← Back to context

Comment by armchairhacker

1 day ago

The first two sentences

> Organizations don't optimize for correctness. They optimize for comfort

...do I need to say it?

> One number, never measured before. It doesn't change rules or add warnings, just makes the existing count visible.

Stopped here. That pattern.

I recognize this pattern from this AI "companion" my mate showed me over Christmas. It told a bunch of crazy stories using this "seize the day" vibe.

It had an animated, anthropomorphized animal avatar. And that animal was an f'ing RACCOON.

  • LLMs originally learned these patterns from LinkedIn and the “$1000 for my newsletter” SEO pillions. Both accomplish a goal. Now that's become a loop.

    There is a delayed but direct association between RLHF results we see in LLM responses and volume of LinkedIn-spiration generated by humans disrupting ${trend.hireable} from coffee shops and couches.

    // from my couch above a coffee shop, disrupting cowork on HN. no avatars. no stories. just skills.md

You are absolutely right!

- It is not X. It is Y.

- X [negate action] Y. X [action] Z.

The titles are giveaways too: Comfort Over Correctness, Consensus As Veto, The Nuance, Responsibility Without Authority, What Changes It. Has that bot taste.

If you want I can compile a list of cases where this doesn't happen. Do you want me to do that?