Jean Louis <bugs@gnu.support> writes: > When you say "LLMs need to access some ground truth to innovate..." > you are talking like those are some kind of living entities, while in > reality it is just bunch of vectors in a place, doing nothing without > software program.
Agreed. And so is the rest of the universe, from a cell, to a forest, to a human, like me or you. At the end of the day, it is all "dumb" local processes following the laws of nature, and so mathematics. > How could any LLM even know what is the ground truth? Programmer must > somehow give input and define those vector types of ground truth. How? It is all just numbers, as you said. > So far... we all want that change to come. Even if not the LLM, be it > some other form of the computing, I would like to see truly new > innovations. Innovation is computation too. Maybe what we can compute today does not meet your definition of "true innovation" and that is *all right*. The sparks do not convince you just yet. Me, I am more optimistic. We are getting there. There is no reason why a neural network could not compute a new theorem, algorithm, or poem. Or, why it could not approximate large functions, such as that of a human brain. We know it can. And now, that is exactly what we are starting to see, albeit in baby steps. Rudy -- "The introduction of suitable abstractions is our only mental aid to organize and master complexity." --- Edsger Wybe Dijkstra, 1930-2002 Rudolf Adamkovič <rud...@adamkovic.org> [he/him] http://adamkovic.org --- via emacs-tangents mailing list (https://lists.gnu.org/mailman/listinfo/emacs-tangents)
