Ask me anything
An assistant that may only reply with answers I wrote. When none of them fits, it says nothing rather than invent one.
§ A.1How it works
Question
You type anything, in your own words.
Selection
The model reads the approved question-and-answer pairs. Its output is constrained to one of them, plus a yes/no: does it really answer you? If not, nothing is shown.
Answer
The server checks the choice against the list, then shows that answer exactly as written. Nothing is generated.
Why is an answer sometimes off, or slow?
The demo runs on free models, which can queue for a few seconds and occasionally pick a less fitting answer. Even then, what you see is always one of my answers, never an invented one.
Where is this useful?
Anywhere a wrong answer is costly: a business website's assistant, support for a product, guidance through a process. The answers stay yours.