1 paper · 1 filter
Haolang Lu, Minghui Pan, Ripeng Li +6
Long chain-of-thought (CoT) reasoning improves the performance of large language models, yet hallucinations in such settings often emerge subtly and propagate across reasoning step…