undefined | Better HN

0 pointsceldon253mo ago0 comments

> We need llms to be able to tap that not add the same functionality a layer above and MUCH less efficiently.

Agents, tool-integrated reasoning, even chain of thought (limited, for some math) can address this.

0 comments

4 comments · 1 top-level

RagnarD3mo ago· 3 in thread

You're both completely missing the point. It's important that an LLM be able to perform exact arithmetic reliably without a tool call. Of course the underlying hardware does so extremely rapidly, that's not the point.

jdjdndnzn3mo ago

The computer ALREADY does do math reliably. You are missing the point.

kruffalon3mo ago

Could you explain why that is?

koolala3mo ago

A tool call is like 100,000,000x slower isnt it?

1 more reply

j / k navigate · click thread line to collapse