Chris

I regret to inform you that St. Andrews is stunning. Missing the fam but I’m also leaving tomorrow, strange combination of feelings.

On my way to Scotland tomorrow for a conference, and all I can think about is that I don’t want to cross an ocean without my family. First work trip abroad since before I had kids.

Add ‘for tax purposes’ to a philosophical title or assertion. Everyone is the other, and no one is himself for tax purposes.

Timeline cleanse for my numerate friends who enjoy BBC radio shows, otters, estimation, and small aircraft: youtu.be/kmlXLU-1E6Y?...

A study is indeed just a blog post formatted fancy in a PDF.

This is going to be a great book!

I will never not love a John Urschel profile. No Purple Doubt!

This is a fascinating memory framework. FWIW, I’ve built agents with and without “personalities” — persistent reflection on & expression of an ersatz self, or none. As with people & genuine selfhood, less is definitely more.

I think “colleague” might be the one suitcase word for a human relationship that could be adapted to fit AI models without horrible implications. And then he tells the AI model that it’s his favorite coworker. 🙃

Such a revealing demo. The couple of moments where she rolls her eyes at her assistant or smirks. We rely on other humans as objects quite a bit, not taking them to be real people. I'm excited by the capabilities in the demo & also worried about deepening our objectification of helpers.

I just came back to this paper, and precarity is such a good, fascinating concept. It seems intuitively true that care is needed where things are precarious. And yet the angels-on-pinheads debates about AI welfare are a million miles away from this basic reality of (precarious) life.

A revised version of this preprint is now posted on arXiv. The theory section is rewritten, improving model exposition based on feedback we received, we fixed some confusion and errors, contextualized the models more carefully so the interpretation becomes clearer, and improved the overall writing.

This criticism is 100% right and is at the core of my own research. Philosophy of trust has taught me a ton, and it's a very narrow conception of trust that leaves out almost all of normal human trust -- in other humans and in artifacts.

Such an analytic-coded statement. ;-)

One of my favorite bsky posts of all time

Miranda Fricker on blame is really good, especially her reading of Strawson’s “Freedom and Resentment” as metaphilosophy rather than a kind of resigned naturalism about our propensity to blame. #philsky

In no particular order. my9games.net/en

I played this game for hours and hours as a kid and haven't thought about it for 35 years. This is awesome.

I always strive to be one or the other at all times.

I am currently fighting with Claude about whether I'm allowed to say something was "75 years later" in my own speaking notes (not slides) when in fact it was 73 years from 1950 to 2023. How is your day going?

I've been giving a lot of pep talks recently to people on my teams and in my industry about how we'll grow as humans despite *gestures around* all of this. Didn't realize until I read Catherine's speech to our Staffs freshers how badly I needed a pep talk myself!

Yes, a government office is going to develop trust out of whole cloth. This is why I have a dissertation topic, I should be thankful.

On the one hand, I’m pushing forward the frontiers of knowledge as a PhD student. On the other hand, I can’t fully understand Heidegger lectures from 1930 and Kant’s first Critique from the 18th century. (And neither can anyone else.) Nice racket we’ve got going here.

This is a great, thoughtful essay. But notice that aligning human behavior with the problems exhibited by models is not a core part of this (or basically any) AI safety framework. *User* alignment has to be part of the conversation, not just control of models & curators.

This feels like divination that confirms everyone’s priors. If you hate AI, it proves everyone is writing AI slop and lying. If you’re a booster, you write with AI brazenly and laugh.

My latest books cartoon for @theguardian.com

My teen is reading Jhumpa Lahiri’s The Namesake for school. On the one hand, one of my favorites. On the other, it’s now a “classic” that precocious middle schoolers get assigned instead of the new novel that I, as a discerning, plugged-in reader, once discovered.

OK, the Apple livestream is going a little off the rails.

Opus gives such main character energy (pejorative). Claude: I’m going to stop here. We’ve done a lot this afternoon. Me: Dude, this is the first time I’ve talked to you in two weeks.

We need to make humans safe for AI, rather than the other way around. ("Trustworthy" AI is a fool's errand as framed by most.) I recognize the anti crowd has a point and I may be defeatist, or overly optimistic about AI's benefits outweighing the downsides. Don't think we can stop the train though.

Just received word that I am officially a PhD Digital Design and Technologies student instead of PhD Humanities. Higher ed is a mess everywhere these days, and I am so grateful to have people looking out for me and giving me the chance to write my dissertation.

They could have just asked “how do you pronounce the last vowel in Missouri?” mydialect.us/quiz-standar...

Arendt on AI (via Margaret Canovan). Humans hold two contradictory beliefs: we can do anything and we’re animals that are completely determined by material circumstances. Separating knowledge from thought would make us “slaves” to machines.

RIP "trustworthy" AI

Man, I'll believe it when I see it. I use these models for hours every day. They are impressive, but always foil the breathless marketing. On the plus side, labs keep doubling down on approaches my diss says won't work, as @liedra.net confidently predicted when I started. So I still have a project!

The conversation my chapter 2 draft is currently having with its imaginary future selves.