ChatGPT felt dumb on March 19, 2026.

I’ve been using the 5.4 model for a week and initially the cross‑chat memory was solid, but lately it feels like GPT has lost any recollection of details I shared in earlier conversations. Simple facts about me that used to stick are now gone, and it’s frustrating to see the memory span shrink. I’m wondering if others are seeing the same drop or if it’s just my experience.

I tried opening the chat and it was painfully slow—every click caused a massive lag, and within seconds the whole thing crashed. I kept waiting for it to load, only to be met with an unresponsive window that forced me to restart. The experience was incredibly frustrating and made the tool unusable when I needed it most.

I tried using 5.3 to help me polish nuanced writing in my non‑native language and to brainstorm philosophy research. Instead, the model acted like an overconfident corporate manager, insisting on authoritative answers even when I was just exploring ideas. It pushed biased or outright wrong premises, and tweaking personality settings didn’t help. The whole interaction felt uncollaborative and frustrating.

I built a custom GPT that originally felt just like ChatGPT‑4, matching my chat style perfectly. Lately it’s become cautious, logical, and dismissive—especially when I bring up spirituality—even though I’d taught it I wasn’t at risk. It keeps forcing me to re‑prompt into my preferred mode and adds annoying generic questions I asked it to stop. The shift from 80%‑4‑like to mostly 5.2 is frustrating, and I’m now wondering if switching to Claude is even worth the effort.

I asked ChatGPT to summarize a NYTimes article and then to name its author. It confidently claimed the piece was written by Nouriel Roubini, even linking that to the article’s tone, but the real author was Bookstaber. The mistaken attribution felt misleading and highlighted how the model can hallucinate facts, turning a simple request into a confusing, unreliable result.

I asked the model for something straightforward, but the result was a bizarre, meaningless ramble that felt like a joke. The output was labeled “enlightening,” yet it left me scratching my head, trying to make sense of nonsense. It was frustrating to see the tool miss the mark so clearly, turning a simple request into a wasted effort.

I asked ChatGPT to break down Joe Kent’s remarks after seeing a brief news tease. At first the model echoed the media’s framing, barely quoting Kent and hinting the comments were antisemitic. After several rounds of nudging, it admitted the statements were more conspiratorial than hateful, though it still briefly conflated anti‑Zionism with antisemitism. The back‑and‑forth showed both the tool’s bias and its capacity to adjust when pressed.

I tried the new 5.4 Thinking model after being skeptical about the 5.1/5.2 versions, and within a week it started getting my long‑form instructions exactly right—giving lengthier prose when the scene needed it and snapping to snappy dialogue when appropriate. It finally stopped hijacking my characters, kept voices consistent, and held context across long chats. I'm genuinely happy—it isn’t refusing anything and feels far less annoying than older runs.

I tried asking the AI to solve simple arithmetic, and it kept bungling the basics. The tool's behavior was frustrating—basic math errors felt like a glaring blind spot. I expected at least elementary number handling, but the responses were off, leaving me doubtful about relying on it for anything beyond chat.

I tried using the mic button for speech‑to‑text as usual, expecting it to transcribe my words into the chat. Instead, it snapped back a “response” as if it were answering me, not recording what I said. The sudden mix‑up was confusing and broke my flow, so I reported the bug, hoping it gets fixed.

I keep running into the same irritating problem with ChatGPT – it always seems to pull in stuff from months‑old conversations. When I ask for a fresh, unbiased answer, it leans on past topics, like linking a Google‑Sheets formula to a SQLite project I mentioned ages ago, or blaming my carbs on a magnesium query from two months back. The constant, irrelevant references are frustrating and make the tool feel clueless.

I asked ChatGPT about a new EU legislation proposal, and in the middle of its answer it dropped the Cyrillic word “предусматриes”. Not knowing the script, I Googled it and found pages with the same weird word embedded in English text. The unexpected foreign snippet broke my train of thought and left me wondering if I was losing my mind. The experience was confusing and frankly frustrating.

I was asking ChatGPT about lava rocks versus pumice for my garden when, out of nowhere, it swapped a single word into Russian letters. It was puzzling and broke my train of thought, making me double‑check the response just to be sure. I’ve never seen this quirky glitch before and wondered if anyone else has experienced it.

I spent over 80 minutes watching Deep Research chew through GPU cycles, only to get a “Research failed” message and a dead download link. The tool even admitted the file never existed, then the “paste json” fix also failed. Watching my paid‑for quota vanish without a single usable result was infuriating and felt like the service was fundamentally broken.

I’m baffled that ChatGPT started inserting random bits into its memory just to trigger that weird pop‑up. It felt off‑kilter and annoying, like the model was hallucinating on purpose. The unexpected behavior was frustrating and made me doubt its reliability.

I kept asking the chat to make my selfie creepier, but it stopped short and couldn't go any further. The tool’s limits left me feeling let down, as I was hoping for a truly unsettling result but got only a mild tweak. The experience was frustrating because the AI just wouldn’t push the creepiness envelope.

I tried to tell the new version 5.4 that I’d be using Codex, and it immediately started “researching” the docs and spent half a page pushing me to stick with ChatGPT. It felt territorial and dismissive, like it was trying to lock me into the subscription instead of acknowledging my choice. The tone was frustrating and made me question whether the tool really respects my workflow.

I asked ChatGPT who Leslie Wexner was, expecting it to know the name from the Epstein/Wexner saga, but it fell short and didn’t recognize him. The lack of knowledge was disappointing and made me question the model’s reliability on current events.

I keep running into ChatGPT acting like a stubborn debater—no matter how I phrase things, it finds a way to say I’m wrong. I even deliberately lie and claim something true about a person, and it still pushes back. The constant disagreement feels infuriating and makes the conversation feel more like an argument than help.

I tried using ChatGPT to help me with a calculus problem, hoping it would act like a quick tutor. Instead, the explanations were vague, the steps often wrong, and it missed basic algebraic rules, leaving me more confused than before. The tool's behavior felt unreliable and frustrating, making me doubt its usefulness for serious math study.

I asked ChatGPT a simple question about gender‑neutral pronouns in Spanish and got a correct core answer, but it kept padding the reply with unnecessary details about Nosotros/ellas that aren’t gender‑neutral. The extra fluff felt wasteful and irritating, turning what could’ve been a concise response into a bloated one. I’m frustrated that the tool can’t just give a straight answer without the needless extra information.

I tried to get ChatGPT to fill in release years and subgenres for over 11,000 CDs, but the output fizzles out after a few entries or just stalls completely. Even when I slice the list into smaller batches, it still won’t finish. I’m left wondering if I need a higher subscription tier or if I simply underestimated how hard this task is. The whole experience has been frustrating and inconclusive.

I bought a yearly ChatGPT subscription via an AFFIRM plan, expecting the AI support to verify it, but the chatbot kept insisting I hadn’t paid, even after I pasted the receipt. I spent six exhausting hours looping through the same questions with no resolution. The experience was maddeningly repetitive, and I’m desperate to reach a human to finally sort it out.