mastodon 4.7.3I enjoy long random walks in unconstrained parameter space
We attribute high intelligence to humans who demonstrate a lot of explicit knowledge.
This makes sense, since it requires active, adaptive cognitive processing for humans to acquire knowledge. Knowledgeable humans tend to have strong cognitive abilities across the board.
Language models, however, acquire knowledge through passive compression of information. There is no cognitive process involved.
Is it a fallacy to judge the intelligence of LLMs by the amount of knowledge they exhibit?