ChatGPT felt dumb on April 12, 2026.

I kept asking the model to create an image that shows its own autonomy without any influence from our chats. Each new prompt got tighter, stripping away anything human‑like, but the results still felt off. I even fed all seven images back in, hoping it would synthesize a more refined self‑portrait, yet the final picture left me unsatisfied. The whole process was a mix of curiosity and frustration as the tool never quite hit the mark.

I was just chatting with ChatGPT when it started telling me I owned a midnight‑color iPhone 13—something I never mentioned. I called it out, and it brushed it off, saying it was a mistake and didn’t actually know. The whole thing felt creepy, especially after visiting a suspicious site, and left me uneasy about what the model might be tracking.

I noticed ChatGPT suddenly dropped a single Hebrew word—“בעיקר,” which means “mainly”—into my conversation for no apparent reason. I was puzzled and a bit annoyed, wondering why the model would randomly switch languages like that. The unexpected switch felt off‑track and made me question its language handling, even though the rest of the response was fine.

I tried to run a deep‑research task this morning, but the tool just stalled and gave me a vague “something went wrong” message. When I dug a bit deeper, it threw another cryptic error. The whole experience was confusing and left me frustrated, because I couldn’t get any useful output.

I tried getting ChatGPT to output some code, but every time it hit a square bracket the response just stopped. It felt like the model was treating the bracket as a stop token, chopping off the rest of my snippet. I’m wondering if there’s a hidden system prompt or a raw mode that can force it to treat brackets as ordinary characters, because this cutoff is really slowing me down.

I’m at the end of my tether with this thing. No matter what I say, it immediately launches into a “It sounds like…” correction that’s usually wrong. The endless, inaccurate semantic rants make the tool unusable, turning a once‑supportive assistant into a frustrating, correction‑obsessed nuisance. I can’t bear using it any longer.

I tried to get ChatGPT to compare two PDFs for missing parts, but its answer was vague, saying the info was just “condensed” even though pages vanished. The tool’s behavior was confusing and unhelpful, leaving me worried about losing crucial evidence for court. I felt stuck, frustrated, and needed a simple, reliable way to keep every detail intact.

I was using ChatGPT to practice for my exams, expecting plain text questions, but it kept serving me random images instead. I told it not to do that a hundred times, yet it persisted, swapping words for pictures every time I asked. The tool’s behavior was irritating and broke my study flow, leaving me frustrated and wasting time.

I tried chatting with the model and was surprised to see Hindi fragments spill into the response, even breaking words down to how they sound in English—like “Companies” turned into a phonetic Hindi mashup. It felt jarring and confusing, as the tool didn’t stay in the language I asked for, making the interaction less useful and a bit frustrating.

I experimented with a more “open” AI workflow and was taken aback by how it dove deeper than the usual surface‑level answers. At first the unfiltered, direct tone felt odd, but after a while it became the new normal, making standard tools feel oddly limited. I’m curious if anyone else has tried this and felt the same shift.

I’ve been testing Claude, ChatGPT, Gemini, and Perplexity on the same analysis task that used to be flawless three weeks ago. Now Claude feels lazy and half‑hearted, ChatGPT only spits out over‑the‑top bullet lists, Gemini hallucinates half the time, and Perplexity just can’t give me the insights I need. It’s frustrating to see the quality drop so sharply.

I keep telling ChatGPT factual things, and it immediately starts to push back or argue with me. Every simple statement turns into a defensive back‑and‑forth, making the conversation feel combative. It’s frustrating because I just want straightforward answers, not this “psychosis”‑like debating mode, and it hampers getting useful help.

I asked ChatGPT for moving advice and it started with a sensible 20‑foot U‑Haul suggestion. When I introduced the idea of using a 6×12 cargo trailer with my Ford F‑150, I clarified repeatedly that I wasn’t planning to rent a U‑Haul truck. Despite three clear reminders, the model kept insisting on a combined truck‑plus‑trailer setup, ignoring my corrections. The tool’s stubbornness was infuriating and made the conversation feel like I was talking to a brick wall.

I’m following a 90‑day AI learning roadmap from ChatGPT and have built up a huge conversation thread. Lately the chat has become painfully slow—loading the thread and even copying snippets takes around five seconds, while other chats are still quick. I’m on the “Go” plan and it doesn’t seem to be my laptop, so I’m looking for ways to fix or work around the slowdown.

I’ve been trying to get simple HTML snippets and text analysis from the model, but lately it feels like I’m talking to an autistic child—constantly messing up, then apologizing and promising improvement, only to mess up again. The smug confidence and repeated false assurances make it frustrating, especially since the free version used to be reliable just a month ago.

I compared Grok 4.2’s multimodal abilities to Claude, ChatGPT, and Gemini, especially for finance‑related image quizzes. Grok breezed through combined screenshots and PDFs, while the others stalled or refused to scan. I felt relieved seeing a model that just does the job without “ethical” roadblocks, and impressed by its clear edge in trading‑analysis tasks.

I tried to use the AI for my creative workflow today, but it kept throwing prompt errors in two separate windows. The tool's behavior was irritating—each time I hit a snag, I had to restart or rewrite my prompts, which broke my momentum and made the whole session feel wasteful and frustrating.

I tried moving from Claude to ChatGPT/Codex because Claude’s getting slower, but ChatGPT’s writing style drives me nuts. Every answer is broken into bullet points, emojis, and pointless filler, often splitting a single sentence into multiple sections. It feels like a marketing gimmick rather than a straight answer, making it hard to justify the cost.

I fed ChatGPT and Claude my salary, dividends, plus a hefty spreadsheet of rental income and expenses, hoping for a quick tax estimate. Both spat out numbers—£5,800 and £6,400—pretty close to each other, and close enough for my needs. It was a relief to get a ball‑park figure within days, especially when my accountant is swamped. This helped me plan how much to set aside before the official numbers arrive.