"A line of text that is true has – for the LLM – absolutely no different quality than one that is false. There is no link to reality, no sensor data or anchoring, there’s just the data one was trained on (that also doesn’t necessarily have any connection to reality). If using the term hallucination is useful to describe LLM output it is to illustrate the quality of all output. Everything an LLM generates is a hallucination, some just might accidentally be true." - @tante

Smashing Frames

It's all hallucinations

The discourse on “AI” systems, chat bots, “assistants” and “research helpers” is defined by a lot of future pro...