Post

I’ll go ahead and make a prediction here. Crucify me if I’m wrong. LLMs don’t help us talk to animals.

ghost malone ergo propter bones@proptermalone.bsky.social

the cynic in me thinks that a lot of this is that we're preemptively maneuvering around the fact that pretty soon a bunch of things-- LLMs and probably some animals we'll be able to understand because of LLMs-- are going to ask us for moral standing, in their own words, and boy that's a hard one

raphus.bsky.social

@raphus.bsky.social

Assuming whales are talking, and that they are (at least sometimes) talking about concepts human language can express, we may at some point be able to deform the representation space of a human-language model to align it with a whale-language model. arxiv.org/abs/2405.07987

arxiv.orgThe Platonic Representation HypothesisWe argue that representations in AI models, particularly deep networks, are converging. First, we survey many examples of convergence in the literature: over time and across multiple domains, the ways...

6/10/26, 8:58 am0 reposts0 likes

No replies yet.