3 Comments

Big, overly centralized LLMs never made any sense to me for 98% of the world's needs. When you have a legal question, which model would you want to query? The one that has targeted the Westlaw database, or one that includes every random tweet on the Internet saying so-and-so politician should be in jail? Garbage in, garbage out.

Expand full comment

This sounds like an absolute claim that "...these models can create maps of fragments and then reassemble those fragments in statistically likely ways. They can do this without knowing what words are supposed to mean or what time is. An output that simulates real thought but does not reproduce real thinking."

Yet, we know for a fact that we're dealing with black boxes here. You sound a bit too sure.

And maybe I say this because there's a growing coalition of scientists who have a contrary belief that these machines are in fact somewhat conscious, building world models from text; Hinton, Tegmark, Ilya, Bengio, etc., are in this school of thought. It's an open debate at this time, but a nice article nonetheless.

Looking forward to engaging further with you, on twitter.

Expand full comment