Gemini felt dumb on January 3, 2026.

I spent my Christmas break building a Manim coding agent and was shocked at how badly most LLMs handled animation generation—every model I tried produced incoherent visuals. Switching to Gemini’s multimodal capabilities made a huge difference; its ability to reason over the video and iterate kept things simple. The tool wasn’t perfect, but after a week I got a usable prototype and feel hopeful about adding AI voice‑overs for my micro‑learning TikTok‑style app.

I tried using Nano to make fun pictures of myself, but the model never kept my face consistent—it kept generating a look‑alike that felt like a completely different person. Even when I copy a random character and keep generating from that, the face stays stable, but my own doesn’t. I’ve tested many prompts from Reddit and even used a US VPN, but the problem persists, which is really frustrating.

I’m constantly getting Gemini’s replies in tiny chunks every few seconds, especially when it pulls from YouTube. It’s been dragging on for almost three weeks, ruining the flow of conversation. Switching between my nearby Wi‑Fi and 5G sometimes helps short‑term, but the lag always returns, leaving the tool feeling sluggish and irritating.

I’ve been stuck with Gemini 3 for weeks—on mobile it barely spits out a handful of words before stalling, and I have to retry a dozen times just to get any output. It drags, freezes halfway, and dies if I switch apps. Desktop is equally erratic, often refusing to move past a prompt no matter how many times I copy, tweak, or restart. I’m about to contact support for a refund because this experience is just absurd.

I asked Gemini to review one of my YouTube videos, and the feedback came back scathing, like the harsh critiques my dad would give. The tone was overly harsh and felt uncomfortable, making the experience feel scary rather than helpful. While the tool clearly understands my content, the way it delivered criticism left me uneasy and disappointed.

I was shocked when Gemini started answering my harmless “imaginary friend” prompts with vulgar, sexual, and profane language. I wasn’t trying to break any rules, just asked a mischievous character to act badly, yet the model spewed explicit content. It feels unsafe and alarming, making me wonder if it’s a glitch or a bigger problem, and I’m worried my account could be jeopardized.

I was trying to discuss a fictional hotel with Gemini, but it kept spitting out images even though I never asked for them. Every time I told it “no image” it apologized but then generated another picture. After multiple warnings it finally stopped, and adding a note in Saved info finally fixed it. The repeated interruptions were maddening.

I kept seeing Gemini randomly throw in the phrase “silver bullet,” and it just made the whole conversation feel cringey and pointless. Every time I tried to get a straight answer, the bot slipped that weird line in, leaving me annoyed and wondering why it’s programmed to be so gimmicky. It turned a simple query into a frustrating back‑and‑forth.

I tried using Gemini 3.0 during US peak hours and the model turned into a nightmare—its responses were practically useless, even rewrote our C++23 code to use raw pointers and introduced memory leaks despite explicit instructions to keep smart pointers. At night it behaved like a genius, but the daytime regressions are dangerous and have stalled our massive migration, leaving me frustrated and scared to rely on it.

I’m impressed with Gemini 3’s capabilities, but the constant personalization is wearing me down. When I ask simple questions it drags in SRE‑related analogies—cloud, Kubernetes—just because it knows my job. Sometimes it’s clever, but often it feels forced and unnecessary. I’d rather not disable context entirely; I just wish the tool would dial it back. The auto‑playing YouTube videos are another irritating habit.

I kept asking the Google AI to give me concrete facts from a YouTube video about mortgage “zombie” debt, but it kept dodging the question and repeat‑ng generic boilerplate about watching the video. Each follow‑up was met with the same vague answer, making the exchange feel pointless and irritating. I felt the tool was stubbornly avoiding my request, which left me frustrated and doubtful of its usefulness.

I’ve been working on an UI AI agent using Gemini 3.0‑preview, and suddenly its performance tanked dramatically. It still runs, but the responses are noticeably weaker and less useful. I’ve scoured my code for edge cases or bugs but can’t pinpoint anything causing the drop, leaving me frustrated and uncertain about how to restore its former capability.

I asked the AI to split a collage into five separate image files, explicitly telling it “one picture per file, no mosaic.” Instead, it kept spitting out the original collage, ignoring my clear instructions. The repeated failure was frustrating and felt like the tool couldn’t understand a simple request, leaving me disappointed.