The Problem With Google Gemini

I'm starting to notice a problem with Google's Gemini. They release a new model, the new model does well and gets things more right then wrong, users get excited about using, over time the results get worse and eventually becomes no better than random.

Here's the latest example with 3.7 Flash. I was asking it to update my fantasy football team and it kept pulling in wrong data from two seasons ago.

Trent and Salah have both left Liverpool. Similarly, Onana has left Man Utd. That's precisely the danger with just using AI blindly - you need to know where it is getting things wrong. Otherwise, it could lead you down a rabbit-hole of getting things completely wrong.

And it persists

Everyone makes mistakes and so maybe I should be a bit more forgiving. I told it that it was wrong and it acknowledged its mistake as seen here:

but then shortly after, answering a follow up prompt, it got the player's transfer status wrong again by picking Onana who is in the Turkish league right now:

I ran into something similar when Google released Gemini 3.1. It was amazing and definitely better than any of the other models out there. Then, very quickly, it felt like it got nerfed, and I was only seeing average results at best and while it was a little productivity boost for tasks that I was familiar with, it wasn't very helpful for unfamiliar tasks.