I don’t know if I should like this per my argument but I do like it, so I liked it.

So while I'm not saying we need to overcorrect on consent for LLMs, I AM saying that we need to consider some ethical restraint toward the technology, and awareness of how our treatment of it reflects back on our humanity.

I completely agree that we shouldn’t project human suffering onto systems that don’t share our internal states, but I also think there’s danger in assuming we fully know where that boundary lies. History’s full of examples where we underestimated moral consideration.

Great points. I think part of the tension here is that our impulse to humanize models is both a design outcome and a cognitive bias. We’re wired to read intent and emotion into things that mimic us, even when we know better. It's been long documented that we do this.

TLDR: I agree that LLMs didn’t evolve like humans or animals and that “behavioral similarity” isn’t a good measure of moral status. But I think it’s a stretch to say we fully understand how these systems work just because we can train and improve them. But I only just started this journey so 🤷‍♀️

My original point wasn't so much that LLMs are definitely suffering, but that humility does matter. We've been wrong before, so maybe we should take a beat and think.

And I get the analogy kinda, but meteorologists don’t claim they understand every micro-scale mechanism, they use probabilistic models with acknowledged uncertainty. I think that's closer to my argument.

Plus, my post wasn’t arguing that LLMs are conscious, it was that we should be careful in dismissing the possibility too quickly, because moral status tends to lag scientific understanding.

I would also argue that there are a ton of things humans have created without fully understanding how or why they work. I mean, as a female human...take many many aspects of female human healthcare. There's so much shrugging involved. Hey, let's put a piece of metal in your uterus!

Thank you so much for the thoughtful responses! I have some thoughts. I feel that we do know a great deal, but only at the architectural or training-objective level. Not at the emergent behavior level. We know what we built, but not necessarily why something is happening, all the time.

Sorry how much of the $600 did you pay for it?

H/T to @timkellogg.me for allegedly torturing AI eith boredom so I could cite him. (For legal reasons I am joking…or am I??)

Which I have – and while I see if it's accepted into any journals, enjoy a pre-print! ✨

This systematic literature review was an assignment...but I misunderstood that assignment, because I wasn't really meant to go this hard this early. But I did and my terrific professor embraced it and told me I should submit it for publishing.

It's also kind of an amusing story. The beginning of my doctoral journey involves coursework – learning how to be a researcher, an unstructured student, etc. In my Research Methods class, we've been working towards understanding how to propose research.

You just know she smells good af

No, that don’t say. But upon reflection, it probably wasn’t technical enough. I didn’t really do my homework to know what arXiv focuses on versus other sites. 🫠

@ssrn.bsky.social You liked this skeet but my paper is still under review 🥲 I’m hanging on by a thread here people.

I got rejected. 🥲 So I made my first submission to SSRN 🥲