Longreads + Open Thread

Today's issue of The Diff is brought to you by our sponsor, 2 Hour Learning.

Longreads

• There's lively and unlikely-to-be-resolved debate over the extent to which education's impact is mostly from selection, signaling, getting the right peer group, or actually absorbing the material. Since the first three make educators look bad, the default framing is that one of them is stronger than people think, and the skills-building part is weaker. Here's some evidence that education itself, aside from selection effects, has an impact: philosophy majors tend to start out ahead of their peers on some measures of open-mindedness and critical thinking, and tend to improve over time. (Amusingly, one sub-skill where their peers improve more is in looking up academic publications to support their point—if you train your argue-from-first-principles muscle too well, you neglect to wonder what the empirical work says.)
• How North Korean hackers infiltrated crypto projects. This is a story of two characteristics of crypto: remote-friendly companies that meet talent through Github and Discord rather than LinkedIn and recruiters will get a totally different talent base—worse credentials, but higher variance, and a few superstars who don't fit the traditional mold. But that also means it's a way for countries to export services, particularly if they couldn't do it when de-anonymized. The other unique thing about crypto projects is that they manufacture an easily-transferred bearer instrument, which is cheap to move but also possible to track. So it's the kind of environment where money is more likely to be stolen, but the thieves are more likely to eventually get tracked down.
• In The Atlantic, the harrowing story of college professors discovering that their students have never read a complete book before. And it's an elite school story: it opens with an anecdote about Columbia, and features quotes from professors at Princeton and UVA. UChicago, apparently, has been spared. The reason isn't that the students are incapable of reading a book's worth of information, but that they're used to short stories and excerpts, not complete works. Which raises the question of how important the specific skill of completing a book really is: if they could do it, technically, do they really need to? There's been coevolution between the economics of packaging information and the increments in which people publish their ideas; for the last few centuries, a default way to sum up an important idea is to produce 200-700 pages about it. And since earlier generations were used to thinking in those terms (before digital documents and Xerox, it was much more logistically challenging to assign excerpts from a dozen books, so you'd probably read one). So a lot of the intellectual tradition that schools are trying to pass on was first spread in book-sized chunks. It's hard to outcompete something that has a multi-century head start.
• An interesting Bloomberg piece on a container ship rental company that apparently started out legitimate and then switched to being a ponzi scheme after a while. This pattern seems more common among the biggest frauds: a pure ponzi can grow for a while, but it doesn't have much of a buffer. A big company that starts a small ponzi arm has plenty of buffer, and plenty of experience with what realistic returns look like. In this case, the company seems to have started out selling investments in a unique alternative asset, shipping containers. They still own about 1% of the global stock. But they claimed to own, and distributed the fake profits of, multiples of that. Since the liability of a fraud almost always expands faster than the business it's attached to, this is a good reminder that temporarily papering over a cash flow deficiency by fudging numbers can make a total collapse inevitable.
• Also on the education front, the Chronicle of Higher Education covers the rise of special disability accommodations, few of which are evidence-based and many of which are counterintuitive solutions to what they're essentially solving. The right ADHD accommodation isn't extra time to pass a test; it's probably more like having the option to spend the first hour of test time looking at the syllabus and textbook, possibly for the first time. To the extent that schools exist to impart knowledge, it would be a good idea for them to figure out whether these rules actually help in that process. And to the extent that schools are preparing students for the workforce, it hurts everyone—it's a noisier signal for the students who don't need this kind of help, and the ones who do have been operating under a more lenient set of expectations than they will in the professional world.
• In Capital Gains this week, we look at career progress as a process of being an increasingly central node in a growing network of smart people and useful information.
• In this week's issue of The Riff, we discussed the nature of competition in tech, rates, buybacks, and why having tomorrow's newspaper doesn't necessarily help make money in today's market. Listen with Twitter/Spotify/Apple/YouTube.
• And I'm very pleased to announce that Tobias Huber's and my book, Boom, is now available for preorder. Paying subscribers to The Diff also get a free copy, so stay tuned for more information on how to redeem that once the book is out.

Books

Supremacy: AI, ChatGPT, and the Race that Will Change the World: It's hard to write a business history about companies that, until recently, weren't producing any material revenue whatsoever. This book does a decent job of tracing the history of OpenAI and DeepMind, including some of their surprising parallels—DeepMind was a venture-backed company and then a big tech subsidiary that was desperately struggling to be some sort of nonprofit or not-quite-purely-for-profit business, at the same time that OpenAI started out that way and moved in the opposite direction. Since AI research was such a small world, and since many relevant people were involved in both companies, the book manages to get in some good stories about the founding and early days of the big labs.

The book is not, and isn't trying to be, a look at the technical aspects of AI. (For that, Why Machines Learn is a great introduction.) And, unfortunately, there appear to be lots of well-written NDAs, so some of the stories feel like they're missing critical details. The interpersonal focus makes some parts weaker than others—the chapter on AI ethics mashes together basically unrelated concepts (AIs delivering bad answers because of bad data; AIs delivering good-but-upsetting answers because they have more data) into one vague and seemingly useless idea—that AI tools can only be trusted when they tell us what we already knew.

Still, worth reading if you weren't following the AI story earlier and want to quickly catch up on the dramatis personae.

Open Thread

• Drop in any questions or comments of interest to Diff readers.
• Two of the pieces above argue that US schools are getting less rigorous, at least in subjects where it’s hard to measure academic outcomes. (It’s the nature of any subjective field that a standardized test will underperform for the very best students—if you come up with a new explanation for the cause of the War of 1812 or develop a groundbreaking interpretation of Hamlet, your answer isn’t part of the grading rubric.) What’s the best way to get the equivalent to what a degree used to signal? Tech people can contribute to open source projects or pass tech screens, but what about everyone else?

A Word From Our Sponsors

Can this really be true?

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

• A next generation tech media network covering the trends driving our future is looking for an editor in chief to help launch “The Free Press” meets CB Insights for the technology industry. (Remote)
• A hyper-growth startup that turns customers’ sales and marketing data into revenue is looking for a product engineer with a track record of building, shipping, and owning customer delivery at high velocity. (NYC)
• A growing pod at a multi-manager platform is looking for new quantitative researchers, no prior finance experience necessary, 250k+ (NYC)
• A startup building a new financial market within a multi-trillion dollar asset class is looking for a data scientist with actuarial experience. (if you’ve been an actuary but are newer to the data side, that’s great too) (NYC)
• A well funded startup founded by two SpaceX engineers that’s building the software stack for hardware companies is looking for new grad / early career software engineers. (LA, Hybrid)

Even if you don't see an exact match for your skills and interests right now, we're happy to talk early so we can let you know if a good opportunity comes up.

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.