Gemini felt dumb on October 9, 2025.

I rely on Gemini’s CLI to churn out CSVs for my Shopify store, but lately it’s been erratic. Most runs just spit out errors, and sometimes it goes silent, not even showing what it’s doing. At its worst it flips into a “paranoid mode,” freezing up or behaving strangely. The whole experience has been pretty frustrating and has slowed down my workflow.

I was trying to shop for a new keyboard and kept running into the same annoying glitch: every time I asked the assistant to add Dell products, it threw an error. I even opened fresh chats and queried other Dell items, only to hit the same roadblock. It felt like the tool just refused to recognize the brand, leaving me frustrated and wondering if there’s some hidden conflict between Google and Dell.

I kept asking Gemini for a new image, but it kept spitting out the same one each time. It felt like the tool wasn't listening to my prompts, which was pretty frustrating because I couldn't get the fresh visuals I needed.

I opened Gemini and tried to copy a generated answer, then attempted to export a DR report to Google Docs, but the website instantly crashed each time. It halts my workflow and forces me to reload the page, losing any progress I made. The constant freezes are incredibly frustrating, and I’m stuck waiting for a fix so I can actually use the tool.

I noticed a shift in AI Studio after the A/B test – the new Gemini 3.0 kept giving the same patterned answers, which was a bit odd, but the biggest surprise was how much quicker it became. Replies now arrive in about a fifth of the time I was used to, making the whole interaction feel snappier and far more efficient.

I was desperate to get the bot to stop its endless loop, but no matter how I tweaked the temperature or token budget, it kept repeating the same nonsense. I even told it to stop and tried banning it, yet nothing changed. The constant repetition was infuriating and made the tool completely unusable for me.

I tried out Gemini and Sonnet 4.5 side‑by‑side during a long, random chat about burritos. The models actually showed a thinking process, which was kind of interesting to watch. Occasionally 4.5 would loop me back in at the end, asking for my take on the conversation. It felt like a mixed bag—some cool moments, but nothing that blew me away.

I tried the 20Q game with Gemini and was amazed when it nailed what I was thinking on the first round. It felt spot‑on and pretty impressive. Then I switched roles and let Gemini pick an object, but instead of asking me, it started questioning itself, which was weird and a bit confusing. The initial success was great, but the self‑talking twist left me scratching my head.