Large language models are having their Stable Diffusion moment

Re:Large language models are having their Stable Diffusion m... by tbrownaw on Saturday March 11, 2023 @08:47PM (#35115437)

At least partly, yes. There's that "7 +/- 2" number for how many slots there are in working memory, and then what actually goes in those slots has something to do with how well you understand whatever topic you're thinking about. IIRC there have been studies on chess experts that more or less boil down to them being able to use their slots more efficiently for game states that make sense.

[ Parent | on HN ]

Re:Large language models are having their Stable Diffusion m... by mafuy on Sunday March 12, 2023 @10:10AM (#35120229)

Yes. It's fewer than 7 (more like 3), and each slot takes one 'symbol'. A symbol is anything that's in long term memory. Symbols are also recursive. It could thus be, for instance, a chess opening of 15 moves. This is why an expert player can memorize usual chess games easily, but struggles with truly random arrangements of chessmen. In the latter case, an expert does not outperform a beginner.

[ Parent | on HN ]

Re:Large language models are having their Stable Diffusion m... by yieldcrv on Sunday March 12, 2023 @12:19PM (#35122007)

It could be interesting if we can create custom neural processors to excel at particular kinds of tasks.

Like temporarily, or permanently, allocate some part of the brain for a certain configuration, beyond what experts can do and closer to what a processor can do.

[ Parent | on HN ]