Just instruct it not to do these things

wgd 14 hours ago | parent | next [-]

Didn't read the linked post award.

The issue covers at least two reasons this doesn't work:

1. It literally doesn't work, Claude rapidly drifts back to this style even when instructed not to.

2. Writing style constraints push the model out of its training distribution and it's very unclear how much of an impact this has on work quality.

	▲	ArtRichards 14 hours ago \| parent [-]
		I always think of it like: let it activate whatever experts to reason/do the thing, then in the next prompt give me your explainer experts. Then, both are in the conversation history.

jadar 13 hours ago | parent | prev | next [-]

That works for a couple turns but the model reverts after just a couple turns. Not to mention after compaction.

That doesn't work.