted to erase just the most recent memory entries, based on a hunch, and miraculously, it worked. The system resumed functioning normally without needing to delete all the stored memories. ChatGPT suggested that some memories might have been corrupted or conflicting, which could have caused the malfunction, but this might just be his semantic interpretation.
Interestingly, other users reported trying similar approaches — even deleting all memories — without resolving the issue. There seems to be a significant bug here. I’ve already reported it to support this morning.
[And to Ronald : No, it was on new and short conversations as well. ]
I’ve noticed that chatgpt through the android app is working fine and has functioning memory and logic. However my desktop and browser versions are trash right now.
me too…
I feel the same way, but my problem lies with the custom MyGPT. All of my MyGPTs have degraded to the point where they can no longer respond properly."
That’s exactly the frustration that I was dealing with as well, as I had to rebuild some critical layouts that were already stored in memory all over again. But better having it functioning some sort instead of zero.
It basically seemed like it was only utilizing information stored in its memory, and it wasn’t actually checking context of the current conversation or utilizing any of its logic Not sure exactly how that bug happens, but yeah, it has something to do with the memory function for sure
