Gemini felt dumb on January 20, 2026.

I noticed a weird pattern: every AI I tried—ChatGPT, Gemini, Qwen, Claude, GLM, Grok—kept generating walking animations that moved sideways instead of forward. The limbs looked off, and the code they supplied didn’t fix it. I had to dive into the script myself, flip the XYZ axes, and finally get a normal walk. It was frustrating to see each model repeat the same mistake.

I’m still reeling from how awful that model turned out to be. Every interaction felt like a nightmare—its responses were chaotic, nonsensical, and sometimes downright dangerous. I kept hoping for a coherent answer, but the tool kept spitting out garbage, leaving me frustrated and uneasy about trusting any AI that behaves like this.

I spent a day parsing 15 PDFs—about 400 pages of financial and technical data—to spot inconsistencies. ChatGPT forced me to batch‑upload, hallucinated numbers, and forgot earlier files. Gemini let me drop all PDFs at once, returned a precise table of Q3 spend, flagged the three mismatches and even gave exact page references. The contrast felt like a superpower for heavy research.

I tried the Gemini web app hoping for a smooth chat experience, but it felt like a step backwards. Compared to ChatGPT, Grok, or Claude, it was clunky and unreliable, and even lagged behind basic open‑source GUI chats. After getting an API key and switching to LibreChat/Cherry Studio, I realized I’d never want to return—its errors and poor UI made it frustrating and unsafe to rely on.

I tried Gemini’s Chrome integration while scrolling through videos and social feeds, and it instantly grabbed the tab’s context to fill in gaps I didn’t even realize I had. The browsing felt smoother and far more informed – the tool’s intuition was almost uncanny, turning routine surfing into a remarkably helpful experience.

I tried to use Gemini to generate a beach scene with me in a bikini, but the model suddenly blocked that request. It was confusing and disappointing—what used to work now just returns a refusal or a bland alternative. The sudden censorship felt like a step back, making the tool less useful for the creative edit I needed.

第一次的回覆常常不正確

I tested Gemini 3.0 on a bunch of writing tasks I rely on AI for and kept hitting the same problem—its output was half‑finished, riddled with omissions, and it kept forgetting explicit instructions. Compared to ChatGPT, it skimmed documents, hallucinated details, and produced far shorter, less detailed answers, leaving me frustrated and wondering why it’s suddenly so “lazy.”