John Laudun

One name: Dwarkesh.

Oh, yes!

I remember early on attempting to build a transformer, with a really small data set, and watching all 8 cores of my little Mac go to 100% and each epoch taking 1 hour. 3 hours in I had to stop because I was worried the poor thing would melt. (And I still had 32 epochs left!)

Apparently one hot trend on Youtube is installing DeepSeek on a Raspberry Pi. The hotrodded GPUs added into the system remind me of garage mechanics, but it works.

That made me LOL. I've sat on those committees.

It's an awkward truth but universities are full of people who study systems (of various kinds) but who themselves rebel against be systematic or systematized.

I'm dumbfounded that, in the face of all these advances, "coding across the curriculum" isn't a thing.* It should be the case that everyone at a university should have a fundamental understanding of algorithmic thinking.

*I'm old enough to remember "writing across" being a thing.

Funny you should post this. I just finished setting up NetNewsWire (again, but after many years). I'm old enough to remember the promise of the blogosphere and RSS the first time they came around.

There's a new, and very distinguished CTO for the Library of Virginia. Is it @scottbot.bsky.social you ask? Why, yes, yes it is. Congrats!

I had never seen those. I like how they thoughtfully build a sump into the bottom, like an inverted punt.

One of the lesser known joys of using a fountain pen, using bottles that could have been ordered by perfumeries! (Perhaps a little dab of moss green behind the ears?)

I'm a folklorist in a film, or symbologist if in a Dan Brown novel, and I can instantly decode obscure characters and cryptic passages. I'm especially fast when I have absolutely no idea of the context.

Agreed, but the sticker shock is pretty intense at $152. (Sadly, this means that my university library won't even be able to afford it, times being what they are.)

Yes, but gay people getting married! Trans people undermining traditional gender roles! Killing babies! As long as the voting public is content to be distracted, they will be misled (in all senses of that word).

The endgame was predicted in 1985's "Real Genius":

I forgot to tag the article's author, @piperhutch.bsky.social. My apologies! Thank you for your dedication!

Louisiana to join conservative higher education accrediting body  • Louisiana Illuminator Louisiana will join six other university systems in the South to form a new alternative accrediting body favored by conservatives

Conservatives can't decide: abolish the U.S. Department of Education or have it authorize the takeover of public higher education by the "right kind of people."

Republicans used to be for what made the country economically strong. They appear willing to sacrifice strength for "correctness."

This reminds me of "The future is here; it's just not evenly distributed" (often attributed to William Gibson and I have no reason to doubt it). But I like how you've brought in networks as a way to think about it.

YouTube video by Actually Not David Every One Piece Character in One Massive Chart

I'm really enjoying some of the interesting approaches to network analyses by @notdavidyt.bsky.social. This one on character appearances across a long-running series is very compelling: www.youtube.com/watch?v=RIZn...

Last Train to Bremen A storytelling game of four doomed musicians and their poor decisions.

It's not a story line (narrative framework?) that I find particularly compelling, but damn do I dig the vibe this game has …

www.backerkit.com/c/projects/j...

Also, why do sheet sizes vary so much by width and length. It feels like there as kind of a standard for a long time, and then, at some point, linen manufacturers were like "We can do whatever we want" (said with a Cartman voice).

YouTube video by The Daily Show Nick Offerman Is Sounding the Alarm: America’s National Parks Are Under Attack | The Daily Show

Must-watch: This admin is gutting everything good about America—our national parks, public lands, public trust.

What we cherish most is under attack.
🎥 youtube.com/watch?v=b7Wv...

… but you now get to see what you thought you knew in a new light. So, J.L. Austen, Gregory Bateson, and Erving Goffman are part of the build up to Paul Werth and @joannagavins.bsky.social. (It has been interesting to see linguists in one subdomain not be aware of other subdomains!)

I'm exploring the idea that different kinds of discourse modes (narrative, informative, argumentative, etc.) create different kinds of text worlds. Spelunking in a new(ish) intellectual history is always fun, especially when it pulls up stuff from your domain that you already knew about … 1/2

A friend posited an experiment to me, where you train an LLM or other AI on physics or chemistry but ONLY up to a specific cutoff date. So no material beyond a specific point in time. Then see if the AI can infer what comes next. That is: can it remake the discoveries we know? (I'd guess not.)

Same. The final assignment/exam in a text analytics course was a hand-written Jupyter-style notebook in response to a novel problem. Pseudo-code was acceptable, but the documentation, which I had tried to get them to practice all semester long, had to be substantial. They hadn't practiced that part.

To (maybe) add to Chris' point, I agree with you, Ted, that I have yet to see the value for writing when writing is also thinking, but maybe our students are seeing writing assignments as engineering problems and not as opportunities to think?

But to your larger point, and drafting this in VS Code, what it would have me write versus what I am struggling to capture diverges radically at almost every turn. (Except that one: it offered the cliche "every turn" and I realized that that worked better than what I had in mind.) 3/3

When I first turned on Copilot in VS Code, I was grateful for the automation of low-level tasks but really wanted it out of the way when I was trying to do something beyond my own abilities or when I was trying to do something "new." 2/3

I am fascinated by what the various models "anticipate" you are going to write. It is almost always conventional bordering on cliche, as one would expect! 1/3

"There is too much garbage coming in at the foundation model level."

Did Elon just admit the quality of discourse on his platform is garbage?

Am I alone in my fascination, and perhaps yes fear, that we have to conduct such studies using techniques from the social science, the observation and analysis of behavior, and not those from the natural sciences? (And I say this as a humanities scholar.)