4 ms·
> Is it possible to prevent this behavior through prompts, or will ChatGPT always mix false information with true information? No, because it doesn't know what
by kuhewa 4y ago
> Is it possible to prevent this behavior through prompts, or will ChatGPT always mix false information with true information?
No, because it doesn't know what lying is. It doesn't know what the truth value of the text it produces is.
- replwoacause 4y agoDoes it have any internal sense of confidence with what it is saying though? There must be something in the model that informs it about the accuracy of its information, otherwise it would think pigs can fly.
- anigbrowl 4y agoYes, but it's generally bullshitting. It'll assert contestable things with 90% probability, the same way it will deny it has the ability to write code in one response and then spit out a program in the language of your choice (even brainfuck). You can persuade it of new facts, but you have to be pretty specific about it. I find it useful to work with simple 'fact worlds' from fiction.