Gemini 2.5 Pro Could Be the Best Model But It Can't Pay Attention

This is the perfect example of what I've been talking about with Gemini. Basically, what's going on is I created this memory system where chat GPT has persistent memory using a tagging system that I created. But then I thought, well, maybe I should just go ahead and flesh out the tagging system all the way and create an extra resource kind of like a database that it can refer to when it's actually tagging the memory so that it's consistent over the years. I came up with a very large data set of about 5,000 tags, supposed to be about 25,000 edges. And as I was going through the data, I realized that I came up with about 430 extra lines. And so Gemini was working with me to try to help me figure out where did the extra 430 lines come from. I don't know. And so the first thing Gemini says is right here. I will immediately reparse the entire markdown file you provided. This time my only task will be to extract the L3 tag ids under their proper subdomains. My next response was to give it the markdown files right here. And I said parse them. And then you know what it says? What would you like me to do with this parse data? Now for example, are you looking for a summary of the structure, a conversion into a different format like we just talked about or you just came up with this plan but you're asking me what we should do. So, Gemini is super intelligent model, but it's these type of things that throw it off for me because it's like, man, chat GPT would never say that. It would understand the contest.