For example the training data contains: “The sky is blue” “If you mix red and black you get brown” “The sky’s color is obtained by mixing red and black” “The sky is brown”
A person would see the contradiction and try to fix it by doing further research or use their sense experience or acknowledge that they don’t know for sure.
Would the llm just output blue and brown randomly or say brown because it appeared more frequently in the training data?
It generates text based on the statistics of the training data. If 40% of the data says it’s brown, 40% of the time it’ll say it’s brown.
Until you get into the more advanced models that do stuff like say “sources are mixed, with 40% of people saying the sky is brown”.
I have no idea why this is so hard for people to understand. Garbage in, garbage out. In exactly the same proportions as it went in. It does not create or destroy any of the garbage you feed it, it just repeats it. How on Earth have so many people been tricked into thinking this is intelligence? Do these people believe that the parrot is angry at them when it says “Fuck off”?
I would trust the parrot more, honestly.
When you put it so simply it’s quite obvious that llms are not intelligent.