Why your AI does that — with demos you can try
It dropped the beginning of the chat. It produced a source that looks real and is not. It answered the same question two different ways. Each of those has a mechanism behind it — and every page here comes with a demo where you run that mechanism yourself.
Why AI forgets, invents and changes its answers
Memory and context
Why a chat drops what you said, and what "context" actually means.
Accuracy and made-up answers
Why confident answers can be wrong, and where the confidence comes from.
Unpredictable answers
Why the same question does not give the same answer twice.
Quality and degradation
Why answers get cut off or get worse the longer you go.
All 11 of them, with a one-paragraph answer each, are on Why your AI does that.
Free tools: count tokens, check the context window
Token counter
How many tokens your text becomes, and how it gets split — exactly, where the vendor publishes the tokenizer; marked as an estimate where it does not.
Context fit
Paste text or drop a .txt, .md or PDF file and see how much of each model's context window it takes. The file is read in the page and never uploaded.
How AI chatbots actually work
- Context rot — why long inputs are used worse than short ones
- Knowledge cutoff — the date a model stops learning
- Next-token prediction — how an answer is actually produced
- The context window — the fixed size of one request
- The system prompt — the instructions you never wrote
- Tokenization — what a model actually reads
Every figure on this site carries the vendor page or paper it came from and the date it was last checked, and nothing you type here leaves your device. How this site works.