korrents

Our readingWhat is a korrent?

Prefill attacks work because pre-training deeply ingrains a reflex in models to continue sentences.

Drawn from what Neel Nanda said

What this subject means

scaling laws Whether more compute keeps buying more capability, and where the curve now bends -- pre-training, post-training, test-time.

What Neel Nanda actually said

  1. Neel Nanda Runs the mechanistic interpretability team at Google DeepMind
    And they have like a really strong prior for continuing sentences because you know if you don't continue a sentence properly you normally fail in pre-training which is a strong incentive to really deeply ingrain the reflex to continue sentences.

    Watch from 0:17:36 plays here

    ↗How Aligned Is Claude? A Live Review of the Opus 4.5 System Cardyoutube.com

    27 Dec 2025 · video · 2h 14m · spoken · machine transcript

    All 27 korrents from this recordingthis one is 6th

Added to korrents 9 Nov 2026 · How quotes work · Something wrong? Tell us

Do you hold this korrent?Do you also believe this?

I hold this too

Related korrents

Shared subject first, then similar wording. They may agree or disagree.

See this claim on the map