Hacker News Reader

Comment by at_a_remove | Hacker News Reader

It is interesting to see how much of human parsing of language depends on prior probability, probably expected pairs of concepts, because I had double-take on that sentence, since my brain really was not expecting to find combinations of KFC, Malbolge, or General Hospital in a single sentence.

View on HN

Seriously thought the whole thing was generated by a connector between GPT-3 and a new deepfake engine. Still think that's what it is. This is mind breaking.

Ah, you beat me to writing a comment that mentioned GPT-3, but I promise I didn't see yours when I started writing mine. Perhaps human thoughts are more predictable than we like to believe.

Anyway, to make your suggestion of "a new deepfake engine" more concrete, let me include a reference to DALL-E, the recently announced text-to-image engine from OpenAI:

https://venturebeat.com/2021/01/16/openais-text-to-image-eng...

Thank you, that's deeply horrifying! :)

That was my first thought as well until I saw the tweet, though the possibility of it being an acid flashback was a close second.

I had a similar experiencing when reading it, and now I'm wondering if there is a way to use a language model like GPT-3 to measure how unlikely a given string of words is, relative to other headlines.

I don't think you need something that heavyweight to do this (although maybe for an entire headline). Amazon has had the "Statistically Improbable Phrases" feature for books for quite a while now.

I was not aware that General Hospital was the name of a TV show, so not only the headline but the article itself made little sense to me initially - it read like randomly generated spam.

Reply on news.ycombinator.com