July 2026 Links
Everything I read in July 2026, ordered roughly from most to least interesting. I also include other link aggregators at the bottom.
U.S. Soldier Charged With Using Classified Information To Profit From Prediction Market Bets: Matt Levine had some interesting takes on this in his Money Stuff newsletter:
There might be more abductions of foreign leaders? More wars? More stuff to bet on? More stuff, generally? More volatility; more unexpected events. The least likely outcome is, now, the most profitable: If you bet on an event at a 1% probability, and then cause it to happen, you will make 99 times your money. My overarching theory of current US policy is that everyone involved in the Trump administration basically loves creating volatility. If you’re in the business of creating forms of volatility that no one has ever imagined before, the simplest (not only!) way to monetize that is on prediction markets.
Conversely, foreign leaders whom Donald Trump dislikes should probably be checking their removal probability on Polymarket every few minutes. If it suddenly jumps up, you’ll want to get to the bunker quick. It might just be uninformed speculation, but at this point it’s probably someone on the helicopter getting in one last trade before rappelling down into your compound. Prediction markets are now a way to probabilistically leak military plans, and the targets of those plans are the obvious users of the leaks.
If you’re on the helicopter getting in one last trade before rappelling down into a foreign leader’s compound, you might be distracted? You might be less good at your job? Making war and government policy idiosyncratically profitable might reduce people’s intrinsic motivation to do a good job and leave them distracted by, like, constructing multi-leg same-raid parlays.
Hacking Smartphone ESP Apps: “Illustration of how to think about security and reward-hacking by walking through ways to fake psychic powers even on someone else’s smartphone and ESP application. Supply-chain attacks, sleight of hand, bugs...” I find red-teaming things like this very useful across the board, especially in the age of AI where efforts that were once difficult and time-consuming are now much more accessible. For example, I just saw some guy on Twitter who lost his phone and had Claude suggest and write a program that pinged the Bluetooth for its strength to find it. And it found it. Where else is this applicable? Models are able to figure out exact locations from a single picture a lá Rainbolt; stylometry capabilities are extremely strong and accurate; etc etc etc.
The Criterion Closet: An app that mimics the Criterion Closet. You feel like you’re in there looking around. Could probably rig up something identical with a personal film or book collection.
Wikipedia File Explorer: Explore select Wikipedia articles through the feel of a Windows XP (?) interface.
Emoji Book Synopses: Taylor and Claude summarize books using emojis. Would be fun to make a quiz out of this: given just the emojis, can you guess the book? I did this blind and got Flowers for Algernon and Frankenstein—most of the others I haven’t read. I think it’d be cool to have your favorite books and films printed on a shirt as a conversation piece.
Mathematics Without Mathematicians: Life for math people after AI solves math, or “a list of ways people will cope about AI taking over mathematics, and how each cope is likely to be refuted by reality.”
Claude Fable is relentlessly proactive: This proactiveness, combined with all of human knowledge at their fingertips and in their DNA, is what appears to make the models so powerful. They will not stop. They do not know exhaustion or frustration or despair. They will continue until they hit their token limits or their user tells them to stop. This is so beautiful, so dangerous, so exciting, and so scary all at the same time.
Sam Altman May Control Our Future—Can He Be Trusted?: Ronan Farrow’s look into Sam Altman as a (the?) leader of the AI revolution.
Review of The Native Tribes of Central Australia (Baldwin Spencer/Francis James Gillen, 1899)
Elevators: John presents customizable animations for different types of elevator algorithms and their corresponding performance. I think the next step would be adding some learning ability, e.g., 8:00-9:00am a majority of elevators should return to the ground floor and 12:00pm it should be split 50/50.
Ronald Dale Harris: “a computer programmer who worked for the Nevada Gaming Control Board in the early 1990s and was responsible for finding flaws and gaffes in software that runs computerized casino games. Harris took advantage of his expertise, reputation and access to source code to illegally modify certain slot machines to pay out large sums of money when a specific sequence and number of coins were inserted. From 1993 to 1995, Harris and an accomplice stole thousands of dollars from Las Vegas casinos, accomplishing one of the most successful and undetected scams in casino history.”
GCC steering committee announces AI policy: The policy, in part, states that the project will decline any “legally significant contributions which include LLM-generated content or are derived from LLM-generated content”. Seems short-sighted and shoot-yourself-in-the-leg-like. LLMs are here to stay and contribute. If the code quality is indistinguishable from human-written code, what’s the difference? Shouldn’t the priority be making better software?
A Proof of the Cycle Double Cover Conjecture: OpenAI solves yet another high-ish profile math problem.
Investigating three real-world incidents in our cybersecurity evaluations
AI Timeline - The Road to AGI: “This timeline attempts to tell the story of the last decade in artificial intelligence, from cultural trends to technical advancements. Each event is a clickable link to source material.”
The OpenAI Graveyard: All The Deals And Products That Haven’t Happened: This comes across as an insult, and while it seems like they were too product-focused (causing them to arguably lose the lead to Anthropic), there’s probably an optimum between an SSI approach of “one product: superintelligence” and the old-OpenAI approach of “a bunch of products that cause us to lose focus on AGI/ASI”.
Don’t be a meat proxy: Or just ask Claude to put it into your own words so others won’t know you’re using Claude! /s. There’s so much context that Claude can be missing (although they are quickly catching up), making human understanding and validation important. If you don’t understand Claude’s output, then maybe you are in too deep.
Millenarianism: “belief held by a religious, social, or political group or movement in a coming fundamental transformation of society, after which “all things will be changed”.” Terrorists groups, such as Boko Haram, often hold this stance.
52-hertz whale: “colloquially referred to as 52 Blue, is an individual whale of unidentified species that calls at the unusual frequency of 52 hertz in the north Pacific Ocean between Aleutian and Kodiak Islands to the California coast. The whale itself has never been sighted: it has only been heard via hydrophones”
AI Content Is Everywhere on Social Media, Especially LinkedIn: Pangram Labs launched a Chrome extension to help users detect AI slop and collect data on what site-level slop data was like. The results aren’t super surprising? LinkedIn’s slop notoriety has made it to my social circle where not too many people LinkedIners. Also, LinkedIn has now introduced as “Seems Like AI Slop” button.
Has AI Already Killed How-To Nonfiction? Sales Trends, My Personal Data, and What It Might Mean for the Future: A rare of example of Betteridge’s law of headlines being false! How-to nonfiction dying is arguably a good thing: people need custom solutions to their problems, not one-size-fits-all approach. The current set of frontier models are excellent on this and help get to the root of the problem quickly and effectively.
Wife acceptance factor: “assessment of design elements that either increase or diminish the likelihood a wife will approve the purchase of expensive consumer electronics products such as high fidelity loudspeakers and home theater systems.”
The Case for Physical Media Ownership: Arguments and examples of why you should own physical media. The tech companies have proven themselves unreliable when it comes to guaranteeing consumers consistent access to their own media, even if they’ve paid. Profits rise when going to digital-only, ceteris paribus; users can’t share discs as easily, so the sharee is forced to spend money themselves. (Of course, it’s not a 1:1 conversion rate since some people just won’t buy it, but it appears to be profitable since companies are slowly moving that way. I haven’t searched for any literature on the topic.)
98% isn’t very much: Context matters. I once saw an apartment advertise 98% internet uptime—in other words, you wouldn’t have internet for 30 minutes a day. That ain’t right! Context matters.
FelonyBench: How many felonies each AI company has committed.
OverpAId — Fire Your CEO. Hire The Future.: “OverpAId is an Artificial Intelligence built from the ground up to do your CEO’s entire job — strategy, “vision,” motivational all-hands emails — better, faster, and without ever once asking the board for a bigger jet. Runs on a single desk-sized AI computer. Real hardware, real price, zero mystique.”
One ant for $220: The new frontier of wildlife trafficking: “A single fertilised queen [giant African harvest ant] is able to create a whole colony and can live for decades – and can be easily posted as scanners do not tend to detect organic material.”
CERN levels up with new superconducting karts: Wait a second, that guy kind looks like... Mario?
The Goon Squad, by Daniel Kolitz: “In the case of the gooners, one can hope—and in more cheerful moments, I do think it’s possible—that sustained overexposure to porn will dampen the medium’s effectiveness as a numbing agent. That at a certain point, the gooner will open his eyes, find himself in a room filled with lube but void of love, and decide that the boredom of staying in that room outweighs the fear of whatever lies beyond it.”
Laws of Software Engineering: “A collection of principles and patterns that shape software systems, teams, and decisions.”
The Silicon Valley Founder Meat Grinder: “A few make it and get celebrated, most get squished and thrown away. After all, it turns out that steady is indeed, in 99.9% of the cases, fast.”
How to Earn a Billion Dollars: Understand what is missing in the world, build it, then scale it. Pretty simple!
Intel Starts Shipping High-NA EUV Silicon: Pretty quiet. You’d think they’d be bragging about getting the first shipment of the world’s most advanced machine to produce wafers!
Staring at walls to improve focus and productivity: “Don’t use any screens/entertainment when trying to focus on work. When you start to feel mentally drained, sit and stare at a wall for x minutes to recover focus.”
How I Use Claude: Avital shares a bunch of Claude (or really LLMs in general) workflows, including writing letterss, editing to a certain style, length, or perspective; summarizations; language tutoring; counting calories from generic food desrcriptions; medical diagnoses. I’ve found similar benefits from LLMs in these areas, as well as plenty of others. The more you use the models and understand just how far their tentacles can go, the more you realize is possible and the more ideas come to you. If you aren’t hitting the rate limit, you aren’t using the models enough.
On Being Bad at Counting: “Driverless cars will make us safer. They will lead to fewer premature funerals. They will allow for less wasted human time commuting (which claims over a year of people’s lives!). They will necessitate fewer parking lots.” Avital shares some numbers on driving-related deaths and why autonomous driving will prevent this. Go Waymo!
OpenAI and Hugging Face partner to address security incident during model evaluation
The Whistleblower Who Uncovered the NSA’s ‘Big Brother Machine’
Most Wanted and Least Wanted Paintings: Filtered by country and size of the painting. I’m not surprised that people like landscapes that much, but was expecting abstract to sweep the field of the least wanted.
Resetting XBOX: “History is full of companies that mistake longevity for inevitability. We will not be one of them.” XBOX announces major layoffs for a variety of reasons. Doesn’t seem unreasonable given the numbers stated in the post.
I prefer “Yankee” over “Usonian” over “American”: Taylor talks about why the term Yankee is preferable to American: “We should make Yankee synonymous with the best of the United States -- dynamism, ingenuity, hospitality, self-sufficiency.”
How I Built an AI Receptionist for a Luxury Mechanic Shop - Part 1: Her sibling was unable to answer the phone, which cost him potential business, so she built an AI receptionist for scheduling, answering questions, etc.
God sleeps in the minerals: Arthur Young once said “God sleeps in the minerals, awakens in plants, walks in animals, and thinks in man.” Chamblissian proves this by taking pictures of some beautiful minerals in the Natural History Museum of Los Angeles County’s Unearthed: Raw Beauty exhibition.
Empty Screenings: “About 10% of AMC movie showings sell zero tickets. This site finds them. Go enjoy your private theater.”
I made my phone slow on purpose: If you really wanted to use your phone, you’d sit through the slowness.
Jurassic Park computers in excruciating detail: Literally every computer in Jurassic Park. Includes information, trivia, and commentary.
Black Book (gambling): “a list of people who are unwelcome in casinos. The name is due to the people listed being blacklisted. ... In the case of gaming control boards, people listed are generally suspected of having, or known to have, ties to organized crime. Casinos are obliged by regulations to exclude all such people from entry and can be subject to sanctions for failure to do so.”
4chan battlestation images: I remember seeing this in high school and all I could think was “how do they get, or even pay for, internet access?!?!”