undefined | Better HN

0 pointsNuzzerino23d 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

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.

jdjdndnzn22d ago

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

kruffalon22d ago

Could you explain why that is?

koolala22d ago

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

1 more reply

j / k navigate · click thread line to collapse