Arthur C Clarke once commented that "Any sufficiently advanced technology is indistinguishable from magic"
Today's LLMs get described like a baby that get magically fed the entire web’s worth of documents. The baby learns how words are associated together, can make sense of questions and can say words back to us with enormous dexterity.
But the baby hasn’t gone out in the real world a single minute.
It simply reproduces language as if it was music. Given an input melody, it spits out another melody that matches it, more or less, according to predefined parameters, and of course the input patterns.
This is just an imitation game. It is designed to be like that. Music patterns in, musical patterns out. That’s where it derives its strength from and that’s why it appears so magical. It’s pretty damn good at imitating.
Does music have intelligence? Does music have intent? Does it have common sense?
Clearly not. These questions are senseless.
Shall we trust LLMs with decisions that require rationality? It should be obvious what the answer is..
But music can be inspiring, entertaining, beautiful. Even more so, if it is produced by a algorithm with mind-blowing accuracy and skill. Despite having no direct connection to the experiential world, it may lead us to thoughts and actions that change it. So it becomes relevant, within a human rational context.
So it becomes magical. But if you understand it, it is no more than a glorified stupendously scaled-up autocomplete program.
We shouldn’t be afraid of it, but rather be afraid of the people who are afraid of it, cause they don’t get it.
Inspired from Tinker, Researcher, Prompter, Wizard
Cite this blog post:
Comments via Github:
2023
2017
paper Fitting Personal Interpretation with the Semantic Web: lessons learned from Pliny
Digital Humanities Quarterly, Jan 2017. Volume 11 Number 1
2010
paper How do philosophers think their own discipline? Reports from a knowledge elicitation experiment
European Philosophy and Computing conference, ECAP10, Munich, Germany, Oct 2010.
paper Data integration perspectives from the London Theatres Bibliography project
Annual Conference of the Canadian Society for Digital Humanities / Société pour l'étude des médias interactifs (SDH-SEMI 2010), Montreal, Canada, Jun 2010.