Sam Tobin-Hochstadt

I totally agree with that. I just don't think it implies your first post.

I don't think that's right -- in general it's bad to express negative judgments of people that are incorrect or wrong, even if the target doesn't hear about it.

Maybe I misunderstood your post with the word "language requirement" in it?

I think you are describing math grad programs (undergrad math degrees do not have language requirements) and Bret is describing undergrad history programs.

How much of the bias is "mil hist bad" vs "sure you're bringing a subaltern perspective to understanding Nepali soldiers in the British army but the students just like battles" vs just not thinking it's a cutting edge area?

Here's literally what happens if you put in Matt's prompt and then what you quoted.

If Lincoln keeps Hamlin as VP the whole of American history goes differently so I think that's the one clear competition.

Dad's side: started a freight forwarding company (which I think was just him).
Mom's side: HS football coach, printed products salesman/manager/executive.

Here's an example: bsky.app/profile/samt...

Shay for the W

Because I can't let this go, here's the Big 10 WBB All-Conference teams plotted by @herhoopstats.com
Win Shares and @barttorvik.bsky.social
PRPG, along with some notable other players.

cc @hooplawyatt.bsky.social

Isn't the Latino reversion part of the D+7/8 that we expect?

That makes sense. I wonder if there's some better way to quantify "# of shots per trip down the court" directly.

Why not more directly quantify number of shots (as # shots/# opp shots or something like that)?

The systems reply is the correct answer for the Chinese room.

That's certainly true for the model itself, but it's easy to write a simple program wrapping it that then does things autonomously (there are lots of experiments in this vein). Maybe you want to call that thing conscious but it's like 20 lines of code.

There are certainly possible null results that would be interesting, but they require a phenomenon that we think we understand but don't fully. Which is rarer than things we don't understand.

I have never used cursor. But "do something intelligent with this pdf" continues to be a hard case.

Oh and tonight "my configured brother printer isn't working, fix it". That was the whole prompt, and the result was the pages coming out of the printer.

For some of the bugs, it's just a matter of "here's a stack trace, fix it", and it works.

- a binary parser for serialized scheme boot files using the TUI library
- working patches for at least a half dozen bugs

Well, I don't know what you're asking for, but here are some things I've made in the last couple weeks:
- a fractal generator/display/animator for a few different fractals to show my kids (they were not that impressed)
- a pokemon themed typing web app (more impressed)
- a TUI library for Racket

This is a great representation of the biggest impact of Claude Code from my perspective. So many things that you wouldn't think to do before because they were only worth 10 minutes of thought can now be realized with a couple of prompts.

This is such a valuable use just of plain LLMs. There are so many idle questions I have that I now get detailed researched answers to, whereas before they would have stayed shower thoughts.

Text and language!

Yep. Bleak.

The Man Who Stole Infinity | Quanta Magazine In an 1874 paper, Georg Cantor proved that there are different sizes of infinity and changed math forever. A trove of newly unearthed letters shows that it was also an act of plagiarism.

www.quantamagazine.org/the-man-who-...
Wow, this rewrites history. I skipped forward a bit, but the parts I didn't know were fascinating.

I don't think you need to write a full model of the publication process. Just that the way it's written it makes it seem like obviously we would not want to select on significance and there's an obvious fix.

The alternative view, which is almost explicit in your post, is that publications exist to produce a literature, which should converge to an approximately true estimation of facts about the world when taken as a whole. But there's a real tension between those two views, because our time is limited.

Maybe it doesn't fit in this post, but I think it would be worth outlining why people would *want* to select on significance.

In particular, if you think publications exist to provide a stream of papers that are useful to read to learn about the world, there's an argument to select on significance.

Me too.