Gemini felt dumb on December 30, 2025.

I tried to put Gemini through the OSINT Exercise #029 as a real‑world benchmark, hoping it would pull together clues and answer the questions. Instead, the model stumbled at the very first step, missed critical sources, and produced completely irrelevant or wrong information. The whole attempt felt like a waste of time, leaving me frustrated and concerned about relying on it for serious research.

I tried using the Gem’s knowledge files today, but it just wouldn’t pull any info out. I opened three separate chats to see if it was a one‑off glitch, even removed and re‑inserted the file each time, and it still failed. The whole thing felt pointless and irritating, leaving me stuck without the data I needed.

I tried to test Veo 2 in Google AI Studio, even though my account shows 10 free slots. Every attempt to generate a city animation ends with “Failed to generate video: user has exceeded quota.” I’ve changed prompts for an hour, but the error persists. It feels useless and frustrating because I can’t create anything despite having available generations.

I tried using Gemini for image edits and generation, but it barely got anything right. Most of the time it repeated the same answer without understanding my prompts, and the conversation died after just two turns. Even simple tasks that should work felt like talking to early Siri—frustrating and unhelpful.

I keep hitting the “loading nano banana” screen even when I explicitly ask for a plain text answer, and it happens far too often. It feels like the app can’t handle the most basic request, forcing me to wait endlessly or restart. The constant loading loop is infuriating and makes the tool feel completely unreliable for anything I need.

I asked Gemini to generate a quiz, but every time the right choice was simply the longest option. It was glaringly obvious, yet the model didn't catch it. I tried tweaking the prompt to equalize answer lengths, which fixed it, but the initial output felt sloppy and unreliable, leaving me irritated that such a basic flaw slipped through.

I’ve been messing with Gemini’s Nano Banana mode and noticed the facial details have seriously slipped compared to the old Nano Banana Pro version. Now it just spits out a bland “Nano Banana” without the crisp features I used to rely on. It feels like the tool’s output has become shallow, and I’m left wondering if this is a bug or an upcoming update that’s throttling the detail. The drop in quality is pretty frustrating.

I tried Gemini 3 Flash to build a personal YouTube recommender from my Takeout data, and it was surprisingly smooth. I could just “vibe code” the whole pipeline—fetching the history, cleaning it, and adding knobs to tweak the feed. The model’s suggestions were spot‑on, the generated code ran locally without a hitch, and I felt a genuine boost in productivity, turning a complex idea into a working prototype in a few hours.