undefined | Better HN

0 pointsblackbear_2y ago0 comments

In principle you could attach a reasoning module to the neural network and only use the LLM part of the network for input/output.

The design of such a reasoning module is an exercise left to the reader, obviously :p

0 comments

1 comments · 1 top-level

daveguy2y ago

Yeah, LLM + Reasoning Module is only as good as its weakest part. And we had more than one AI winter around the development of reasoning modules. I think you're 2x correct -- we need both and the exercise to create it is worthy of :p

j / k navigate · click thread line to collapse