@UlrikeHahn @sarahafisher @ct_bergstrom
The other advantage of the term "bullshit" over "hallucination" is that it addresses the issue of how we (the general public) are mis-using these #LLMs. If they are providing bullshit (good-sounding but unrelated to truth) language, then they are not appropriate to use for news reports, essays, or search engines.
And, also, hallucination has a very specific meaning in #neuroscience and #psychology (based on, once again, process), which is nothing at all like what these #LLMs are doing.
For anyone interesting in understanding these issues of process, I would point you to my 2013 book __The Mind within the Brain__ (Oxford Univ Press), which details the intersection of neuroscience and psychology and how what we know about process explains a lot of these issues.
Basic process definitions aligned to neuroscience and psychology:
hallucination = activation of sensory system neurons without external incoming signals and without top-down "imagination" markers. (A good example is migraine aura, which is activation of the neurons in the visual cortex by migraine processes.)
imagination = activation of sensory system neurons with some (currently unknown, but almost certainly frontal cortex) activation indicating that it is "simulation". (BTW, this is a good description of the "activate output neurons, work it backwards, see what shows up in input neurons" in these #connectionist models - again, it's an issue of process.)
https://www.amazon.com/Mind-within-Brain-Decisions-those/dp/0190263172