undefined | Better HN

0 pointsmhmmmmmm1y ago0 comments

They also recommend to use it to iterate on your own prompts when using Claude Code for example

0 comments

3 comments · 1 top-level

otabdeveloper41y ago· 2 in thread

By "rigorous" I mean peeking under the curtain and actually quantifying the interactions between different system prompts and model weights.

"Chain of thought" and "reasoning" is marketing bullshit.

How would you quantify it? The LM is still a black box, we don't know what most of those weights actually do.

Yes, that was my question.

j / k navigate · click thread line to collapse