We need to give LLMs a stream of inputs and outputs that are replaced over time

2 pointsposted 3 months ago
by tokenomics
(matml.bearblog.dev)

3 Comments

Marshferm

3 months ago

You have to discard 90-97% of machine vision, which has little to do with mammalian vision

tokenomics

3 months ago

The point is less about vision, and more about needing a continious stream of inputs and outputs.

Marshferm

3 months ago

That’s not what human vision is.