Live data from Hacker News

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

matml.bearblog.dev

Post reply on HN

Print view View original

1–4 of 4 posts

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

#1

Post by tokenomics » Tue, Nov 04, 2025, 4:52 PM UTC

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

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

#2

Post by Marshferm » Tue, Nov 04, 2025, 4:54 PM UTC

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

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

#3

Post by tokenomics » Tue, Nov 04, 2025, 5:06 PM UTC

Marshferm wrote: post #2

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

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

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

#4

Post by Marshferm » Tue, Nov 04, 2025, 5:09 PM UTC

tokenomics wrote: post #3

Marshferm wrote: post #2

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

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

That’s not what human vision is.

Post reply on HN