1 point
Both the paid version of OpenAi and co-pilot are able to search the web if they don’t know about something.
The biggest problem with the current models is that they aren’t very good at knowing when they don’t know something.
The o1 preview actually solves this pretty well, But your average search takes north of 10 seconds.
0 points
They never know about something though. They are just text randomisers trained to generate plausible looking text
0 points
3 points
*
The problem isn’t that the model doesn’t know when it doesn’t know. The models never know. They’re text predictors. Sometimes the predictive text happens to be right, but the text predictor doesn’t know.
