Dropsafe

by Alec Muffett

Stochastic Nerds

I keep reading criticisms of LLMs as mere “autocomplete engines”, excellent at predicting what people will say next, framing this as systemic weakness.

As a person with ADHD who has spent much of his life waiting for others to finish saying the perfectly obvious thing that they will not actually finish saying for the next 25+ seconds, I find this… problematic?

Cue: screams of rage that these are not the same at all.

Still true, though.

⊞
Fediverse reactions

Comments

2 responses to “Stochastic Nerds”

  1. Edu

    @alecm The autocomplete critique undersells it for text, but for code it's dead on. It predicts syntax beautifully. Whether the logic actually holds is a separate question, and that's still a human's job to check before it ships.

Leave a Reply

Your email address will not be published. Required fields are marked *

More posts