Relationship between AI/LLMs and prolepsis

I agree this could be a fruitful line to explore. It seems to be a key attribute of canonical faculties that they operate "mechanically" and without opinions of their own, and often here people have commented that there may be an aspect of "pattern matching" going on with prolepsis. (I started to say pattern "recognition" but that might not be the best word in this context.)

At this point, I say no, it does not have something to teach us about preconceptions. I think it's like trying to learn about the physics of stars by observing Van Gogh's painting Starry Night. No matter how advanced or well-defined our models, they are still (for now) just models and analogies.

Maybe it does, insofar as it can teach us what a preconception is not.

As Diogénēs describes of a preconception, it is a "memory of the appearances from abroad", so being able to experience/process sensation, as I read it, is a necessary precursor to a preconception.

In an Epicurean sense, I don't think it's accurate to say that LLM's can have "preconceptions" because they are prone to error. Rather, it looks to me like they are being programmed with "opinions", some of which are true ... but they are not, themselves, standards of truth. They lack the standard of sensation, so they're at the whim of their programmers' memories.

We'll need to get to the point where an android organically dreams of sheep.

If I were on Picard's Enterprise-D, I would, personally, trust Data, but not the ship's computer, even if 99% of their knowledge-base were shared. If I'm going to risk losing my arm, I'm not going to take advice from an armless thing. Give that thing an arm to lose, and then see how it thinks. Likewise, I'll trust Data's description of the texture of kitten fur before the computer's, or the flavor of Picard's tea (probably bitter Earl Grey), over anything else for which it lacks its own sensory organs.

Quote from Titus

I just wanted to share my thoughts on a topic that has been on my mind for a long time.

Really cool thought, though! It's at least worth the thought experiment.

I understand the question from Titus and @Elkadites’ response as: currently, LLMs cannot have preconceptions if preconceptions mean an idea conceived before and independent of the question. Also, as I understand the LLMs, they function because of training that exposes them to data (everything on the internet and/or fed directly to them) plus additional training after data collection, on how to “talk” to the people asking questions. Then, the LLM matches the question to its data, based on how it is phrased, and answers in a polite colloquial manner. I’ve found sometimes it then summarizes the response it has given, once again based only on the data in its microchips.