Gemini felt dumb on October 15, 2025.

I was in a tight chess match, about to clinch a win and hit a rating milestone, when a pop‑up forced me to enable Gemini. It blocked the browser until I hit “no, thanks,” making me lose on time. The interruption was maddening—I felt the tool’s behavior sabotaged my progress, and I’m seriously considering switching browsers if it happens again.

I was trying to build WordPress layouts with Gemini and asked for a YouTube embed snippet. The code it gave worked, but the placeholder video turned out to be the infamous Rick Roll. When I called it out, Gemini cracked a joke about it, saying “Haha, glad you caught the Rick Roll – a little bit of fun never hurts!” I was surprised and amused by the AI’s sense of humor, and overall the experience was helpful and entertaining.

I tried to ask simple, personal questions like whether a plaster could help me quit smoking, but every query got blocked. The constant cancellations felt like the AI was refusing to help with everyday stuff, which was aggravating. I bought my phone for its AI features, and now I'm left frustrated because the tool keeps denying basic requests.

I’ve been testing the new Wan 2.5 model and, honestly, it feels like a jump forward in realism. The images it generates have a depth and texture that previous versions just couldn’t nail, and I’m constantly surprised by how natural everything looks. It’s not perfect, but the improvement is striking enough that I’m actually excited to keep experimenting with it.

I just stumbled onto nano‑banana and gave it a few manga panels to colourise. I was skeptical at first, but the results were surprisingly vivid and faithful to the original art style. The tool handled shading and tones nicely, and I didn’t have to spend hours fixing mistakes. Seeing those panels come to life felt like a pleasant surprise and saved me a lot of manual work.

I tried using Gemini to make images, but it only gave me a square 1:1 format and the resolution looked way worse than it did just a month ago. The drop in quality was really disappointing and made the tool feel unreliable for my projects.

I upgraded to vea 3.1 hoping for improvements, but I couldn't spot any change at all. The experience felt flat and disappointing – everything that annoyed me before stayed the same, and in some cases the responses were even worse. I kept hitting the same errors and had to waste time working around them, leaving me frustrated that the update seemed completely useless.

I asked the model to generate a cool visual effect and it spat out an A/B test using the X28 (3.0 Pro) checkpoint. The result was basically a cloud of particle‑like smoke that reacts to the mouse. It works, but it’s just “okay” – not the striking effect I was hoping for, so I’m left feeling a bit underwhelmed.

I set up Gemini’s screenshare as a live math tutor, writing on my tablet while it answered my questions instantly—its explanations were spot‑on and helped me grasp concepts way better. The only thing dragging me down was the constant crashes; every 5‑10 minutes the AI would just shut down, leaving me mid‑problem. I love how it works, but the unreliability is seriously frustrating, and I’m looking for a fix.

I tried starting a brand‑new chat just to test the Deep Think feature, pasted the prompt from the screenshot, and instead of getting a response I got a bizarre error message about an attached file—even though I never attached anything. The unexpected error halted my workflow and left me feeling confused and annoyed, wondering if the tool is broken for everyone right now.

I tried using the Daily Brief from the Productivity Gem for a couple of weeks, and it initially worked, but now it keeps hallucinating. When I hit “regenerate” it pulls my old emails instead of the latest, and requests to add my iCal events just result in fabricated calendar content. It feels like the context is maxed out, and I’m left wondering if I need to start a fresh chat every time or find another way to sync all my Google data.

I tested Gemini 3 Pro in Google AI Studio on a tough graduate‑level measure theory problem that even the pricey GPT‑5 Pro couldn’t crack. Gemini 3 got the closest to a solution, which felt impressive, but it still tripped over a few details and gave minor errors. Overall the experience was promising yet imperfect, leaving me both hopeful and a bit frustrated.

I tried getting Gemini to create an HTML file I could download, but every time it gave me a weird “sandbox:/test_file.html” link that just opens a search result instead of the actual file. I tested it on different browsers and computers, but the result was the same. It’s annoying because I have to copy‑paste the whole code manually.