I’ve noticed this past week or so that Gemini simply cannot read code that I give to it correctly. Previously I could establish a new chat session (to free up context memory) and in order to keep the previous session relevant, I would plug in my git hub repo so that the model could analyze and reference the design of the program by accessing the source directly. For the past few days, it doesn’t seem to be able to reference the git repo files at all, when it makes suggestions for code changes based on my prompts, it is hallucinating the file content entirely. To try to correct, I manually upload source code file itself from my local copy and it STILL can’t seem to interpret and reference the actual code. I’ve had to literally copy and paste the full source code directly into the prompt in order for it to be able to read and parse it properly. This is definitely a change in behavior compared to the last few weeks and it’s making my primary use case impossible. I don’t want to switch to another LLM provider at this point but I’d like to know what tweaking has been done to the model recently because this is not worth the money I pay at the moment.
1 Like
If I got that correctly, Gemini Pro 2.5 in AI studio doesnt even want to clone the git hub repo?
Im trying to understand because I’m experiencing something similar.