>Q: What happens when output from LLMs start feeding back into training new LLMs?
Humans do this all the time. For example, fake news spreading and most people believing in it.
Humans do this all the time. For example, fake news spreading and most people believing in it.
If I said my calculator gives false numbers you wouldn't reply "humans are sometimes bad at math".
LLMs will only help fake news spread faster because they're dumb --- they can't tell the difference between real and fake.
They are indiscriminate multipliers of both real and fake data.