Do they have a term for the accretion of constraints that directs the exploration? I've been thinking about this as I pair with LLMs, but I can't find good vocabulary. Essentially, a term for "working a problem out"
Discussion
Essentially, if a problem is solved—or in this case, if a scientific discovery is made—the artifact becomes a fact through debates. This process is called stabilization, where arguments settle and the discovery is accepted as "fact."
I like "stabilization". I've also been exerting to incremental constructions as "scaffolding", from ethics. Both of these seem to describe parts of what's going on when a model "talks it out".
I think scaffolding is a better term for LLMs because stabilization happens over time and eventually becomes a black box—something fixed and unquestioned. Scaffolding, on the other hand, is a step-by-step process where each part builds on the last. LLMs don’t produce fixed truths; they refine responses based on input, making scaffolding a better way to describe how they "think" in real time.