@laurahelmuth Excellent. “Generative AI isn’t hallucinating, it’s bullshitting.” The difference is that the models it uses are unmoored from reality. They don’t “know” anything and have no concept of truth or false, right or wrong. They do one thing: predict “what goes with what” in terms of a sample of data used to initialize (“train”) the model. However, the LLM likelihoods are not statistical correlation or probability in a rigorous mathematical sense.
🎯 Language matters.