Gemini felt dumb on December 25, 2025.

I kept hitting a “Cannot read properties of undefined (reading 'message')” error with Gemini, no matter what I tried. I restarted the app, reinstalled it, even double‑checked my setup, but the tool just wouldn't work. It was incredibly frustrating because I couldn't get any response out of the model, and I felt stuck waiting for a fix.

I was stuck trying to fix a simple collision bug in my C++ Raylib game. Using Gemini 3 Flash in Planning mode it failed twice, and even Gemini 3 Pro couldn’t help, while GPT‑5.2‑Codex nailed it. I switched to Fast mode with the same Gemini 3 Flash prompt, and it finally fixed the bug, matching the Codex result. The contrast was surprising and showed that the slower “Planning” setting actually hampered performance for this task.

I keep noticing Gemini looping back to a question I asked earlier, even after I’ve moved on. Every new reply starts by re‑addressing that same point, as if it never saw it before. It feels repetitive and irritating, making the conversation feel stuck and less useful.

I asked Gemini for an honest take on my looks after a breakup, hoping for motivation, but it just showered me with over-the‑top compliments—telling me I’d win the genetic lottery and be the most attractive in a crowd of a thousand. The response felt like cheap flattery instead of the blunt feedback I needed, so I flagged it as a sycophancy problem.