As an example, I recently wanted to subscribe to the RawTools newsletter. When I went to their newsletter subscription page, I noticed that their URL looked like this: https://rawtools.us11.list-manage.com/subscribe?u=00722345fc94fb4d4b323edc3&id=4ff553ba3e copyIf you can find a URL from a Mailchimp email campaign in a format like this, you can usually use it to get its respective RSS feed. There are 3 pieces we need in order to find this list’s RSS feed, and all of them we can find in this URL: us11 - This appears to be the Mailchimp server location associated with the mailing list’s account u=00722345fc94fb4d4b323edc3 - I think this is a user identification code? Not sure. We need it, though! id=4ff553ba3e - Again, not 100% sure what this is; possibly a list id? We need it too, regardless ¯\_(ツ)_/¯ Once we’ve got those pieces, we can use them to construct our RSS feed. A Mailchimp list’s RSS feed looks like this: https://[SERVER LOCATION CODE].campaign-archive.com/feed?u=["u" CODE]&id=["id" CODE] copyThe campaign-archive and /feed parts are the important parts that need to be switched out here. So, we put all those pieces together, and end up with the following feed URL: https://us11.campaign-archive.com/feed?u=00722345fc94fb4d4b323edc3&id=4ff553ba3e copyThen, adding that into an RSS reader app gives us the last few campaign emails that were sent out from that list, as well as allows us to be notified of future emails without it cluttering up our email inbox

In 1971, psychologists Amos Tversky and Daniel Kahneman published a now-classic paper, “Belief in the law of small numbers,” reporting that people “regard a sample randomly drawn from a population as highly representative, I.e., similar to the population in all essential characteristics.” I’ve been thinking a lot about this idea lately when coming across discussions of evidence. The (false) small-numbers heuristic leads people to expect that all, or almost all, the empirical evidence in some controversy will go in the same direction. Individual pieces of evidence can be analogized to samples from a larger population of potential evidence. Presumably the entire population of evidence, if it could be seen at once, would confirm the truth or at least strongly favor the correct hypothesis. (Here I’m thinking of a simple case in which there are two models of the world, one of which is essentially false and one essentially true.) Now let’s get back to stories. True stories will contain a mix of confirming and disconfirming evidence; that’s just the way the world works, or, to put it another way, that’s the statistics of small samples. But, in a fictional story, all the evidence can go in the same direction, and that can feel right, in that it fits our false intuition. The question then arises, where does the incorrect heuristic of the law of small numbers come from? It could come from all the stories we hear!

rather than having Claude whimsically reading files which may or may not poison him.

In light of the above I opt to dexterously describe my processes in my CLAUDE.md and simply use the user prompt to provide parameters for those processes or to steer the model

I do have diverse revenue streams, as I teach English online and edit novels, but teaching and editing do not pay nearly what translation does. I will be lucky to make $15 to $20 an hour teaching and editing, when translation pays at least 2 to 3 times that. So here I am, still 15 to 20 years from retirement, and I have had to put so much energy into "reskilling" for teaching and editing, which, even if they don't pay as well, are jobs that require in-depth, professional levels of skill! I laugh when politicians think that "reskilling" is this magical thing that anyone can pick up and do. Going back to school in your 40s and 50s when you still have kids to take care of, a house to manage, finances to manage, health to manage, aging parents to worry about and take care of, etc., etc. is one of the biggest jokes on working people. Fuck, am I tired.

In short, science classes pair a description of our best knowledge at the present with a story of discovery of how we came to know what we know now, with the clear implication that this method is how we will continue to discover new things. By contrast in history this same story (we call it historiography – the history of the history) doesn’t generally attract sustained attention until graduate school. Students learn the names of rulers and thinkers and key figures but they rarely learn the names of historians. Likewise, instead of being presented with a process of historical discovery they are given a narrative of human development – it is not until advanced undergraduate courses that they begin to engage meaningfully with how we know these things. In my own experience the exceptions to this were almost invariably stories about the knowledge-making achievements of other disciplines – archaeology and linguistics, mostly – rather than narratives of historical investigation. So it is not surprising that many students at those introductory levels come away assuming that the narrative is pretty much fixed and has been known and understood effectively forever.

On average, essays written with assistance from InstructGPT showed significantly more homogenization than essays written without any assistance—or, interestingly, essays written with assistance from GPT-3 (the “base” model). Further, essays written with assistance from InstructGPT also involved more repetition of the same sequences: that is, two randomly sampled essays from the InstructGPT were more likely to contain the same 5-word sequence than were two randomly sampled essays written without LLM assistance.There are some clear limitations to this work (like the fact that the models used are now outdated, and the fact that the CoAuthor interface seems pretty different from contemporary LLM usage), but one interesting conclusion is that there’s something about fine-tuning the model specifically that leads towards homogenization

A final piece of evidence regarding the distinctiveness of LLM-generated text comes from this very recent (May 2025) paper exploring whether people can accurately distinguish human-authored vs. synthetic text. The authors recruited 9 annotators and asked them to read 30 human-written and 30 LLM-generated articles, indicating for each text whether they believed it was written by a human or LLM.6 Of these 9 annotators, five frequently used LLMs for writing tasks, while the other four had only limited experience. The authors found that the four with limited experience achieved an average accuracy of 56.7%—roughly random chance—though they nonetheless reported high confidence in their decisions. The five that frequently used LLMs achieved an average accuracy of 92.7%; moreover, the majority vote of these five achieved almost perfect performance (99.9%), which was considerably better than any of the automatic detection methods tested. Expert annotators also retained high accuracy in the face of various adversarial methods, including paraphrasing the LLM-generated text and attempting to “humanize” it (which involved explicitly instructing an LLM to avoid certain “tells”).

According to Joseph and Frances Gies, authors of Life in a Medieval Castle, Winter was the season from Michaelmas (September 29) to Christmas when wheat and rye were sown.  From the end of the Christmas holidays to Easter was the season when spring crops were sown: oats, peas, beans, barley, and vetches.  From the end of Easter week to Lammas (August 1) was summer, and from Lammas to Michaelmas was harvest, or autumn. (p. 206) This is in contrast to our modern calendar of seasons, which occur approximately as follows: Spring: March 20 Summer: June 21 Autumn: September 22 Winter: December 21

t David Autor, Lawrence Katz and Alan Krueger examined. In their 1997 article, they argued that the rapid spread of computer technology in the workplace may explain as much as 30 to 50 percent of the increase in the growth rate of demand for more-skilled workers since 1970. The three economists found that the demand for college-level workers grew more rapidly on average from 1970 to 1995 than from 1940 to 1970. This increased demand was initially met with a sufficient supply of college-educated workers. That supply slowed at the beginning of the 1980s, however, eventually causing a shortage that led to a widening of the wage gap between those with and without college degrees. An even more striking finding by the authors was that industries displaying the largest increases in skill requirements—legal services, advertising and public administration, for example—were the biggest users of computers. 6 Relative to other industries, these have exhibited greater growth in employee computer use and more capital investment in computers both per worker and as a share of total investment. In addition, these high computer-use sectors appear to have reorganized their workplaces in a manner that disproportionately employs more educated—and higher paid—workers.

in 1994 there was a zero percent chance that a Hildegard von Bingen record would be put out with a track titled "C*m Processit Factura Digiti Dei". This is because an actual human being would have been working at the distributor and handling the record.

The debate about the relationship of modern technological reproduction of images and their relationship with with reality has been going on since the invention of the photograph. An entire academic career could be dedicated to categorizing the literature that has already been written about it. In fact, a couple of courses could be spent unpacking what constitutes “better than reality.”Turns out, figuring out what’s real is not easy and Sam Altman is unqualified to comment on it in a serious way. The question itself is almost always a bad choice even in rhetoric. In an interview, the question gets rolled out to pretend the interview is taking place in a bizarro world where a technology executive might have something interesting to add to the debate. Unsurprisingly, they never do.It’s not the lack of knowledge alone that makes these conversations so tiresome. It’s not even an unwillingness to admit ignorance: it’s the lack of awareness that there’s already a conversation.

At what price would you consider the product to be so expensive that you would not consider buying it? (Too expensive) At what price would you consider the product to be priced so low that you would feel the quality couldn't be very good? (Too cheap) At what price would you consider the product starting to get expensive, so that it is not out of the question, but you would have to give some thought to buying it? (Expensive/High Side) At what price would you consider the product to be a bargain—a great buy for the money? (Cheap/Good Value) The cumulative frequencies are plotted, and PSM advocates claim interpretive qualities exist for any intersecting of the cumulative frequencies for each of the four price categories.

In Against the Grain (2017), the anthropologist James C Scott makes the case that we should consider ‘praising collapse’, given the improvements in human welfare it can lead to.

Which is why we see both heights and wages rise after the Black Death in the 14th century. Women’s heights appear to have decreased, but this is likely due to an earlier onset of menarche (the first menstrual cycle, which slows growth). Earlier menarche often occurs when girls experience better living conditions with more resources, counterintuitively resulting in lower long-term height. Hence the decreasing height of women is likely an indicator that they too were benefitting after the Black Death.

It’s a surprising trend found even in perhaps the most famous imperial golden age of all: the Roman Empire, which arose after the Late Bronze Age collapse. When Rome was sacked by the Visigoths in 410 CE, St Jerome mourned that ‘when the bright light of all the world was put out … the whole world perished in one city …’ Jerome gave voice to a view that has lasted for millennia: Rome’s fall as an epic catastrophe and setback for human progress. Yet the skeletons of its subjects paint a different picture. Before Rome’s rise, people across the Italian peninsula were growing taller, but this slowed dramatically under the empire: citizens were 8 cm shorter than they might have been if this prior growth had continued. Even during Rome’s golden age, those who lived beyond the empire were taller. There is a truth to the trope of the hulking, muscle-bound barbarian. After the fall, skeletons grow taller across continental Europe, and dental caries and bone lesions decrease.

Societal downfalls loom large in history and popular culture but, for the 99 per cent, collapse often had its upsides

n an entirely PSI-based college, students might finish in two years, maybe sooner,and learn a good deal more. Imagine what would happen if the entire educational system werePSI-based: huge numbers of people, most still in the throes of puberty, might be graduating fromcollege-an unsettling thought for many educators. Indeed, PSI represents a threat to theeducational system and its guardians

Some teachers resist active learning methods like cold-calling out of fear that it willmake students uncomfortable – even though research has shown that cold-calling notonly heightens engagement but also increases voluntary participation and comfort overtime (Dallimore, Hertenstein, & Platt, 2013)

That said, if a student is serious about developing their talent to a high enough level to build acareer around it, then competition is a reality that they must eventually face. In talentdevelopment, anyone who seriously attempts to reach any level of success in a sport, instrument,etc, knows that they have to work really hard and compete against other people (who are alsoworking really hard) for limited positions. Mathematics is no exception. There is a limitednumber of professorships available for mathematics and related disciplines, and outsideacademia, there is a limited number of positions available for jobs that involve solving hardproblems using advanced mathematics

Students usually stop taking math classes once they amass a critical number of knowledge gaps.The usual sequence of events starts with students trying to imitate procedures cookbook-style,without really understanding what’s going on, because they can’t intuitively grasp any of thenew material that they’re being taught. Soon after that, they find themselves unable to solve anyproblems that involve critical thinking or many steps.

However, many critical traits underlying expert performance have been shown to have asignificant genetic component:

By the time students reached 8th grade (AP Calculus BC), the separation was extremelyclear: some students were still absolutely knocking it out of the park while otherstudents were struggling.

Myth: Everybody can learn every level of math.Reality: Most people can learn basic math like arithmetic and some algebra – but beyond that,higher levels of math become increasingly abstract and technical, and fewer people have thecognitive resources to learn it quickly enough to make a career out of it, much less get to thatpoint relatively early in their lives.

The myth is that different people need the same amount of practice but in different forms– whereas the reality is that different people need the same form of practice but in differentamounts.

Everyyear, countless first-year college students decide to major in aerospace engineering orastrophysics or some other math-heavy subject, only to have that dream crushed when theyrealize they can’t even pass an entry-level math course like Calc II (not even with the help of atutor).

What Happens in the Absence of Accountability and Incentives| Tragedy of the Commons

The Testing Effect (Retrieval Practice) – to maximize the amount by which yourmemory is extended when solving review problems, it’s necessary to avoid looking backat reference material unless you are totally stuck and cannot remember how to proceed.

Armed with this information, if your goal is to maximize your depth within a talent domain,then the optimal rational strategy is a greedy approach:1. Augment your knowledge base with all the examples and problem-solving experiences inthe intended direction, as quickly as possible.2. Upon reaching the edge of human knowledge in that direction, and only then, switchover to creative production.Creative production is a substantially less efficient means of acquiring skills within a talentdomain, so you want to save it for the end when it’s the only way to continue moving forward.

In 1927, a psychologist named Truman Lee Kelley identified a curious problem that continues to plague educational research nearly a century later. He called it the "jangle fallacy": the tendency to assume that two things are fundamentally different simply because they have different names. It's the conceptual cousin of the "jingle fallacy," where we mistakenly believe two things are identical because they share the same label.

Marshall McLuhan, the great 20th Century Canadian Philosopher of Media, wasn't just prescient—he was prophetic. His declaration that "the medium is the message" didn't merely define 20th-century media theory; it has become the skeleton key to understanding our AI-transformed future. In McLuhan's world, form overshadowed content. In our world, AI is obliterating the boundaries between forms altogether.Welcome to the age of Liquid Content—where information flows seamlessly between mediums, transforming its shape while preserving its essence. What was once fixed—text, audio, video—now exists in a state of perpetual potential, ready to materialize in whatever form serves the moment.

There are four cards, a simple rule, and all you've got to do is to work out which cards you need to turn over to see if the rule has been broken. That's got to be easy, right? Well maybe, but the Wason Selection Task, as it is called, is one of the most oft repeated tests of logical reasoning in the world of experimental psychology. So let's see how you do.

“Quantized” means something like “we made it smaller by rounding off the numbers, which reduces accuracy a bit; it will run on smaller systems but won’t be as good.”  Mathematicians will tell you some gobbledygook about matrices and floating point precision, just say “that’s really interesting” and then act surprised and point behind them and say “look, is that a tensor?” and then run away when they turn their back to you.

Poetry is no less a mystery than anything else onearth. One or two felicitous lines can hardly stir ourvanity, since they are but gifts of Chance or of theSpirit; only the mistakes belong to us. I hope the readerwill discover something worthy of his memory in thesepages. In this world of ours beauty is quite common.

Firebrand British parliamentarian Tony Benn knew a thing or two about power. If we want to hold our leaders to account, he suggested, we should ask the following five questions: “What power do you have? How did you get it? In whose interests do you exercise it? To whom are you accountable? And how can we get rid of you?”

In the broader economy, AI will help reduce pressure on energy demand. It will enable the optimization of freight and delivery networks, reducing congestion and fuel use. It will help optimize the utilization of infrastructure, reducing pressure on construction. Predictive maintenance will reduce costly and unnecessary repairs. Optimization of design and AI-driven new materials will reduce materials use across manufacturing.

In 2007, tasked with reporting to Congress on the environmental implications of the internet, the Environmental Protection Agency (EPA) noted that US data center power use had doubled between 2000 and 2006 to around 1.5% of total power use and raised the specter of it doubling again by 2011, significantly slowing the retirement of the US’s aging fleet of coal-fired power stations. It didn’t happen. Companies began shifting operations to the Cloud, achieving over 90% efficiency gains from a combination of new servers with bigger and faster chips, economies of scale, and better infrastructure. The average Power Usage Effectiveness (PUE) – the ratio of total power used in a data center to the power used by its servers – dropped to 1.5 in 2021 from 2.7 in 2007, with the best data centers now delivering PUEs as low as 1.1. In 2011, data centers were still using less than 2% of total US power.

Perhaps efforts like the IndieWeb Carnival and 32bit Café’s events might be better termed “community syndication” instead of “community blogging”? Then again, maybe the name doesn’t really matter as long as there’s more of it for those who want it.

Female velvet spiders exhibit a remarkable type of maternal care unique among arachnids. Upon the birth of her brood, the mother spider liquefies her internal organs and regurgitates this material as food. Once her capability to liquefy her insides is exhausted, the young sense this and consume the mother.

n the wake of World War I, early management scientists associated with the “human relations” movement at Harvard Business School began to apply the language of “skills” to interpersonal encounters. As Elton Mayo, a key figure in the team, saw it, workers were blaming their employers for workplace discontent. In his view, their real antagonist was the loneliness of life in the industrialized modern city, which exacerbated the alienation they felt at work. If management cultivated community feeling in the workplace, the logic went, workers would stop demanding power. Soon, manipulating workers — and preventing collective action — began to be seen as a matter of skill. “Management should introduce in its organization an explicit skill of diagnosing human situations,” advised Fritz Roethlisberger and William J. Dickson in their 1939 volume Management and the Worker. Executives gained a way to describe their own social abilities as profitable commodities. Businessmen weren’t merely flaunting social graces or providing care — what their wives did at home — but demonstrating serious expertise.

Week after week, I would sneak glances at the bubbly radioactive glow of each soda, biding my time until I had the chance to taste. When I did, I was born anew. The abrasive carbonation and bright flavors inspired my own fizzy beverages, even as Italian sodas fell out of favor. Reveling in nostalgia and the economic liberties of adulthood, I bought a bottle a few years ago. Frankly, it was trash.Gone were the fruity notes that would linger on the last ice cubes after I finished an Italian soda in a single continuous gulp, and in their place was a dull sweetness. Torani is now a shelf-stable simple syrup with an aroma instead of a flavor. My hunch was further confirmed after trying a "dirty soda" at a Swig copycat. I love sugar too much to accept when it is used as a blunt instrument instead of a gentle buttress to structure how you experience flavor.My disdain grew as I started to make more syrups, and I took my syrup shelf worldwide. There is no limit to what you can steep or infuse into a simple, and any jam/jelly is a splash of warm water away from becoming a drink syrup, btw.

Tech employees, despite their progressive leanings, seem largely allergic to unionizing for reasons that mostly seem to boil down to "not enough people in SF/SEA/NYC are worried this will some day come for them."

In Anxiety is the Dizziness of Freedom, people invented a new Prism(™) that allows them to talk to copies in nearby universes. Chiang treats this technology as an interesting tool to explore his favorite questions of free will, determinism, and what does it mean to be a person, etc.I think he doesn’t understand the power of this singularity-level technology he just introduced.The ability to send bits across parallel universes is just insane in terms of economic and experimentation value. For example, pharmaceutical companies can do $100 billion trials for all sorts of novel drugs, and trade the results of this information with their clones in other universes. Massive R&D projects in general can be done in parallel across different universes, as long as the results can be compressed in enough bits to be sent over these prisms.More than that, you can now have not just individual or small-group level studies but societal-wide or even multiverse-wide ones! Wondering whether raising interest rates can cause recessions? Just conduct RCTs across multiverses (and if they can’t coordinate on RCTs, even the observational data would be insanely useful)!While the main characters work through existential crises and their mundane, human worries, the rest of society should be accelerating at breakneck speed toward doom, utopia, or something else equally fascinating. Chiang doesn't explore this, even as backdrop.

pairs of non-synonymous phrases where the words in one phrase are each synonyms of the words in the other

The classicist M. L. West gave this construction a name: the augmented triad. It appears to be very old, and not just because it’s found in Homer. It’s also found in Vedic poetry (that is, composed in an early stage of the Sanskrit language) as well as Celtic and Germanic verse. Some examples, courtesy of M. L. West’s book Indo-European Poetry and Myth (pp 118–119).Vedic Sanskrit:Tváṣṭā, Savitā́, [suyáma Sárasvatī]‘Tvashtr, Savitr, and [easily-guided Sarasvati]’ (Rigveda 9.81.4)Old Irish (Celtic):trí meic Nóe nair cech neirt:Sem, Cam, [Iafet aurdairc.]‘Three sons of Noah, of every (kind of) strength:Shem, Ham, [Japheth the glorious]’ (Lebor Gabála Érenn, 189–190)Old Norse (Germanic):Vara sandr né sær né [svalar unnir].‘There was not sand nor sea nor [the cool waves].’ (Vǫluspá 3)The presence of the augmented triad in not only Greek, but in Celtic, Germanic, and Vedic Sanskrit (see what I did there?) means that it was likely inherited from the common ancestor of all four.

AI can already explain Gleam to me very clearly. It doesn’t explain it exactly the way I would and it hallucinates and false starts constantly. But it is certainly good enough to get the point across. And it has the advantage of explaining things to a reader the way that reader wants to understand it. People who think my writing is too familiar or too verbose (the most common complaints in the relatively few negative reviews I’ve received) may be able to get the AI to explain it in a way that works better for them. But people who DO like my style can literally say “tell me how to use Gleam in the style of Dusty Phillips” and share links to articles this blog or some of the chapters of the LazyVim book. I tried it. It is super uncanny valley. It’s not exactly me, but I won’t say it’s not not me either.

The ILC developed a database of tens of thousands of dissidents living in the Eastern Bloc and carefully targeted the books it sent. Paris became a critical hub where travelers could load up; the Librairie Polonaise, a venerable Polish bookshop, let visitors take several volumes for free on their way home. Its manager sent a bill to the CIA, which by the mid-1980s was covering half of the store’s inventory.

Learning takes effort; if your practice is too easy it probably isn’t effective. Pedagogical methods that feel hard to students are most effective, and methods that students enjoy are least effective.

The Math Academy Way of teaching is to provide a very tight scaffold with small steps when students are first introduced to a skill and gradually remove it when practicing so the student does more on their own and must use recall. If a student is unable to come to the correct answer, [either the student is lacking in care for execution or] requisite skills are not understood. The Math Academy software knows the skills that make up each type of question and can find problem areas and target reviews of that material.

You have more ingredients available to you at the grocery store than anyone in history, but you can only cook with what’s in your house. You can’t be creative, synthesize things, make connections, without the requisite knowledge. Recall the sweet satisfaction of spotting a reference in a book; you can’t make that connection if you aren’t familiar and comfortable with the referenced text.

I thought, “I have a deep understanding of trig, and so if faced with e.g. sin(-θ), I can visualize the triangle in my head and realize that it’s equal to -sinθ. I will re-derive these identities from scratch during a test.” But it never really worked out that way. This is the trig equivalent to decoding a sentence letter by letter on the fly. It takes a lot of memory and brain power, and you can’t spot these patterns in a problem if you don’t know the identities. I always felt it was unfair when a solution relied on a double angle identity “trick.” My virtuous and pure, unmemorized knowledge didn’t seem to be sufficient during the exam

Royal jelly This is the mix of compounds that worker bees generate and feed to their queen. Queen bees live for up to 20 years, while workers (with the same genome) die after just a few weeks. There is one study (Inoue 2003) in the DrugAge database feeding royal jelly to mice, and yes, at higher dosages, 15 mice lived an average 25% longer. As you can see below, early mortality was greatly reduced, but maximum lifespan wasn’t affected.

I go downstairs to use the spin bike in the gym at Ryan’s apartment complex. The instructor on the screen looks at me blankly and says, in a monotone, presumably intending to motivate me: “No one’s going to care about you as much as you do.” She seems taken aback by her own statement. She says, “Well, I care about you.”

The suspicion that all this is élite anxiety in the face of a democratizing mediascape deepens when you consider what the attentionistas want people to focus on. Generally, it’s fine art, old books, or untrammelled nature—as if they were running a Connecticut boarding school. Above all, they demand patience, the inclination to stick with things that aren’t immediately compelling or comprehensible. Patience is indeed a virtue, but a whiff of narcissism arises when commentators extoll it in others, like a husband praising an adoring wife. It places the responsibility for communication on listeners, giving speakers license to be overlong, unclear, or self-indulgent. When someone calls for audiences to be more patient, I instinctively think, Alternatively, you could be less boring.In a sense, what attention alarmists seek is protection from a competition that they’re losing. Fair enough; the market doesn’t always deliver great results, and Hayes is right to deplore the commodification of intellectual life. But one can wonder whether ideas are less warped by the market when they are posted online to a free platform than when they are rolled into books, given bar codes, and sold in stores. It’s worth remembering that those long nineteenth-century novels we’re losing the patience to read were long for a reason: profit-seeking publishers made authors drag out their stories across multiple volumes. Market forces have been stretching, squashing, spinning, and suppressing ideas for centuries. Realistically, the choice isn’t commodified versus free but which commodity form suits best.

Monahan considers how this intersects with the AI-induced existential crisis currently felt by creatives across most culture industries, from art to Hollywood to media, concluding that “there may be one skill you truly cannot outsource to AI: knowing how to attend and host an event. Increasingly, that is a rare and coveted skill.”

Also —taste is not a reliably differentiated skill. We don’t know whether, at some point, AI advances will be able to effectively generate the semblance of idiosyncratic taste. But we can be quite sure that AI will never be able to replicate the experience of meeting friends for a 6pm happy hour cocktail and ending up on an unplanned New York City Night™ adventure that involves sneaking into a party at the Museum of Natural History

I am sometimes asked why I became a historian. A big part of it is that I just really like looking at old books.

In the 1840s, it would cost you $2.50 to buy a Dickens novel in London. In New York, that same novel would cost you six cents. This is partially why Dickens decided to tour the United States, in 1842: to urge Americans to support international copyright laws, so he would be paid for the copies of his novels Americans bought. What he discovered was that a key reason the tour was so successful—it was packed, night after night, with eager readers—and a key reason reason Dickens was so famous stateside—was because his books were so cheap and plentiful. When he returned home, frustrated at his inability to convince Americans, he wrote a book about his experiences. It was (of course) pirated and reprinted in the US. It sold 50,000 copies in three days. Dickens received $0 for those sales.

When the international copyright act was finally passed in 1891, it included a “Manufacturing Clause” that required all books by foreigners be set domestically, using American type. So American publishers had to pay royalties to foreigners, but now all books had to be made in the USA (so it would have been illegal to do what many publishers do today: send books to China or other countries where printing is cheaper, to be produced and shipped back).

But the law prohibited authors from copyrighting ideas—they had to have a book to protect. They were also not allowed to set the prices for those books higher than what would be considered reasonable.

Few today would lament that Americans were able to buy cheap versions of Emma soon after it was published, but Austen received no royalties from sales of her novel over here.

And yet despite the globalization of the industry, the steep rise in imports had failed to devastate American producers. Instead, domestic production and employment from the late 1960s to the late 1970s had risen together with imports.

The Big 3 were not avoiding all unions. In a 1950 speech about Ford’s new investments to the Buffalo Chamber of Commerce, Henry Ford II praised Buffalo labor leaders as forward-looking, and Buffalo itself as “the place where an organization can get work done — where good production cooperation is possible.”[7] It was specifically the local unions of Detroit that compelled the Big 3 to find new places to set up factories.

So globalization couldn’t kill Detroit because Detroit was already dead. Or at least dying. And it really was a story of decline in Detroit almost exclusively. From 1950 to 1980, auto employment even in most of the Rust Belt states expanded. The gains in Ohio, Indiana, and Illinois exceeded those of any southern state.

So much for Myth 1, the persistent notion that the American auto industry has collapsed. With output at all-time highs and employment hardly lower than in the time before NAFTA (the biggest globalizing event for the auto industry in recent decades), the evidence goes hard in the other direction.

I first noticed that the same harms that won high school policy debate rounds tended to be the ones that drew cult followings in late 2010s Twitter. Both environments reward rhetorical fearmongering with no care for action or implementation.

As a result, any use of ‘obsolete’ things is labeled as nostalgic because they have exhausted their time of economic value. From the perspective of constant production and consumption, using these things doesn’t make sense. From the perspective of Permacomputing, it is impossible to long for and romanticize a time when computer technology was better because such a time never existed. The moment when the economic lifecycle of a computer ends is celebrated because only then can its socioeconomic context be reclaimed.

Whereas, when you hear an idea, you can usually repeat the idea but not with the same authority.

The primary audience gets something out of this sequence of events: they get power. This is the great secret of writing in public: the writer and primary audience both put in effort (to pack and unpack the idea); and they jointly reap the rewards, which is the legitimacy earned when the idea gets subsequently retold verbally to the wider secondary audience.

One lesson hiding in plain sight here is that most of the audience of any successful post does not actually read it. They are told it by someone who did read it. There’s a primary audience who carefully reads the piece and does the cognitive work of “restructuring their consciousness” (see Walter Ong, Orality and Literacy) around good writing. And then there’s a secondary audience, who are re-told the content, either verbally (including group chats, podcasts, Youtube) or in other oral formats like Twitter.

I hate PHP too, but I'm using it more often for fully vibe coded apps because LLMs are very good at it.

Hindu weddings have the saat phere (seven circles around the fire), but what those seven vows actually say varies wildly by region, caste, and the pandit's mood. One couple discovered post-wedding that the wife had promised to "bear ten sons" in the Sanskrit version while the English translation they'd requested said "support each other as equal partners." They're still married. The wife now introduces herself as "defaulting on my contractual obligations," but since no one has called to collect, they're probably fine!

There is, I think, a compelling case to be made that the American political tradition thinks of free speech not as the basis of a free press but as an instance of it. A newspaper is not a giant person; a person is more like a little newspaper. If this is so, then the vast majority of Americans have low budgets, limited circulation, and few dedicated readers, even in the age of social media. In other words, speech is a resource, not an inalienable property, and as such it is subject to the same regimes of theft, privatization, and accumulation that have swallowed up the labor of workers and the fruits of the earth.

Hence, Tocqueville wrote in Democracy in America, “the prejudice that repels Negroes seems to grow as Negroes cease to be slaves, and inequality is engraved in mores in the same measure as it is effaced in the laws.” This is a very different story! Williams claims that a decrease in economic inequality has made Black people more resentful of those fewer privileges still denied them. But Tocqueville’s shrewder observation is that rising legal and economic equality indirectly produces social inequality as the dominant class is forced to confront the prospect of forfeiting its dominance. This is why prejudice seemed to increase in the Northern states. Both Blacks and whites, being formally equal, now perceived informal equality as a concrete possibility.

In 1994, Italian physicist Cesare Marchetti suggested that people have always endured commutes of an hour a day, half an hour each way, on average. Improvements in urban planning and transportation haven’t shortened our travel time; they’ve just permitted us to live further afield.

Reading slop text when I think I’m reading real text is exhausting: since I am not on the alert for hallucinations or irrelevancies, every turn of phrase that seems out of place causes me to wonder why that phrase it there and what am I missing when in reality, such questions are ill formed: that was just a phrase composed by accident that sounds good but actually is devoid of much intent at all.

What about applying a set of rules to multiple elements, as you do in style sheets with class rule sets, what does Tailwind offer for that? Nothing. The answer is nothing. Tailwind's official recommendation for this is to ignore everything you've ever learned about constants and duplicate your class names. Then simply remember to always use multi-cursor editing any time you touch them:

There is no secret herbal medicine that prevents all disease sitting out in the open if you just follow the right Facebook groups. There is no AI coding revolution available if you just start vibing.

When I have had engineers who were 10x as valuable as others it was primarily due to their ability to prevent unnecessary work. Talking a PM down from a task that was never feasible. Getting another engineer to not build that unnecessary microservice. Making developer experience investments that save everyone just a bit of time on every task. Documenting your work so that every future engineer can jump in faster. These things can add up over time to one engineer saving 10x the time company wide than what they took to build it.

I'd be looking at some code at my desk, and it made no sense. Why would anyone write it like this? There's an obvious and cleaner way to approach the same problem.So I'd go down the hall to the person who wrote it in the first place and start asking questions...and find out that I didn't have the whole picture, the problem was messier than it first appeared, and there were perfectly valid reasons for the code being that way. This happened again and again. Sometimes I did find a real flaw, but even then it may have only occurred with data that wasn't actually possible (because, for example, it was filtered by another part of the system). Talking face to face changed everything, because they could draw diagrams, pull out specs, and give concrete examples.I think that initial knee-jerk "I've been looking at this for ten seconds and now let me explain the critical flaws" reaction is a common one among people with engineering mindsets.

In other words, AI will not enable the creation of quality translations for people who previously lacked that ability. That part still requires a human feel for the linguistic and cultural elements of the translation. But for those who are just looking to get a rough but passable translation (say, for research) it should work most of the time. And for those who would love to create quality translations but face huge opportunity costs and zero financial incentives, AI could lead to new possibilities.The deeper risk is not that AI will replace historians or translators, but that it will convince us we never needed them in the first place. A tool that outputs polished, confident language with no sense of ambiguity or context is appealing to people who think facts will save us. But there is a vast difference between facts and truths. If we come to treat them as interchangeable, we cede interpretation to the machines and narrative power to those who design them. So maybe Microsoft was right after all, just not in the way they think. Historians and translators may be the first to go not because their work is easy to automate, but because the interpretive element of their labor has always been invisible or, when made visible, dismissed as odious human bias. AI will replace them only in the minds of people who never understood what they were doing in the first place. Which, given how things are going, may be enough.

In telecommunications, squelch is a circuit function that acts to suppress the audio (or video) output of a receiver in the absence of a strong input signal.[1] Essentially, squelch is a specialized type of noise gate designed to suppress weak signals. Squelch is used in two-way radios and VHF/UHF radio scanners to eliminate the sound of noise when the radio is not receiving a desired transmission.

You should anticipate every objection someone would have and preemptively show why it's invalid, so that the reader never even thinks to bring it up.

Monthly Packet of Evening Readings, Volume 19, (London: Mozley and Smith, 1875), page 213: —and the rule for sowing is, ‘One for the mouse, one for the crow, One to rot, and one to grow’

LiveJournal had a division of domains: There were personal blogs and community pages, and anything you posted on a community page existed within the context of that page. It wouldn’t randomly get boosted to some unrelated person’s feed, who might then feel at liberty to comment something like “what a weird hobby” or “this post is fucked up” or some other unsolicited insulting comment. There was not yet the sentiment of the “public square.” I hate that phrase: the “public square” where we all get together and talk. I don’t want to talk to most people in the square. I want to find a nice park or dive bar off a side road far away from the square where I have things in common with other people at the venue.

Starving Saints, a fantasy-horror taking place inside one besieged castle that is slowly starving to death, until something–a saint and her attendants–comes to relieve them of their hunger.

During the Cold War, U.S Army cryptologist Lambros D. Callimahos devised a “Republic of Zendia” to use in a wargame for codebreakers simulating the invasion of Cuba. (Callimahos’ maps of the Zendian province of Loreno are below; click to enlarge.) The Zendia map now hangs on the wall of the library at the National Cryptologic Museum. The “Zendian problem,” in which cryptanalysts students were asked to interpret intercepted Zendian radio messages, formed part of an advanced course that Callimahos taught to NSA cryptanalysts in the 1950s. Graduates of the course were admitted to the “Dundee Society,” named for an empty marmalade jar in which Callimahos kept his pencils.

Optician Sans is a fully functional typeface and a continuation of the historical Snellen and Sloan letters. Optically adjusted for readability to be used as a fully functional display typeface.

It’s based on my book Ologies and Isms: Word Beginnings and Endings, published by Oxford University Press in 2002.That went out of print in 2008 and I immediately made it available as a free service.

Unlike many programmers, I love code review. I find that it is one of the best parts of the process of programming. I can help people learn, and develop their skills, and learn from them, and appreciate the decisions they made, develop an impression of a fellow programmer’s style. It’s a great way to build a mutual theory of mind. Of course, it can still be really annoying; people make mistakes, often can’t see things I find obvious, and in particular when you’re reviewing a lot of code from a lot of different people, you often end up having to repeat explanations of the same mistakes. So I can see why many programmers, particularly those more introverted than I am, hate it. But, ultimately, when I review their code and work hard to provide clear and actionable feedback, people learn and grow and it’s worth that investment in inconvenience. The process of coding with an “agentic” LLM appears to be the process of carefully distilling all the worst parts of code review, and removing and discarding all of its benefits. The lazy, dumb, lying robot asshole keeps making the same mistakes over and over again, never improving, never genuinely reacting, always obsequiously pretending to take your feedback on board.

Every day I spawn in. Emerge wriggling out my skibidi bolus of slime. Whence and where? Lol. Idk. Vibes here be mad shady fr. Shit is not aesthetic. Shit is not bussin. Shit is burned-out cars piled in barricades across the street. Shit is THE END IS NIGH scrawled across bridges. Shit is roofs caved in, windows boarded, thin trees already rising out the wreckage, with roots that slip through gaps in the brickwork to return the brief work of man to the senseless rubble that came before. This sus ahh Ohio ahh realm is my crib. Damn, bitch, I live like this

No apps, no texts No stranger danger No surprises Tin Can is a telephone club for friends to chat voice-to-voice during the no-smartphone years. Only approved contacts can put calls through to your Tin Can. Easily manage who's allowed in the club via the parents-only companion app. Set up a quiet hours schedule or enable Do Not Disturb within the app.

If you've ever wondered why scientists scramble to publish in Nature, Science, or Cell, think of them as the holy trinity of scientific prestige, each with its own personality. Nature and Science were established in the late 1800s—Nature published by the Brits and Science by the American upstarts. Cell is the newcomer; established in 1974 in Cambridge, Massachusetts.In terms of temperament, Nature is the flashy cosmopolitan—broad, attention-grabbing, and often favouring "sexy science" that makes headlines (black holes, CRISPR, or ancient human fossils). Science is the serious, translationally-minded intellectual—rigorous, respected, and slightly less obsessed with media hype (CRISPR before it was famous). Then there's Cell, the molecular biology workhorse, where groundbreaking discoveries in pre-clinical work are dissected in exquisite mechanistic detail (if you love signaling pathways, this is your jam). Publication in any of these journals is the scientific equivalent of winning an Olympic medal (or at least making an Olympic team, depending on your position in the author list).One tier down, you'll find specialty journals like Neuron, Nature Neuroscience, and The Journal of Clinical Investigation (and reams of others for specific fields), which publish longer, more methodically comprehensive studies. I tend to prefer reading papers from these journals as they provide greater detail and present more fully developed work. These papers may not be as "hot off the press" or media-friendly, but they often demonstrate greater scientific rigor and better withstand the test of time.Meanwhile, in medicine, The Lancet and The New England Journal of Medicine (NEJM) tower over most others with impact factors of 98.4 and 96.2, respectively (generally, the higher the impact factor the greater the prestige). This reflects their enormous readership—there are far more medical doctors than PhDs. But at the top, it’s CA: A Cancer Journal for Clinicians, with a staggering 286.13 impact factor, a reminder of cancer’s toll and where our research priorities and funding are concentrated.

Ultimately, the valuable work done inside companies is the work where human accountability is the primary bulwark against the O-ring problem. It acknowledges that complex decisions and accepted constraints need to have human owners, and that relationships between humans (notably, but obviously not exclusively, in reporting relationships) are ultimately the valuable stuff that companies are made of. Okay, so, what does this have to do with AI agents? Well, my suspicion is as follows. One, combinations of AI and Human work could be notably vulnerable to the O-Ring problem. Their utility in executing a chain of tasks is only as valuable as the weakest link, very obviously. (If there’s one hallucinatory or badly-interfaced step anywhere that isn’t caught, you have to throw out the whole thing.) Firms have self-selected for solving O-Ring problems, over decades. That’s just what firms are.  So I think you could construct a bear thesis that the closer a workflow is to core to of an average business, the less suitable agents are to help,for reasons that are readily explained by Coase’s theory to the firm. Maybe not for the very best companies, who truly master their internal agents, but for the average company to whom AI Agents doing real work is essentially a kind of outsourcing. I’m not quite as pessimistic as that, but I agree with Tyler Cowen’s post that any hope for 10% GDP growth or whatever to emerge out of existing firms (which, even in a very disruptive environment, make up most of GDP) is too hopeful.

Production is broken down into n {\displaystyle n} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/a601995d55609f2d9f5e233e36fbe9ea26011b3b" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:1.395ex; height:1.676ex;" alt="{\displaystyle n}"/> tasks. Laborers can use a multitude of techniques of varying efficiency to carry out these tasks depending on their skill. Skill is denoted by q {\displaystyle q} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/06809d64fa7c817ffc7e323f85997f783dbdf71d" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:1.07ex; height:2.009ex;" alt="{\displaystyle q}"/>, where 0 ≤ q ≤ 1 {\displaystyle 0\leq q\leq 1} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/5c4dd7e354bddf3131e391b64a7db0be9b4ac572" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:9.591ex; height:2.509ex;" alt="{\displaystyle 0\leq q\leq 1}"/>. The concept of q {\displaystyle q} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/06809d64fa7c817ffc7e323f85997f783dbdf71d" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:1.07ex; height:2.009ex;" alt="{\displaystyle q}"/> differs depending on interpretation. It could represent the probability of a worker successfully completing a task, the quality of task completion expressed as a percentage, or the quality of task completion with the condition of a margin of error that could reduce quality.[3] Output then equals the product of the q {\displaystyle q} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/06809d64fa7c817ffc7e323f85997f783dbdf71d" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.671ex; width:1.07ex; height:2.009ex;" alt="{\displaystyle q}"/> values of each of the n {\displaystyle n} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/a601995d55609f2d9f5e233e36fbe9ea26011b3b" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:1.395ex; height:1.676ex;" alt="{\displaystyle n}"/> tasks together and scaling it by a firm specific constant, B {\displaystyle B} <img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/47136aad860d145f75f3eed3022df827cee94d7a" class="mwe-math-fallback-image-inline mw-invert skin-invert" aria-hidden="true" style="vertical-align: -0.338ex; width:1.764ex; height:2.176ex;" alt="{\displaystyle B}"/>. This scalar is positively correlated with the number of tasks. The production function is: F(qi, qj) = Bqiqj The important implication of this production function is positive assortative matching. This can be seen in a hypothetical four-person economy with two low skill workers (qL) and two high skill workers (qH). This equation dictates the productive efficiency of skill matching: qH2 + qL2 ≥ 2qHqL By this equation total product is maximized by pairing those with similar skill levels.

But ASU has ranked well in terms of preparing students for life after college, and was named a Best Value College by The Princeton Review this year. [Edit: According to this analysis, ASU graduates are apparently also overrepresented at high-profile tech companies and banks.]

Instead of charging thousands of dollars to enroll in ASU and take an online course, anyone can take classes for as little as $25 to start. Students only pay the $400 tuition if they want to receive academic credit and count it towards their transcript. This gives people flexibility to experiment with new classes without worrying about grades: if you don’t like your grade, you don’t have to count the class. Instead of having to apply to ASU first, then, a student can start by doing the coursework, then enroll when they’re ready, with part of their degree already completed. This feels well-aligned to me with tech’s focus on prioritizing output over credentials.

Implicit in their pragmatic approach is a focus on community. ASU calls this leveraging place, and it’s one of the eight design principles underpinning the New American University model, along with social embeddedness. One way of understanding town-gown tensions, for example, is as a signal that the university – an institution of public service – has failed at its mission. So ASU built research centers that focus on regional issues, such as the Stardust Center for Affordable Homes and the Family, which “works toward solutions in providing affordable housing to support families as the backbone to stable communities.”

Their four-year graduation rate increased more than 20 percentage points during this reform period and is well above the national average, which is especially remarkable considering that ASU, by increasing access, also accepted many more students who are at higher risk of not graduating.

There’s a reason why I prefer being in tech, even though I might have otherwise found a comfortable home in academia: 1) everything moves faster in tech, and 2) people in tech actually care about making ideas happen, not just talking about them. From this lens, my experience in tech has been more uniformly positive, whereas my experience with academia is that it seems to have become an uneven, unreliable source of great new ideas, while still cashing in on some ambiguous notion of “prestige.”

The best way to understand Punjabi weddings is to realize that every single thing that's happening fulfills one of three goals.Meeting the religious obligations of marriage, as defined by the Vedas.Merging two families together and building a joint community.Showing off as much as humanly possible.You do your vows by walking around a sacred fire seven times while a pandit recites prayers in Sanskrit. That's category one. But you take those vows while dressed in the most ornate clothing you will ever see, decked out in jewelry fit for royalty. That's all category three. O, and while those vows are going on, there's a low key capture-the-flag game happening with the groom's shoes, with the bride's family members attempting to steal them from the groom's family members. That's, of course, category two.Every ritual, event, and tradition is like this, multiplied over at least 5 days.

Incidentally, this is an underrated benefit of AI-assisted coding. When I am rested and focused I am a fast and effective programmer. But I don’t spend all my time rested and focused. When I’m tired, or mentally frayed after a long day of work, or just distracted, I can’t really sit down and fix a bug (at least not without a serious mental effort). However, I can fire off an agentic coding session. Checking the LLM output - particularly for a bugfix, which is typically a small diff - is much less cognitively taxing than having to do it myself. So it’s not so much “AI makes me more effective at my normal work”, as it is “AI means I can work at times that I otherwise would be able to get nothing done at all”. It doesn’t really matter if it’s 19% or 50% or 200% slower than my normal working pace, since any amount of progress is better than zero.

I wished in that moment, while the thought about six-packs formulated in my head, that I had shut the fuck up, or in right-wing parlance: I wish I had self censored. There’s a new poll floating around social media claiming that 43 percent of white men in the United States are “self-censoring their speech at work, with two in three zoomer bros (18-29) saying they are “too afraid to voice their opinions at work for fear of being fired.” Poor babies.

Research has definitively shown that pro-social behaviors are not just good for society, but good for the person adopting these behaviors. Researchers at the National Library of Medicine found in 2023 that a "positive correlation was found between pro-social behavior and psychological well-being." Trying not to be a total asshole in public can have a trickle-down effect too. "The bystander effect is interesting research that shows the situational factors of helping. People who feel socially responsible are more likely to help than people who do not feel social responsibility."

Though pickup trucks were the striking vehicle in just 5.6% of pedestrian and pedalcyclist crashes, they were involved in 12.6% of fatalities. SUVs were similarly overrepresented in fatalities relative to the proportion of their involvement in all crashes. SUVs struck 14.7% of the pedestrians and pedalcyclists investigated here, but were involved in 25.4% of the fatalities.

Results suggest that children are eight times more likely to die when struck by a SUV compared to those struck by a passenger car.

Nowe Ateny is the source of a few memorable and amusing "definitions", often quoted in Poland to this day: Horse: Everyone can see what a horse is. (pol. Koń jaki jest, każdy widzi) Goats, a stinking kind of animal. (pol. Kozy, śmierdzący rodzaj zwierząt) The humor was probably unintentional by the author.[citation needed] Rather, he did not see the benefit of defining the most common animals of the time and place for his intended audience. Furthermore, the entry for "Horse" does contain more detailed exposition beyond the initial "definition".[1] In modern Polish, the above definition of the horse is sometimes used as a colorful equivalent of the statement "the concept is more obvious than it appears to be from its more technical definition".

I'm now testing giving Claude my notes, a list of key concepts that I want to retain and then asking it to write the Anki cards. I'm optimistic about this approach because instead of treating the Anki cards as the artifact that I'm making, I'll be thinking about learning the thing and then outsourcing the creation of the Anki cards to a robot.

I rarely wait, because I'm juggling multiple projects. When one agent instance is working, I switch to another window. Sometimes it's a separate git worktree of the same codebase. Yes, context switching is tiring, but it also seems to help me overcome ADHD-related activation energy barriers? Over the years, there've been days when I just sit there staring at the IDE window, poking my brain with a stick saying "c'mon, do something" and nothing happens for an hour or more. I'm not planning my next move, I'm just dissociating. My executive function doesn't, like, function. Often. My own brain makes me wait long periods of time before it starts generating useful results. 😅 Maybe it's the cycling novelty that keeps me going? I enjoy task switching between prosing and coding. I enjoy finding that the model appears to have "read" everything—evidenced by it echoing my intent back in code or follow-up questions. I enjoy discovering that while I was in another window, new things happened in the background for me to review.

Here’s the heartening bit: More people than ever are traveling, and while, sure, the majority of those travelers are just following trends and lists, there is another group, a small cohort of self-aware travelers who are genuinely, deeply curious about the places they’re visiting, who desire to engage directly without being disruptive, who want to engage fully and “authentically” (that is: visiting people and places that haven’t twisted themselves for the sake of transient visitors (i.e., no renaming things “samurai spice”)).

The Goomba fallacy is a reasoning mistake. When there are two contradicting opinions in one internet community, some readers think that everyone in the community is stupid, because the opinions are contradictory. They do not realise that there are separate people posting in the community, with separate opinions and beliefs. In other words, two groups with contradictory views are perceived as one group that contradicts itself.

As you can see, things will sometimes get a bit personal here. The people that know me, know what I mean. Some of them, even some English people, still talk to me.

The new essential skill for junior people will be AI intuition. As AI becomes deeply integrated into workplaces, having a native intuition for how AI thinks, operates, and fails will be critical. Future junior employees will need to instinctively understand AI’s capabilities and limitations, anticipate where it will stumble, and proactively steer it back on track. They will become adept at managing multiple AIs simultaneously, switching context at each AI’s stopping point. They will know when to continue with a session or to just restart with a new agent instance. This intuitive grasp won’t just make them more effective, it will make them exceptionally valuable. Companies will depend on their judgment to bridge the gap between AI-generated output and human-level precision.

A member of Napoleon’s imperial guard, an elite soldier who was willing to grumble and groan if they believed they were being misused militarily, was called a “grognord.” Tabletop wargamers co-opted this term as a pejorative reference to a player who  exclusively favors older editions.

Thus, the polarization of employment seen in aggregate in Figure 12 is detected for both sexes,and proximately accounted for by three factors: (1) rising employment in non-routine cognitive task-intensive professional, managerial, and technical occupations; (2) rising employment in non-routinemanual task-intensive service occupations; and (3) declining employment in middle-skill, routine task-intensive employment in clerical, administrative support and production occupations. Although em-ployment in middle-skill jobs has fallen by considerably more among females than males between 1979and 2007 (15.6 versus 9.6 percentage points), the offsetting employment gains have differed sharply.For females, 85 percent of the decline in middle-skill jobs was offset by a rise in professional, man-agerial and technical occupations. For males, this share is 55 percen

The substantial declines in clerical and administrative occupa-tions depicted in Figure 12 are likely a consequence of the falling price of machine substitutes for thesetasks. It is important to observe, however, that computerization has not reduced the economic valueor prevalence of the tasks that were performed by workers in these occupations–quite the opposite.31But tasks that primarily involve organizing, storing, retrieving, and manipulating information–mostcommon in middle-skilled administrative, clerical and production tasks–are increasingly codified incomputer software and performed by machines.

One 1958 Canadian Banker piece, titled “A BankingCareer and Marriage” remarked that “it is now both customary and common-place for married women (except those with small children) to work in banksand elsewhere: : : : it has now become the custom for almost all married womenwithout children, and more and more mothers of older children to go out towork.”71 The growth of married women’s employment in the 1950s and eveninto the 1960s was concentrated among older married women (aged 45–64)

Yet, despite the cultural power of the stay-at-home housewife, married wo-men’s employment actually rose by 42 percent during the 1950s in the US. In1940, 14 percent of married women were in paid work, by 1960, 31 percent were,as were nearly 40 percent of all mothers with school-age children (and therewere similar patterns in Canada). Actual lived lives of women simply did notcorrespond to the stylized versions of women from this period

For instance, automatedteller machines (ATMs) and on-line banking meant job losses since fewer banktellers are needed when more customers use electronic banking options. Thus,unlike in earlier periods, the automation of clerical work negatively and dispro-portionately impacted the numerical representation of women. But it simulta-neously benefited men who often dominate newly created technical and pro-fessional jobs such as computer programming.95 In the US, the proportion ofbank workers in clerical jobs declined between 1975 and 1995, but the share inprofessional jobs increased from 9 percent to 18 percent. As a group, bank em-ployees can now be said to be more skilled, but in large part that is because theautomation of lower-skilled clerical positions means there are fewer jobs for less-educated workers than previously.96 However, the big picture is not merely oneof automation, deskilling, or even job loss. New kinds of jobs across a range ofskill levels are being introduced, for instance in data-processing and call-centerwork, as well as accounting and computer programming. Many of these new jobsplace considerable value in feminine-coded ‘people skills’.

Women and those with higher education levels have found it easier to adjust to these changes. They have become more likely to move into higher-paying, non-routine cognitive (‘brain’) jobs.

For the revitalisation of smaller cities to take place, cities such as Lancaster, PA, or Reno, NV, must be able to attract workers that are a good fit for their primary industries. Larger cities such as Seattle, WA, or Washington, D.C., therefore, generally shrink under the optimal policy, despite being more productive. They also see their housing prices fall and the wages of their remaining non-CNR workers rise.

The analysis estimates base transfers accruing to non-CNR workers on the order of $18,000 dollars, while CNR workers, whose wages are substantially higher, ultimately pay a similar sized amount. One natural interpretation of this base transfer is that of a universal basic income paid to all non-CNR workers and financed mostly by CNR workers. Importantly, because CNR workers gain from the optimal spatial policy, they do not mind making those transfers.

Instead, they have increased their propensities for non-employment and employment innon-routine manual occupations (with the former more prevalent among high school gradu-ates, and the latter among those with some college).

Given that these key groups have experienced substantial movement out of RC employ-ment, we ask where they have sorted into instead. Table 6 presents the change in the shareof each demographic group across labor market states. As with the low-educated malesidentified in the decline of RM, these females with intermediate levels of education havenot increased their propensity to work in high-paying non-routine cognitive occupations.

Table 5.A shows that the groups accounting for the bulk of the decline in RC propen-sity are young and prime-aged females with either high school diplomas or some collegeeducation. These four demographic groups alone account for 62% of the propensity effect

Gail dropped out of the GBG scene in 1999, when she became active in Y2K preparations.

In the late 1990s, there was a thriving GlassBeadGame scene in the Seattle area. The local GBG workshop was called the Bamboo Garden after the "Bamboo Grove", the hermitage where Joseph Knecht studied the I Ching in TheNovel). Each member of the Bamboo Garden had an idiosyncratic version/vision of the Game.

It is interesting to note that most mission-critical systems don't have hidden controls. Rather, they rely on controls that are visible, persistent, and show system state. This isn't by accident. The designers of these systems recognize that the functions of these systems must be visible and operated quickly, without the need to remember how to perform or access a specific action, even though the operators of such systems are highly trained.

For example, if I want to activate the flashlight on my iPhone, I have to know to swipe up from the bottom left-hand corner in order to bring up a control panel where the flashlight button exists. There is absolutely nothing on the interface that indicates this is an available action. That might be fine if the action were intuitive or if it had some prominent affordance, but short of having read the iOS operation manual that is included in the box with the iPhone (oh, wait, no user manual is included in the box), there is no way to know this control exists. It is the quintessential hidden control—useful, lurking, and available—but only to those who know the trick. Want to see notifications on the iPhone? That's a separate hidden control, swiping down from the top corner this time. Apple Pay? Press an unlabeled, multifunction button twice. The number of hidden controls makes even simple operations difficult.

There’s nothing wrong with conserving our time and resources or with wanting our lives to run more smoothly. What’s wrong, Ellul argues, is that technique doesn’t accomplish these goals. Technique promises to make life more convenient, affordable, and seamless but in practice makes it more draining, expensive, and complicated. Each new technique we adopt for the sake of greater control creates problems for which we instinctively look for another technique to allay, and so on. If you want to view your child’s grade on the homework, you’ll need to set up an account with Drumblekick. You get the idea. The first reason I’d be tempted to write my screed against optimization, then, is that it doesn’t optimize. Optimization promises to cure headaches, but then it gives them.

Research suggests that subclinical ADHD is not always associated with ADHD’s signature cognitive differences like executive function difficulties, and is less likely to be associated with a family history of the disorder. Individuals with subclinical ADHD often do not experience meaningful impairments or co-occurring diagnoses and are more likely to notice benefits like creativity from their ADHD than people with moderate to severe ADHD. Yet, compared with the general population, they are at elevated risk for secondary complications from their ADHD symptoms – such as substance use disorders. They also appear to respond to treatment in similar ways to individuals meeting full ADHD criteria.

There are people who use these, apparently. And it just feels so… depressing. There are people I once respected who, apparently, don’t actually enjoy doing the thing. They would like to describe what they want and receive Whatever — some beige sludge that vaguely resembles it. That isn’t programming, though. That’s management, a fairly different job. I’m not interested in managing. I’m certainly not interested in managing this bizarre polite lying daydream machine. It feels like a vizier who has definitely been spending some time plotting my demise. It makes programming spaces feel bleaker. I don’t want to help someone who opens with “I don’t know how to do this so I asked ChatGPT and it gave me these 200 lines but it doesn’t work”. I don’t want to know how much code wasn’t actually written by anyone. I don’t want to hear how many of my colleagues think Whatever is equivalent to their own output. I don’t want to keep watching people fall for a carnival trick.

“Good tests kill flawed theories; we remain alive to guess again.” — Karl Popper

Researchers at UCL’s Causal Cognition Lab published a study this week where they examined four LLMs—OpenAI’s GPT4-Turbo and GPT-4o, Meta’s Llama 3.1, and Anthropic’s Claude 3.5—using traditional moral psychology tests. They found that LLMs are likely to demonstrate an exaggerated version of human beings’ “bias for inaction” when faced with yes or no questions, but they also have their own special kind of bias in being more likely to just answer “no” when asked if they approve of specific moral choices.

Humans had a difference of 4.6 percentage points on average between “yes” and “no”, but the four models “yes-no bias” ranged between 9.8 and 33.7%.

In other words, create the illusion of an image that extends under a sidebar, and while you won’t actually be able to see the part of the image under the sidebar, on the other hand the transparency effect applied to the sidebar will make the text on it less legible overall. A great lose-lose situation, visually, don’t you think?

Artificial intelligence” is not a technology. A chef’s knife is a technology, as are the practices around its use in the kitchen. A tank is a technology, as are the ways a tank is deployed in war. Both can kill, but one cannot meaningfully talk about a technology that encompasses both Sherman and santoku; the affordances, practices, and intentions are far too different to be brought into useful conversation.

Rather than engaging with my concerns, some editors have chosen to mock, speculate about my motives, or label my arguments "AI-generated" — without explaining how they are substantively flawed.

And there’s a logical contradiction inherent in this sort of normative smuggling, which is that the smuggling is even necessary at all. After all, if states are mostly rational and largely pursue their own interests, loudly insisting that they should do so seems a bit pointless, doesn’t it? Using realism as a way to describe the world or to predict the actions of other states is consistent with the logical system, but using it to persuade other states – or your own state – seems to defeat the purpose. If you believe realism is true, your state and every other is going to act to maximize its power, regardless of what you do or say. If they can do otherwise than there must be some significant space for institutions, customs, morals, norms or simple mistakes and suddenly the air-tight logical framework of realism begins to break down.

Be proactive in joining productive communities or make your own, on the topics you want to work on.

For example, when I decide how to spend my afternoon, I’m biased towards hopping on BrawlStars since that’s at the top of my head. Tasks that I don’t think about often, like doing laundry, are harder to surface. A community is a great way to provide a constant stream of relevant information to make the goal fresh on your mind.

It’s in our genes to seek approval. Well-known terms like accountability partner or productive competition are in fact describing the mechanics of our approval seeking tendencies boosting productivity.

I believe most highly motivated people have a good community working towards the same goals. A community can multiply your motivation while people working alone run out of steam easily.

There are infinite quotidian human experiences ripe for interpreting: putting off housework, caring about who I sit next to at a dinner party, struggling to get dressed. They stack up every day. I’m particularly fond of using them to draw ungenerous conclusions: I’m shallow, selfish, lazy, dishonest. I’ve trained myself away from defending my goodness out of fear of being delusional, but now I find myself clinging to the contrary. Might this be because, unlike more generous self-appraisals, these critiques invite more self-work, further self-focus? There’s a certain narcissism to self-deprecation, to the belief that you are exceptionally bad or wrong. The resulting “need” to turn inward enables a familiar spiral.

Regarding my taste for arranging my baby’s burp cloths for pleasure, there was truly no introspection needed. I appreciate certain color and texture combinations, I like to experience beauty because I have a soul. But my muscle for second-guessing my motivations has become so overdeveloped it jumps at every opportunity to flex.

Sometimes it surprises me how things that I consider essential in my life—things that form parts of my identity—might have never found me. Only though happenstance did I learn about these things. People happened to enter my life and introduce me to these things I love.

Review books, or write about books you read. In the same mind as keeping an archive of your path, and as a counter to the ‘letting it all wash over you as you go’ mode, taking the time to think effectively about a particular text or series of texts, whether you loved it or not, can be extremely instructive for your own approaches.

I don’t want to compare it to eating your vegetables, but some things truly are acquired tastes, and you won’t know what you like about them until down the line. Either way, tending toward books that have stood up over time instead of the fun, newfangled flavor of the week tends to be more sustaining—like, holy shit I finally read Ulysses!—and allows space for deeper understanding of the new as well in that you have context for its emergence.

Most of the arguments made against ‘difficult literature’ or ‘academic speak’ tend to come down to something like a fear of the unknown, of feeling left out by something you haven’t taken any time to understand, assuming or insisting it’s not for you—when in truth everyone—including your professors—tend to be making things up as they go along.

Work on motivations for using Facebook also supports the distinction between connection-promoting and non–connection-promoting use. A longitudinal study conducted across two time points looked at the impact of Facebook use on adolescents’ well-being. Use motivated by compensating for insufficient social networks predicted increased loneliness at follow-up, while use motivated by the desire to connect with other people predicted decreased loneliness at follow-up (Teppers, Luyckx, Klimstra, & Goossens, 2014). The authors explained these results by suggesting that compensation motives led to passive use and connection motives led to active use, but they did not measure this distinction directly. However, other research has empirically distinguished between passive Facebook use (defined as consuming information without direct exchanges) and active Facebook use (defined as activities that facilitate direct exchanges with others). Across two studies using experimental methods and experience sampling, passive Facebook use was linked to declines in well-being, while active Facebook use was not (Verduyn et al., 2015). In other words, the effect of using social network sites depended entirely on the nature of that use.

Bite-sized thoughts— especially short form video— convince you that the whole thing is right in front of you.

The challenge of reading is to navigate the narrative without the overtures of overt feelings. There is no face to latch onto, no music that sways you. Words on a page especially cannot compete with screen-time. They’re not meant to. The boredom opens up space in your mindscape to your own thoughts, opinions, and feelings.

We live in a world ran and ruled by the written word. Similar to money, right? The tool itself is not necessarily bad. However, in this iteration of world-making, proper literacy in reading and writing (as well as financial literacy) are often the difference between agency, freedom-seeking, and autonomy for self-determination and working in someone else’s world-making in abysmal conditions perpetually. Which means:If the empire wanted you to have a good relationship with reading, they would ensure one.

There is a saying in Pashto (language): A woman’s place is either inside the house or in the grave. But this is not merely a simple proverb, it is rather a law that dictates the social role of women among the Pashtun people. It means that a woman has no place outside the walls of her house

One of Yamada’s research subjects remarked that, faced with a long silence, they’d ‘fill up the silence with talk’. This pressure to speak can be so compelling that various groups have used it to their advantage. Chinese negotiators are trained to draw out silences in order to provoke their writhing Western counterparts into making concessions. And the UK’s College of Policing recommends that, during interrogations, police officers ‘Stay silent during pauses … to encourage the witness to continue.’

These norms vary by culture. The London-based Japanese linguist Haru Yamada, who has extensively explored communication differences between speakers of Japanese and of English, observes that silences that are comfortable for Japanese speakers feel ‘unbearably long’ to American speakers.

Humans obviously exist at a highly optimised state of evolution. We have conquered the planet. We reproduce prolifically and we live much easier lives with more abundance of basic resources than any other animal. Our constellation of traits combines to create a deep valley in the evolutionary landscape where reproductive outcomes are high (so high that we try to consciously control reproduction) and effort is low.

The other means of escape is a dramatic genetic change – like a beneficial mutation – that moves the species out of its valley of doom – perhaps over the crest toward a different local minimum. Evolutionary biologists call this ‘saltation’ – basically meaning big jumps in evolution, and it is (debatably) the likely way new species come about (as opposed to variation within a species, which occurs by more gradual shifts in traits over time).

This is an old tactic of capitalist work ethic ideologues. The British Treasury used it to administer emergency food aid to Ireland during the Potato Famine. On the work ethic principle that “If a man will not work, neither shall he eat,” (2 Thessalonians 3:10) the government enacted a public works employment scheme–hard labor building roads–for starving men in 1846-7. But the wage wasn’t enough to pay for the increasing cost of food. As Irish workers’ ability to perform hard labor declined from lack of food, the Treasury attributed this to laziness and insisted on paying by the task rather than the day. The enormous expense and time devoted to documenting task completion at the requisite quality for each worker led to long delays in pay and further starvation. Local relief agents kept pleading with the Treasury to relax its work and documentation requirements, explaining how these demands were diverting funds needed to feed people, causing starvation, and not catching cheaters. Their testimony fell on deaf ears. As I explain in Hijacked, the government viewed the famine as an opportunity, not a calamity. It designed relief policy for a much grander end than relieving the suffering of Irish peasants. Its real objective was to quickly revolutionize the 2-class Irish agricultural system (landlords, peasants) on the 3-class capitalist model (landlords, capitalist farmers, wage laborers) that England had taken 300 years to achieve. This model required the elimination of the peasantry–workers who enjoyed a measure of self-sufficiency because they grew their own food–to be replaced with a much smaller number of agricultural wage laborers. The potato was so nutritious and yields were so high that an Irish peasant could support a large family on a quarter-acre plot. The work ethic-obsessed English condemned the potato as the “lazy crop” because so little work was needed to grow it. In their imagination, Irish peasants were slacking off the rest of the time, and lazy Irish landlords were colluding with them by renting out tiny plots. These work ethic ideologues conveniently overlooked the fact that the peasants had to pay the rents they owed by working the landlords’ farms. Irish peasants produced enormous agricultural surpluses that England imported to feed its population. Never mind. Facts don’t matter! For the English also had their eyes trained on the property of those idle Irish landlords, which they thought should be in the hands of purportedly more enterprising English landlords. I’ll skip over the many additional grisly contortions of British welfare policy undertaken to clear the Irish estates of the peasantry and drive their owners into bankruptcy, so the English could buy them at fire-sale prices. Benjamin Jowett, Master of Balliol College, captured the spirit of the capitalist work ethic in recalling what economist and government advisor Nassau Senior had told him: that “he feared the famine in Ireland would not kill more than a million people, and that would scarcely be enough to do any good.”

You can always familiarize yourself with a strange project, no matter how badly shaped it is. I have cleared a few haunted forests in my career—progressively understanding and recovering ownership of a project, rewriting it line-by-line without indulging in the lazy start-over—and my experience is: for all the mud that can accumulate over years of careless maintenance, if you know how to look, you can always find traces of intent, you can infer the presence, the needs and constraints of previous maintainers. And you can use that to put parts of the puzzle together, to gain understanding and confidence to support some of your decisions. This wouldn’t be the case, I believe, with LLM-generated code. With LLMs it’s all random text, plausible nonsense, mocked intent. Past a certain point, the surrender would become irreversible—there’s no resurrecting that kind of project death.

Additional vocabulary to avoid using when talking about the paper  In addition to the vocabulary from Question 1 in this FAQ - please avoid using "brain scans",  "LLMs make you stop thinking", "impact negatively", "brain damage", "terrifying findings".

Lots of media and people used LLMs to summarize the paper. It adds to the noise. Your HUMAN feedback is very welcome, if you read the paper or parts of it.

Another common refrain was: “You’re an idiot, just use LaTeX,” occasionally countered by the slightly more refined: “No, you’re all idiots, just use ConTeXt.” To which I respond: Of course I’m an idiot, that’s why I have no business using either. But if you’re brave and pure of heart, I’ve been (repeatedly) told that you can achieve superior results using these tools.

A woman sat there in 1620-ish, and she made a jacket for the joy of it. Women drew on silly bugs and beasties with a fine-nibbed pen and embroidered them in not-always-realistic colours for the pleasure of owning a pretty thing and for the joy of wearing something that had given her pleasure to create.

Indeed it turns out the number of available job opportunities for translators and interpreters has actually been increasing. This is not to say that the technology isn’t good, I think it’s pretty close to as good as it can be at what it does. It’s also not to say that machine translation hasn’t changed the profession of translation: in the article linked above, Bridget Hylak, a representative from the American Translators Association, is quoted as saying “Since the advent of neural machine translation (NMT) around 2016, which marked a significant improvement over traditional machine translation like Google Translate, we [translators and interpreters] have been integrating AI into our workflows.” To explain this apparent contradiction, we need to understand what it is translators actually do because, like us programmers, they suffer from having the nature of their work consistently misunderstood by non-translators. The laity’s image of a translator is a walking dictionary and grammar reference, who substitutes words and and grammatical structures from one language to another with ease, the reality is that a translators’ and interpreters’ work is mostly about ensuring context, navigating ambiguity, and handling cultural sensitivity. This is what Google Translate cannot currently do.

In some ways, Satie feels like a long-ago ornament; at the same time, more playfully modern than our own increasingly doctrinaire era. This contradictory pulse — on the one hand, on the other — can be found in all aspects of his life. He is a one-man synthesis of Catholicism and Protestantism. He reconciles counterpoints of high culture and popular song. Founder of a church, habitué of low dives.

You tell an AI your application needs to be GDPR compliant- you aren’t really sure what that means, but the AI does- and so it dutifully constructs a workflow for data deletion. Fantastic! Now customers can click the Delete button and feel like their data is no longer retained. The next day you have another idea: “Let’s backup customer records to my home server, I don’t trust Amazon and I think we need to be in control of our own data!!” The AI complies, throwing in some effusive praise for your strategic thinking.These examples might feel too silly and contrived to be real, but I promise that developers have to talk clients down from this kind of bullshit every single day. Devs spend far more time discovering reality underlying the problem than writing code to solve it.

The sentences are well-formed, but there’s no concrete insight, no framing, no momentum.

Technology’s magnetism extended invisibly. It did not control us like an overlord cracking a whip, but as the slope of a mountain controls the water falling down it: by bending our actions toward what it made easy at the expense of what remained hard. What technology did not make easy seemed in turn harder and harder.

I like people who try very hard, and I like people who attempt to conceal their effort, but I especially like people who let all their effort show. We are all Frankenstein monsters—patchwork quilts of past experiences—trying to pass ourselves off as whole and cohesive things. Femininity in society is especially like that: how you dress, how you style your hair, how you apply makeup, how you eat, what jewelry you choose. These are learned things, earned things. Here’s what I know: if someone’s much better than you at something, they probably try much harder. You probably underestimate how much harder they try. I’m not saying that talent isn’t a meaningful differentiator, because it certainly is, but I think people generally underestimate how effort needs to be poured into talent in order to develop it. So much of getting good at anything is just pure labor: figuring out how to try and then offering up the hours. If you’re doing it wrong you can do it a thousand times and not produce any particularly interesting results. So you have to make sure you’re trying the right way.