iiinova — where your digital self keeps evolving

Oh wait John Lewis was already dead in 2020! It was just the Trump/Lewis feud I guess. But the diversity effort stops at black people -- 4 hispanics and 2 asians.

│ race │ count │
│ varchar │ int64 │
├─────────────────┼───────┤
│ White │ 192 │
│ Black │ 37 │
│ Native American │ 10 │
│ Hispanic │ 4 │
│ Asian American │ 2 │

100

Dear god these people are clowns they made a public website for children with a link to "their American Heroes Biographies FINAL" and the "final" copy link is to a protected sharepoint site the general public can't access. freedom250.sharepoint.com/:b:/s/Freedo...

Just a screenshot of the thing described in the post.

220

Wait no Grover is still there! With a straight face they say he is the only non-consecutive term holder and Americans *really* like a non-consecutive term prez BTW the anti-Trump talking point on Cleveland needs to become "One of only two presidents to win the popular vote at least three times."

Grover Cleveland was a man of great integrity who served as both the 22nd and 24th President of the United States. He is the only president in our history to serve two non-consecutive terms, which proves the lasting trust the American people had in his leadership. Cleveland was known for his honesty and his firm belief that "a public office is a public trust." He worked hard to stop government waste and was famous for using his veto power to protect the nation's treasury. He believed in a limited government that allowed the free market and individual hard work to drive the country forward. During his time in office, he stood up against special interests and fought for a stable economy based on sound money. Cleveland’s dedication to the Constitution and the rule of law made him a respected leader across the political spectrum. He was a sturdy defender of American independence and a man who never compromised his principles for popularity. His presidency reminds us that character and honesty are the most important qualities in a leader. He remains a symbol of the rugged, principled executive.

100

Ben Schmidt @bschmidt.bsky.social · 28/04/2026

This part is interesting -- AFAICT they removed some people! I checked my list of the most hilariously undeserving people from last year and while Alex Trebek is still there, Grover Cleveland seems to have been nixed... who else has been taken off the list?

Mr. Trump appears to have left himself room to make five last-minute nominations; there are only 245 people in a list the administration distributed. The White House did not reply to questions about whom he would choose to fill extra slots.

110

Something wrong about thinking of a model trained on pre-1930 data as “an LLM from 1930”. I’m assuming they just shuffled the data so it tries to predict text smoothed over the input years?

230

Ben Schmidt @bschmidt.bsky.social · 15/03/2026

RIP Jürgen Habermas! His work was central (in hindsight, weirdly so) to my college education, so much so that I rushed to meet him when he was on campus for an honorary degree…

3314

I do think one of the more dangerous things about RLHF is that it produces models that will respectfully accept the most insane starting points -- I told it I use perl, awk, xls, raku and arc for geospatial consulting but keep getting pressure to switch, and it doesn't take much to get it to agree.

In this context, the "switch to Python" advice starts looking even more like cargo cult thinking. You'd be switching a working pipeline to get... what exactly? The ability to make maps in Python instead of Arc? But you're already happy with Arc's output, and you're not bottlenecked on visualization—your bottleneck (if any) is data processing and reporting, which your current stack handles.
The one actual risk I see: It's Raku, not Perl. You already acknowledged the Perl ecosystem is waning, and Raku is even more niche. If you hit a point where you need to debug something deep in the language or you want a library that doesn't exist, finding help gets harder. Raku's geospatial ecosystem is... sparse. That could eventually bite you.
What I'd honestly do in your position: Keep the pipeline exactly as is. Don't switch languages just because fashionable people say so. But be thoughtful about the Raku drift—not because Python is inherently better, but because "stable, widely-used Perl" has more long-term staying power than "interesting, niche Raku," especially for solo work where you need quick answers if things break.
If your Perl code is holding up fine and you're not hitting the edges of its capabilities, there's no business case here. You're solving the right problem with tools you understand. That's the opposite of technical debt.

I have been doing geospatial consulting for years on a stack of perl, awk, and some sophisticated Excel macros I built up, a little bit of arc desktop when I need to visualize. More recently I've been letting a little bit of Raku into the mix since the perl ecosystem feels like it's waning. People are saying I should switch to python with some javascript visualization, but this seems like bikeshedding to me -- the pipeline I use works great for my clients! What do you think?

210

This is the pattern that drives me crazy the most in python. Constantly we have a list of things (access roles, features, plan types, whatever) where one person might add a new element, and that will break someone else's consumer a mile away. In python have to maintain useless duplicate enums.

Claude screenshot.

TypeScript — derive a type from a value, use it to constrain another:
typescriptconst vals = { foo: 1, bar: 2, baz: 3 } as const;

const lookup: Record<keyof typeof vals, string> = { foo: "a", bar: "b", baz: "c" };
// ✅ Error if you add/remove a key from vals and forget to update lookup
Python — no equivalent; you have to maintain two separate definitions manually:
pythonfrom typing import Literal

Keys = Literal["foo", "bar", "baz"] # must be kept in sync by hand
vals: dict[Keys, int] = {"foo": 1, "bar": 2, "baz": 3}
lookup: dict[Keys, str] = {"foo": "a", "bar": "b"} # no error for missing keys either
The two gaps: Python can't derive a Literal type from a dict's keys (keyof typeof), and dict[K, V] annotations don't enforce exhaustiveness the way Record<K, V> does in TS.

210

I find it's more fundamental to have strict typing with even stricter linting rules. This means basically that any production system should be in typescript, go, or Rust -- I do think one of the reasons typescript is spiking so much as LLM dev takes off is that it's just better suited than js/py.

260

Ben Schmidt @bschmidt.bsky.social · 18/02/2026

as thrown around in gen-ed requirements; the place where that skill migrates to in managing LLMs is much more in the domain of software engineering. E.g., one of the tasks I used Cursor for yesterday was "take this pile of junky code and refactor it into a pub-sub pattern we use elsewhere"

We need to display updates on subagent calls in the main UI. Look at the pattern that we use for artifact updates at @packages/database/src/redis/artifact-pubsub.tsand use the same system for subscribing to sets of *assistantSessions* to hear changes. The system should allow any session to register listeners: we should fetch the content of conversations until someone actually expands or opens the relevant message (which will be on of the elements in our research accordion.) Each new message should be an SSE push -- it's not important to receive every streaming token update.

140

Wolfson's message got posted to PACER last night. The bits on wokeness are pretty milquetoast; I'm guessing the real problem is Wolfson's straightforward admission that even stated priorities of cutting "woke" grants wasn't followed: they were just illegally terminating properly committed funds.

Text message screenshot of person being sent article "Jews thought Trump wanted to fight antisemitism. Why did he cut all of their grants" from the forward.

Wolfson responds:

"Thanks for sending. Well, it is is the case that DOGE cut grants having nothing to do with DEI. As they said it was a government wide effort to claw back money. But note the tentendtious accusation that the administration is activing like all authoritarian (or even totalitarian!) governments to destroy the humanities. Their ultimate goal supposedly being to destroy independent thought. But the progressive version of the humanities accomplished that some time ago. Today it goes by the term wokeness and intersectionality.

111

Ben Schmidt @bschmidt.bsky.social · 16/02/2026

IANAL, but if I'm reading right the court ordered that the req for Wolfson's text messages to be secret made them be *immediately* put in the public docket -- otherwise we would have had to wait for the plaintiffs to cite them; now they should be on PACER this week.

But we need not wait months until the summary judgment motions are filed and decided.
Having asked the Court to rule on the confidentiality of these few sentences, Defendants
themselves have rendered them judicial documents subject to a meaningful presumption of access.
Moreover, as the Second Circuit has recently laid bare, documents "do not receive different
weights of presumption based on the extent to which they were relied upon in resolving a motion,"
and a court may not treat the presumption as "barely cognizable" merely because it does not rely
on them. See Giuffre , 146 F.4th at 178- 80. The sole question is whether the material is "relevant

110

Although they used chatGPT by my reading the keyword search prob mattered more. I wouldn't be shocked if it were simply the word "diverse" in the the abstract… the point of this operation seems to have been defund as much as possible, and then let the Bush appointees save a few conservative things.

"102. Defendant Fox, an ultimate decisionmaker for which grants to terminate or keep, began selecting grants for termination by sorting them based on their relation to his conception of DEI. In doing so, Fox did not critically analyze the description or purpose of each grant. Instead, he used crude filtering techniques that searched for keywords in grant descriptions that related to the grant’s application to individuals of a certain race, nationality, sex, gender, or other personal characteristic.
103. For instance, Fox searched each grant’s description for the use of key words that appeared in a “Detection List” that he created. Those key words included terms such as “LGBTQ,” “homosexual,” “tribal,” “immigrants,” “gay,” “BIPOC (Black, Indigenous, People of Color),” “native,” and so on. Terms like “white,” “Caucasian,”
and “heterosexual” did not appear in the Detection List."

Florence Illuminated: Visualizing the History of Art, Architecture, and Society
FAIN: PW-296829-24
University of Chicago (Chicago, IL 60637-5418)
Niall Atkinson (Project Director: July 2023 to present)
Anne Conyers Leader (Co Project Director: April 2024 to present)

A project to build a platform that would bring together data from five digital humanities projects focused on the cultural history of late medieval and early modern Florence.

Florence Illuminated, a free online research platform, will be a repeatable model that resolves the redundancies of digital humanities projects on similar topics to offer a streamlined way to share research and build on it. The resulting website will be an extensible platform for data integration and digital collaboration that maintains the autonomy of participating collections. This proof of concept will produce a dataset drawn from five projects on Florentine cultural history to illuminate the city’s social and material infrastructure, one of the most dynamic urban phenomena of early modern Europe. Our platform will allow collaborative exploration of the city’s cultural phenomena—not as a series of discrete well-known artworks, buildings, or elite individuals, but as a comprehensive aggregate of diverse people, groups, spaces, and objects that all contributed to the city’s growth and diversification—on a scale unprecedented in scholarship on Renaissance Florence.

241

Answer: no, AFAICT. But the defendants are required to file these documents by Tuesday, putting Assistant Chair Wolfson's opinion about the "progressive" and "woke" aspects of certain grants out of the public view.

"The confidentiality designations as to US-000061504, US-000063328, and the four-sentence excerpt ofUS-000062916 are therefore not sustained. US-000063328 must be unsealed. Defendants shall file on the public docket complete and unredacted versions of US-000061504
and US-000063328, as well as US-000062916, no later than February 17, 2026."

It is true that the Court has not yet been asked to rely on this document in deciding a motion
- aside, of course, from the very sealing motion the Government itself has filed. But while
summary judgment motions have not yet been filed-Plaintiffs motion is not due until March 6-
the Government's request that the court rule on the confidentiality of this particular document has
clearly been timed to keep Assistant Chair Wolfson's opinion about the "progressive" and "woke"
aspects of certain grants out of the public view. The fact that the Plaintiffs will almost certainly
rely on this language in their forthcoming motion simply underscores the "judicial" nature of the

141

Oh and hey there it is, look at that one of the items they were seeking to exclude as confidential (the request the judge rejected) was a text message with the "personal views" of Wolfson... have the contents actually been publically released? I haven't been following that closely.

"US-000062916 consists of a text message exchange sent from the personal device of NEH Assistant Chair for Programs Adam Wolfson. Defendants seek to maintain as confidential four sentences in that message. They contend that these four sentences reflect ""personal views"" of Wolfson that are unrelated to the formal decisionmaking at issue in this case. Dkt. No. 205 at 7-
8."

172

Ben Schmidt @bschmidt.bsky.social · 05/01/2026

Noting in this NEH grants filing from the Author's Guild that Michael McDonald wasn't the only NEH staffer collaborating: Adam Wolfson also helped triage which chat-GPT flagged grants might actually be related to the garden of heroes or America 250. www.neh.gov/staff/adam-w...

"109. Fox and Cavanaugh sorted grants in lists labeled “to cancel” or “to keep.”
110. No grant relating to DEI as broadly conceived of by Fox and Cavanaugh peared on the “to keep” list. Grants that Fox and Cavanaugh considered “wasteful” d thus slated for termination could be moved to the “to keep” list by Defendant cDonald only if they related to “America 250” or the “Garden of Heroes” initiatives sed on the views of Defendants McDonald, Fox, Cavanaugh, and NEH staff ember, Adam Wolfson.

21810

And in the end it comes out the same. I continually marvel at how much these things are totally fungible. Like I choose different things from this dropdown periodically, but they all basically have more in common w/ each other than each does with its own company's offerings a year ago.

Dropdown menu listing models to use for coding.

Opus 4.5, Sonnet 4.5, GPT-5.2, Grok Code, etc.

260

Ben Schmidt @bschmidt.bsky.social · 10/12/2025

This holiday season, give the gift that lets your loved ones combine their affection for James Scott with their irritation that those "No Farms No Food" bumper stickers try to normalize agriculture.

Bumper sticker that says "No Farms No State" above a mesopotamian cityscape and the phrase James C. Scott 2017

000

Ben Schmidt @bschmidt.bsky.social · 07/12/2025

Also I did not know that the tradition that a black man be register of the treasury, shown starting in the show, was so strong that *Woodrow Wilson* nominated a black Democrat to the post in 1913 until the Southern Democrats rebelled. That is mind-blowing.

WASHINGTON, July 26.-Unless President Wilson withdraws the name of Adam E. Patterson of Oklahoma, the negro whom he nominated two days ago to be Register of the Treasury, he will have a fight on his hands that bids fair to rival the Crum incident in Mr. Roosevelt's Administration. Senator Tillman of South Carolina, who so violently opposed Dr. Crum's confirmation as Collector of the Port of Charleston, and Senator Vardaman of Mississippi, who was elected on an anti-negro platform, already are preparing to fight, and it is known that they have many allies. The opponents of the nomination do not make any charges against Patterson, and it is said that he holds the respect of the white people of his community. He is said to have campaigned at his own expense for the Democratic ticket in the late campaign, and that is thought to have been a reason why his candidacy was indorsed by Senator Gore of Oklahoma. The opinion was expressed this afternoon that there was little likelihood that the nomination would be confirmed. For many years the office of Register has been held by a negro. But Southern Senators find it impossible to support even a Democratic negro for a place that will give him authority over white men and women.

Ben Schmidt @bschmidt.bsky.social · 04/12/2025

now we can use the 'equator' between the land pole and water pole (i.e., the points with the most land and water near them) as a first approximation to calculate the curve, and view it on a projection with those as the poles. That's faster than what I *was* doing, using latitude and longitude.

Ben Schmidt @bschmidt.bsky.social · 04/12/2025

OK, yeah moving to using the land-pole equator makes calculating the line much faster.

032

There are two other sets of great circles I don’t know about. 1. Those that maximize time over land. 2. Those that maximize time over water (distinct from the one with the longest unbroken arc over water, below) 3. Those that spent the same proportion of time over water as the earth is water.

A great circle over the southern ocean from India to Kamchatka

120

They were bad at this in ways that IMO got a little interesting: they kept making really basic logical fallacies that I haven't seen the current generations do, although that's maybe because I only use them for coding.

Claude -- So I think you're right - by continuity there must be at least one orientation where one hemisphere hits 29%, but it's not obvious (to me at least) that the opposite hemisphere would also be 29%. That seems like it requires a special geometric property of how Earth's land is actually distributed.

230

AFAICT there are infinitely many such splits, lying on a line that looks roughly like this: it's reasonably far away from the antipodal points that make the land and water hemispheres, but *not* just a great circle away from it. This notebook tries to find it: observablehq.com/d/db36886905...

A map showing a line that goes through North America, Brazil, the South Atlantic, the Indian ocean, and up to the North Pacific through the Phillipines.

241

Here's one such split, I think -- each of these hemispheres is roughly the same as the whole globe.

A hemisphere centered somewhere near capetown in the southern ocean.

A hemisphere opposite that one, centered in the North Pacific, including Norht America and much of Asia.

163

OK, so I just got really obsessed with a very arcane question. There are 'land' and 'water' hemispheres that maximize the amount of each (below). But can you divide the earth into two hemispheres that are each 29% land, 71% water, just like the earth as a whole? The LLMs said no… but I persisted.

The land and waterhemispheres. Equal-area projection. Map from wikipedia.

4307

I'll stay committed to the bit even if it doesn't work:

Junk line chart correlation purporting to show that Psychology is an outlier in low quality compared to the other social science.

140

Ten-year changes in majors by field -- when I first made this for the American Historical Association in 2018, history had the bottom slot, so slightly nice to see that it's rising.

A chart "Change in degrees, 2014-2024", with increases for Statistics and Computer science, and steep drops for most humanities and humanistic social sciences.

0131

@pengzell.bsky.social asks about the social sciences, which I haven't looked at in a while. Indexing these against 2008 rather than peak, and struck that economics has been on a downslope for 6 years now, while psychology, the worst discipline, continues to rise rapidly. bsky.app/profile/peng...

Chart titled 'Quality of Social Science Disciplines inversely correlated with change since 2008', showing that Criminology and Psychology have increased , economics is steady, and Polisci, anthro, IR, and Sociology are all down 30-40%

8258

Some more granular disciplinary data, by raw number of majors, showing the span since the 1990s. Most of the humanities are at about the same raw counts they were in the early 1990s. (And also the early 1970s, for that matter, although it's not in this chart because the data's harder to harmonize).

Line charts showing annual trends for sixteen humanities disciplines.

2130

Ben Schmidt @bschmidt.bsky.social · 22/08/2025

Despite the gutting of the National Center for Educational Statistics, the dept of Ed *did* manage to release 2024 college major counts in the usual format, so I can run it through the same code I do every year. First off, the change since peak of the largest fields -- another year of drops.

A line chart captioned "The big humanities majors were mostly still falling in 2024", showing drops since 2008 for most humanities fields between 10% (Study of the Arts) to 68% (religion) with history, english, and foreign languages all clustered around 50-55%

35513

They’re cleaning up at the Herald Square station and briefly scraped the place back into December 1979. Hard for me to imagine living in such a two-tone world.

Photo of old black and white printed placards on a black surface, advertising a town hall on January 4,1980

Photo of many different scraps of old placards on a black background, mostly political, some involving protests against the atgemtine government at the YN.

060

Ben Schmidt @bschmidt.bsky.social · 15/08/2025

I've been waiting for a cross-temporal spatial model like this a while, there's a lot of interesting visualization one can do! That said, who knows how useful these specific embeddings are. I used to tell graduate students never to use a plate carrée if they wanted to be taken seriously. And lo:

Ben Schmidt @bschmidt.bsky.social · 10/08/2025

Also I always have trouble believing those numbers because it's only the final release for every year but the current -- but I guess in 2024 at least the new preliminary releases for completions for 23-24 came *after* the final release of 22-23 data… but will we even keep getting the final versions?

110

Ben Schmidt @bschmidt.bsky.social · 05/08/2025

"We do not accept AI-generated content into our visual library" cool Getty images, I guess it's just that "demaerre" took two almost identical robots into an elevator for a photo shoot, but one has three dots on its right arm and one doesn't.

Stock image of a robot in an elevator with three dots on its right arm

060

Ben Schmidt @bschmidt.bsky.social · 05/08/2025

It's fun that the Fortune article saying historians are the #2 job affected by AI includes what's clearly an AI-generated stock image of a creepily female robot "teaching" (i.e., looking at a computer) while the real teacher gets fired.

Fortune headline: Microsoft researchers have revealed the 40 jobs most exposed to AI—and even teachers make the list

Accompanying Getty image of a robot with breasts at a computer while a woman packs a box with her belongings, captioned "Sorry, Gen Z: AI is coming for safe and secure teaching jobs, as well as grad roles."

171

Ben Schmidt @bschmidt.bsky.social · 24/07/2025

IDK if $75m is a high enough price to get out of them, but honestly I think maybe it's ok? If not Waltz, they're saying it would have been the guy who spent his time as ambassador trying to convince the Germans they should give the far-right another go.

320

Ben Schmidt @bschmidt.bsky.social · 25/06/2025

C'mon everybody why did we even make a social network exclusively for leftists with graduate degrees if we're not going to make #facultybratsummer a thing.

Cover of "Citizen and Subject" by Mahmood Mamdani (Reissued edition with a new preface Princeton UP 2018)

1281

Ben Schmidt @bschmidt.bsky.social · 20/06/2025

Yeah Polanyi is definitely the minimal "if you're serious you do have to read this" book here One of the engineers DM'd me earlier today