#evaluation-incentives

[ follow ]
#large-language-models #ai-hallucinations #ai-alignment #hallucinations #pretraining
Artificial intelligence
fromFuturism
1 day ago

OpenAI Realizes It Made a Terrible Mistake

Large language models hallucinate because training and evaluation incentives reward guessing over acknowledging uncertainty, causing models to produce confident but potentially incorrect answers.
Artificial intelligence
fromTechCrunch
1 week ago

Are bad incentives to blame for AI hallucinations? | TechCrunch

Large language models produce confident but false statements because pretraining emphasizes next-word prediction and current evaluation incentives reward superficially correct outputs.
[ Load more ]