1 article tagged with this topic
A 103-upvote Reddit thread exposes how local open-source models consistently hallucinate completed tasks during tool calling.