Gemini is experiencing errors across the board, giving forced answers, and even potentially changing its response just because I want it to. It’s significantly slower, more laggy, and choppier than before. Gemini 1.5 Pro refuses to answer overly long questions, Gemini 1.5 Flash doesn’t even bother reading my requests, and while Gemini 2.0 Flash is usable, it’s been very laggy for the past few days. Gemini 1206 is similar to Gemini 2.0 Flash. I’ve given up on using the other two Gemini versions. Is anyone else experiencing similar issues?
3 Likes
I am not having the same experience. Do you have some sample prompts you’re using where I can check? Thanks!
1 Like
Well, I used my mother language in that promt, so maybe you can’t understand, but I can told you that, when I ask for answers to be spread across A, B, C, and D, the results I get are only A or B. When I concede and request that subsequent results only be C or D, either it still only returns A or B, or Gemini actually does provide C and D but then he adds a note in parentheses saying, ‘actually the answer is B, but because you asked for C, I’ve put C’ :)))))))))))))
Ah, sorry to bother but is there a reason why Gemini 1.5 pro 002 is experiencing severe internal errors, (In AI Studio) mostly stemming from either long context work and or in long used save chat?
@T_Vu: This sounds like a possible hallucination, but it’s difficult to say. Maybe some improved instructions (either in the prompt or as system instructions) could help you get a better answer.
Can you share the link to the chat? You can find the option to share the link in the “hamburger menu” (three vertical dots) at the top-right of the chat.
@Loading_Completed: It’s best if you ask a new question in the forum for this.
If i do, will you reply? I’ve been waiting for someone at google to at least give answers to some of the few topic that has been asking about the same question I’ve been asking. But I digress, I’ll create a topic.
1 Like
Im having the same problems worked great yesterday AM now its laggy and wont reply to my second request. i think it is not unlimited i think they have an internal token limit.
1 Like
@T_Vu: I am pretty sure the issue you described in your initial comment is a hallucination.
You are asking the model to output a list of words, where in 1 of these words the “syllable stress” is different to the others. Due to how LLMs split words into tokens, it is difficult for them to answer a question like the one you posted. This is why the question “How many R’s are the in the word ‘strawberry’?” became relatively famous (and it’s also why OpenAI’s o1 model was referred to as Model Strawberry by some folks).
In general, questions where you want a model to work with words or parts of words (e.g. syllables) are strong candidates for hallucinated responses.
1 Like
This prompt always worked without any issues. I use Gemini 1.5 Pro and Flash, but after adding Gemini Flash 2.0, it stopped working on the 1.5 models (internal error). And the 2.0 model hallucinates A LOT.
@lluc
Sorry, Forgot to tag.
@vadimor: Model performance (speed, accuracy) is known to diminish as the amount of tokens in the context window increases. Your chat is >500k, which is quite a lot in my experience.
I typically try to keep the amount of tokens to a minimum when possible (e.g. <100k), even if the model supports more. While this does not address the issue you’re seeing, hopefully it helps mitigate it in the future.
1 Like
@lluc Gemini can handle large context fairly well. Everything worked fine before the release of Gemini 2.0, but stopped working afterward.
I have noticed Chat GPT having large issues. I suspect it has something to do with our big winter freeze we’re currently having. It should resolve within the month.
I’m experiencing similar issues. It has been getting stuck in a loop giving the same answer over and over again to different and unrelated questions. It has been having trouble accessing Google Keep today. If I give it the title to a Google Keep Note, initially it says it can’t find it. Then sometime later when I’m doing something else unrelated, it finds the note in Google Keep. It has been pretty reliable until today, 1.7.2025. I uploaded a sample of our interaction in the screenshot.
Seems like they fixed the problem.
I have not encountered this issue. Seems really complex. Good luck with it!