Portable LLMs with llamafile

"while avoiding any assumptions" about sharing code... Aspirational, but not realistic. Ultimately, the assumptions have to be based on reality. In reality both copyright and regulatory rules matter. This is why the scope of the information the statistical model mirrors must be finite and therefore auditable. "Garbage in; garbage out" is not automagically optimized away by AI. Quite the opposite. I also don't buy that I should trust C (or C++) with a modeling generator meant to operate on infinite input. PLT (and history) shows this is a recipe for disastrous machine states.