|
|
|
|
|
|
by js8
232 days ago
|
|
|
I just had a similar discussion with a coworker, he was advocating that LLMs are practically useful, but I argued they are kinda bad because nobody knows how they really work. I think it's somewhat return to pre-enlightenment situation where the expert authority was to be taken for their word, there was no way to externally verify their intuitive thought process, and I believe success of science and engineering is based on our formal understanding of the process and externalization of our thoughts. Similar in mathematics, formalization was driven by this concern, so that we wouldn't rely on potentially wrong intuition. I am now in favor of formalizing all serious human discourse (probably in some form of rich fuzzy and modal logic). I understand the concern for definition, but in communication, it's better to agree on the definition (which could be fuzzy) rather than use two random definitions and hope for their match. (I am reminded of koan about Sussman and Minsky http://www.catb.org/jargon/html/koans.html) For example, we could formally define an airplane as a machine that usually has wings, usually flies. This would be translated into a formula in fuzzy logic which would take, for a given object, our belief this object is a machine, has wings and flies, and would return how much it is an airplane under some notion of usually. I freely admit this approach wouldn't work for dadaist literary writers, but I don't want lawyers or politicians or scientists to be that. |
|
|
https://metarationality.com/sort-of-truth
Formalism isn't the right tool for a lot of semi-factual fields like journalism or law. Even in business, numbers are of course used in accounting, but much of it depends on arbitrary definitions and estimates. (Consider depreciation.)