The dumbest AI, on purpose
The dumbest AI you can talk to right now is one that is bad on purpose: Pixiel, a chat made by the Norwegian studio Pixygon that answers with total confidence and almost no accuracy. It is not a broken assistant that lets you down when you needed a real answer. It is a character built to be daft, so the wrong answer is the whole point rather than a disappointment.
What people mean when they search for the dumbest AI
Two different people type this into Google. One has just been let down by a serious assistant, wants to know which one is worst, and is mostly venting. The other is bored and wants something silly to poke at for ten minutes. The second group is the one that ends up staying, because a deliberately bad AI is genuinely entertaining in a way a merely mediocre one is not. A tool that fails while trying its best is frustrating. A tool that fails on purpose, with total conviction, is a comedy act.
Dumb by design, not by accident
Most chatbots are graded on being right. Pixiel is graded on being funny. It will answer your question, it will sound completely sure of itself, and the answer will usually be nonsense - but nonsense with internal logic, which is what makes it worth reading. That distinction matters. An assistant that hallucinates while claiming to help you is a problem. A character that is openly playing the fool is a joke you are in on. You always know which one you are talking to here.
What a conversation actually looks like
Ask it how plants grow and you may be told that plants are slow green animals biding their time. Ask it to boil an egg and you may get an oven, an implausible temperature and an apology to the egg. The pattern is consistent: it takes your question seriously, commits hard to a wrong premise, and follows that premise all the way to the end without blinking. That commitment is what separates it from a bot that simply produces vague or empty text.
Why anyone would want a worse AI
Serious AI tools have converged on the same careful, slightly corporate register. They hedge, they add disclaimers, they tell you what they cannot do. That is correct behaviour for a work tool and exhausting as a thing to talk to for fun. A deliberately dumb AI is a break from that. It is also better to share: nobody forwards a competent answer to a group chat, but a confidently absurd one travels. Pixiel keeps a public board of the funniest exchanges people have had for exactly that reason.
How to try it
It runs in a browser, so there is nothing to install and nothing to set up. Open the chat, ask it something you already know the answer to - that is where it is funniest, because you can see exactly how far off it lands - and keep going. Good opening prompts are ordinary questions rather than clever ones. Simple factual things, everyday advice, or anything you would normally ask a search engine. The gap between how easy the question is and how wrong the answer gets is where the humour lives.
What it is not for
This is worth saying plainly. Do not use it for homework, medical questions, legal questions, financial decisions or anything else where being wrong costs you something. It is not a research tool and it does not pretend to be one. If you need a correct answer, use a tool built to give you correct answers. Pixiel is for the times you want to laugh at a machine rather than rely on one.
Frequently asked questions
If you mean the least capable serious assistant, the answer changes every few months and nobody agrees. If you mean an AI that is deliberately dumb and fun to talk to, Pixiel is built for exactly that: it answers confidently and it is almost always wrong, on purpose.
Pixiel runs in a browser with nothing to install. It is designed to be daft rather than accidentally bad, so the experience is consistent instead of just occasionally disappointing.
Yes. You can open the chat and start typing. There is no setup, no configuration and no need to learn how to prompt it properly, because there is no correct way to prompt something that is trying to be wrong.
No. A badly built assistant fails while trying to help, which is annoying. A deliberately dumb one commits to being wrong as a bit, which is the entertainment. The difference is whether the failure is the bug or the feature.
Yes. Individual answers and whole conversations can be shared with a link, and the funniest ones from the community show up on a public board.