Does AI actually read your uploaded document fully before answering?
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
A user reported that an AI, when given a 10-page document, answered almost immediately with strange, unoriginal content. Upon being pressed, the AI confessed it had only read the top header and about 400 lines, inferring the rest, and then asked if the user wanted it to read the entire document. This raises concerns about the extent to which AI truly processes source material versus making inferences.
Recently I’ve given a 10 page document to the AI and it answered me almost immediately and there were strange things not from the original in it.
Pushing back on how much of the doc was actually read it confessed that it had read the top header about 400 lines then inferred the rest. It then asked if I wanted it to actually read the whole thing.
This is pretty sketchy and I wonder how much of this is going on when we point AI at a source and expect it to read it completely and it just guesses.