you're missing the point. SAT multiple choice negatives for random guesses, fine, you could trivially use this sort of a strategy for assigning cost functions to a classifier and backpropagate. how do you give negative weight to a wrong answer when training a transformer?
And OpenAI has induced hallucinations in o3 with RLVR mistakes, not with a failed pre-training run. They used o4-mini as an example - similar training to o3 and similar issues.
Conversely, they have also designed a post-training system that has successfully reduced hallucinations in GPT-5.
isn't this just related to the question "how do you train a transformer"? you give it wrong examples, and use optimization algorithms to move away from that kind of completions
thats quite hard for the reasons i explained. might be solvable using q learning techniques, but those are not easy in the context of transformers iiuc
