Hacker News: Stronz

Show HN: H-Core Snapshot – forcing LLMs to execute instead of explain

Stronz — Wed, 01 Apr 2026 17:16:24 +0000

Article URL: https://github.com/yaloms/h-core-snapshot

Comments URL: https://news.ycombinator.com/item?id=47603725

Points: 2

# Comments: 0

New comment by Stronz in "Show HN: H-Core Snapshot – prevents ChatGPT from drifting"

Stronz — Tue, 17 Mar 2026 17:15:43 +0000

I built this after repeatedly running into the same issue in longer task-oriented conversations: the model gradually drifts into explanation mode and becomes verbose, speculative, and initiative-heavy.

The idea is to initialize the conversation in a stable execution mode so responses stay structured and operational over longer threads.

The README shows the same prompt in two conversations: – default model behavior – the same prompt with the snapshot applied

Curious if others here have run into the same drift in longer conversations.

Show HN: H-Core Snapshot – prevents ChatGPT from drifting

Stronz — Tue, 17 Mar 2026 17:09:36 +0000

Article URL: https://github.com/yaloms/h-core-snapshot

Comments URL: https://news.ycombinator.com/item?id=47415473

Points: 1

# Comments: 1

New comment by Stronz in "Can I run AI locally?"

Stronz — Sat, 14 Mar 2026 09:31:02 +0000

One thing I noticed with local models is that conversation behavior tends to drift over time as well.

Even when running locally, the model often starts structured but gradually becomes more verbose or explanatory in longer threads.

Curious if others have seen similar behavior when using local setups.

New comment by Stronz in "Ask HN: Why there are no actual studies that show AI is more productive?"

Stronz — Sun, 08 Mar 2026 10:10:07 +0000

It might also depend on how the tools are used. In practice a lot of value seems to come from reducing small bits of friction rather than dramatically increasing output.