Yes, hallucination is the now standard term for this, but it’s a complete misnomer. A hallucination is when something that does not actually exist is perceived as if it were real. LLMs do not perceive, and therefor can’t hallucinate. I know, the word is stuck now and fighting against it is like trying to bail out the tide, but it really annoys me and I refuse to use it. The phenomenon would better be described as a confabulation.
For me, I make things like key and wallet discipline a muscle memory. I literally practiced putting my keys where they belonged like it was some sort of challenging skill to learn. As a consequence when I put my keys down without thinking as normal, they always end up where they belong.
I have something similar. I practice doing certain routine micro-habits until they become ingrained in muscle memory and always do them.
For example, I still set my keys down without thinking most times they are in my hand, but thanks to spending several hours practicing the motion years ago, I now always unthinkingly set them where they belong: clipped to my beltloop and tucked into my pocket. Anytime I identify a need to add one of these to my life I spend an hour practicing experiencing the trigger and then doing the motion. To learn the keys-in-pocket habit, I held my keys, clipped and tucked. Pull them out, note the feel of them in my hand, and repeat, over and over. It feels silly to practice doing something so easy, but once it becomes muscle memory, it doesn’t rely on my faulty thinking memory. I’ll do several sessions of practice every few days until I can feel that it’s fully ‘set’ as an unthinking motion. They’re a pain to establish, but they are well worth it and have saved me a ton of grief over the years.
One of these automatic habits saved me this morning. I always pat my keys when closing a locking door behind me (even if it isn’t locked), and this morning I had missed swapping my keys to my new pair of pants. I would have been locked out of my house and late for work if patting my empty pockets hadn’t alerted me just before a pulled the locked door close behind me. I have some other ones that I haven’t mentioned, because I can’t think of what they are. I’d notice the problems they prevent coming back if I stopped doing them, so I can only assume they must still be working.
Thanks, that is a better word there.
I can see the argument that it has a sort of world model, but one that is purely word relationships is a very shallow sort of model. When I am asked what happens when a glass is dropped onto concrete, I don’t just think about what I’ve heard about those words and come up with a correlation, I can also think about my experiences with those materials and with falling things and reach a conclusion about how they will interact. That’s the kind of world model it’s missing. Material properties and interactions are well enough written about that it ~~simulates ~~ emulates doing this, but if you add a few details it can really throw it off. I asked Bing Copilot “What happens if you drop a glass of water on concrete?” and it went into excruciating detail about how the water will splash, mentions how it can absorb into it or affect uncured concrete, and now completely fails to notice that the glass itself will strike the concrete, instead describing the chemistry of how using “glass (such as from the glass of water)” as aggregate could affect the curing process. Having a purely statistical/linguistic world model leaves some pretty big holes in its “reasoning” process.
Professors don’t always teach in their actual area of expertise. I had a German language professor whose PhD was in Philosophy and activity published in that field, in English, German and French journals. It does seem like an odd combination, but probably not a lot of students signing up for a class in usability of buttons, even from the fields you would expect to study them .