Phil Haack

Frankly, to the extent that an LLM has something consciousness-like, expect it is pretty alien. And I don't think it's self-reflective in a helpful way. We're in "What Is It Like to Be a Bat?" territory, at best.

I agree with this completely! I am a proponent of @sifu.tweety.fish's posts about, e.g., how much neural power goes into brain-body feedback loops. In my mind, it aligns with the William James notion that our emotions stem from bodily feelings and actually precede conscious experience of them.

But that doesn’t mean I think you can ask Claude why he bombed a school and get anything resembling a useful answer.

I also don’t have a good definition of what that means for me. So I allow for some possibility that there is something I don’t understand going on while token streams are being processed that might have qualia or interiority of some kind.

Interesting point. I’m sympathetic to it but personally lean toward agnosticism on it. I don’t think Claude is conscious is the same way that I am, but…

one-banana-michael.gif

I think OP is not saying stateless like HTTP—obviously there’s a context window—but pointing out that there’s no shared state between different Claude instances. They are “different people” in the anthropomorphic sense.

I’m sympathetic to the presumed goal here, but making its private right of action seems like a terrible idea. I’d hate for this to turn into something like all the CIPA lawsuits in California.

But I had so much fun explaining it to my wife!

See, this is why we’re going to be able to use it for legal work.

I’m not entering any business that requires me to sell to lawyers, and especially not one that would require seamless conversion to and from OOXML.

I had a conversation (maybe with @lu.is?) a while back about how we need an IDE for legal writing and maybe it should use something like typst.

And I shudder at the idea of using GDoc.

It is both true that:
1️⃣ lawyers have very good reason to demand high reliability (errors are caught very expensively in law; there is no linting (yet))
2️⃣ lawyers overrate their own reliability, and vastly overrate the reliability of the median lawyer

*People still quote this Mars quip, but I'm actually rather keen to see somebody really build a functional Walden Pond retreat in a chilly and waterless desert

*That wouldn't be a "retreat," more of a genuine advance

somehow Ford Motor Co. has patented dominion over God

patents.google.com/patent/DE102...

But that’s because 90% of the GOP platform is: “Do you have a profound inferiority complex? Let us help you project that on national politics instead of getting therapy.”

Reason 1,327 why I’m happy to be outside counsel.

Particularly in light of the fact that Anthropic’s position has been pretty public. Not like OpenAI’s counsel was likely to ignore the context.

Oh yes. Still pay what you want.

www.whicken.com/scorch/index...

Happy birthday!

I, too, am in shorts and dismayed by it.

I can't believe the headline isn't "OpenAI entirely on board with mass surveillance of Americans, signs up enthusiastically".

How have we not had Statham in a Predator movie yet

This reminds me of the famous Bruce Sterling comment from ~20 years ago: ”I'll believe in people settling Mars at about the
same time I see people setting the Gobi Desert.”

I’m very curious about your process here. You ramble to machine, let it ramble back, and then… edit from the transcript? Have Claude summarize?

FWIW, she has a lot of training and interest in language acquisition and literacy; LLMs as artifacts are pretty interesting viewed through that lens.

This was such a good post that I got my AI-skeptic wife interested enough to read it.

I read this as “Player Characters” and was rather confused by the image.

I submit The Mother by Brandi Carlile for the mom/daughter quadrant.

Blindsight is now required reading for AI critics.

The framers were wrong about a lot of stuff but they were definitely onto something with juries