2026-04-15 10:45 UTC
2/
From min. 16: #LLM training data "basically grabbing everything" on the internet.
"there's a lot of garbage there, right?
And there's no documentation. So if we don't know what's in the data, we are in no way positioned to actually mitigate the harms.
[...] the synthetic text that you can create using a language model recreates the #SystemsOfOppression that are in the training data and can also mislead people if it gets fluent enough.
We think that maybe there's a thinking mind behind it"
Replies (1)
-
@dderigo@hostux.social 2026-04-15 10:53
3/ #StochasticParrot [2]: make vivid "difference between how we use language and what's coming out of what I now like to call #SyntheticTextExtrudingMachines. So once that paper was out in the world [...] I was frequently asked «How do I know that you're not just a stocastic parrot?» And I after being asked this question a couple times decided that I'm not going to be in conversation with people who won't posit my humanity as a basic axiom of the discussion. This is a #dehumanizing question"