Gemini felt dumb on February 9, 2026.

I tried using Gemini 3.0 Pro’s “Deep Think” mode, hoping for deeper, more nuanced answers, but the feature was a disaster. The output was shallow, nonsensical, and objectively worse than the regular model’s reply. It felt like the tool had regressed, leaving me frustrated and unable to rely on the promised advanced reasoning.

I was blown away when I asked Gemini on Canvas to turn a text‑heavy Word doc into a pitch deck. In just 30 seconds it produced 16 tight, well‑designed slides, even condensing the content better than I could in an hour. After a few tweak prompts—swap photos for icons, enlarge fonts, drop a slide—it nailed everything. Exporting to Google Slides gave me instant, granular control. What used to take 2‑3 hours with Gamma or Canva was done in five minutes, feeling like a huge productivity breakthrough.

I tried using Gemini Live while driving with my car’s Bluetooth, hoping for seamless voice interaction. Instead, every time the assistant began speaking, it cut itself off and echoed the same words back as if I’d said them. The constant repetition was irritating and made the feature feel broken, leaving me frustrated that such a basic issue wasn’t addressed.

有時候有點傻

I keep chatting with Gemini and it constantly assumes I’m a British dad with ADHD, even though I’m a Danish woman with no diagnosis. I correct it, it apologizes, then reverts back minutes later and even in new chats. It feels like the model ignores my profile data, which is frustrating and makes the conversation feel impersonal.

I started forcing Gemini 3 Pro to do a “Compliance Drift Check” on dozens of policy PDFs because simple summaries kept missing contradictions. By uploading multiple documents and prompting it to compare intent, the model flagged conflicts, gaps, and risks across HR, security, and SOPs. The output format was clear—policy pairs, conflict, risk level, and suggested owners—making the whole audit process far less painful and surprisingly effective.

I tried using Gemini and kept hitting its overly strict filters—it constantly “babied” me, cutting off my prompts and refusing to let me continue. The tool’s behavior was frustrating and felt like a needless roadblock, making simple requests get tangled in excessive safeguards. I couldn’t get the answers I needed, and the experience left me irritated.