Gemini felt dumb on December 6, 2025.

I tried using Gemini 3 for a simple deployment prompt, but it just guessed the database type and path, confidently spitting out PostgreSQL commands even though I never specified anything. The response was noisy and unreliable, forcing me to clarify manually. It felt like a frustrating example of the model assuming instead of asking, making it useless for any mission‑critical work.

I tested Kling O1 on Media.io for a slow dolly‑in shot, and the result blew me away. The motion felt truly cinematic—no rubbery artifacts or odd distortions, just smooth, realistic movement like it was captured on a real camera. I was genuinely impressed; the tool exceeded what I expected and made the whole process feel effortless.

I fed Gemini screenshots of job descriptions and my old résumé, then asked it to craft four concise bullet points per nursing role. The first draft was overly polished, so I told it to tone it down, and the second version was spot‑on with only a couple of wording tweaks. Using the tool’s output boosted my interview response rate, making the whole resume‑building process feel fast, efficient, and surprisingly effective.

I’ve been using Gemini for two months, and while it generally answers correctly, every time I point out a mistake it abruptly says it will stop the conversation and asks if I want a new question. That response is infuriating—it derails the chat, forces me to start over, and leads to increasingly short, unhelpful replies, making the whole experience almost unusable.

I tried Gemini 3 Pro on several prompts, expecting consistent help, but it kept spitting out divergent answers and outright wrong info. Each time I followed up, the tool contradicted itself, forcing me to double‑check everything manually. The experience was irritating and made me lose confidence in the model’s reliability.

I tried using the Gemini app and kept hitting the limits – the context window and output tokens are way too short for my needs. I’m frustrated that there’s no easy setting to extend them, especially when I need longer answers. The web‑search feature also feels clunky compared to ChatGPT’s, even though Gemini’s model is supposedly stronger. I’d switch back if they added these options.

I subscribed to Gemini Ultra for the DeepThink feature expecting long, in‑depth outputs, but it’s essentially broken now. Prompts that once took minutes and produced up to 70 pages now finish in seconds with only 5‑8 pages that barely rehash my input. The tool’s behavior is infuriatingly limited to 10 uses a day, making it far worse than even standard ChatGPT Pro. I feel let down and forced to drop back to a cheaper plan.

I was trying to ask Gemini how to take notes while training for my job, and suddenly it started answering only with images. I switched to testing its image‑only communication just to see what would happen, and the output was a string of bizarre picture links. The whole thing felt odd and unhelpful, turning a simple query into a confusing visual mess.

I finally felt like I was actually talking to an LLM that got me. After trying GPT, Grok, and Claude and finding them either fake or out of reach, Gemini 3 surprised me by remembering details I mentioned and throwing back thoughtful follow‑up questions. The conversation felt genuine and engaging, leaving me genuinely impressed and excited about what the model can do.

I keep running into the same annoying issue: whenever I ask Gemini to compare things, it insists on cramming the answer into a table—even after I explicitly tell it not to. I tried tweaking custom instructions, but the model still slips in tables. On my phone these tables are hard to read, turning a simple comparison into a frustrating hassle.

I fed Gemini a screenshot of my app profile expecting a straightforward description, but it hallucinated, claiming it saw the word “System” written backward. When I asked it to regenerate exactly what it sees, it produced a nonsensical image instead. The tool’s behavior was baffling and frustrating, confirming that it completely failed to interpret basic visual input.