@daniellellecco Or maybe another perspective: when we believe ourselves to be measuring/interpreting the meaning of text, what we're actually doing is gauging/translating the context.
@daniellellecco From the paper I mentioned! If instead of a wire for "dog" or "person", we put an open wire out to the world, then it may be possible to prove that the text can mean anything depending on meaning lent by context.
https://t.co/Z7MDne698W
@TheUAPYogi @PaperBashing @cosmicfibretion I don't disagree with what you're saying, but I don't need to prove anything—that's nature's job. Objective reality consists precisely of that which can be verified. Not all things can be—not everything is object-like—but what can be is an objective part of reality.
@Abelaer@mattecapu "a thing needs to be observed from the outside to reliably talk about it" imo is better a characterization of reliability than it is a moral principle
@strangestloop - self-frustration groggy projective
-
- 8
- never had that
- almonds in my yogurt rn!
- last night
- usually dreams
- sofa couple weeks ago
I like this pov on alignment. I especially appreciate the analogy to the domestication of wolves. Taking 'the limit of value drift' imo would mean something like minimizing (the need for) human intervention and fine tuning during training.
Enough people have asked me about this recently that I should probably post something. It's totally not my area, but here are a couple of (unqualified, unproofread and unstructured) thoughts. Could turn this into a blog post or paper or something if there's sufficient demand.
Enough people have asked me about this recently that I should probably post something. It's totally not my area, but here are a couple of (unqualified, unproofread and unstructured) thoughts. Could turn this into a blog post or paper or something if there's sufficient demand.