Rosh Hashanah starts this Friday at sundown and opens ten days of celebrating the Jewish New Year, combined with reflection and atonement, ending with Yom Kippur. This time is meant to resolve and forgive outstanding disputes, debts, and grudges.¶
This was on my mind as I compared Uber ratings with colleagues at a meeting last week. I am a 4.93, surrounded by a 4.86, a 4.95, and a 4.92. Uber's rider score averages the most recent 500 rides. None of us could recall which ride, which driver, or what we did to lower our average. Speculation ranged from talking too much or too little to sitting in the front or closing the door too hard (or not hard enough). Nor could we account for the two decimal places. Does a 4.93 meaningfully differ from a 4.95? This is the tyranny of pseudo-precision, like a lab value without a reference range or diagnosis.
In the Uber home from BWI, I asked my driver. He recounted rude, messy, and drunk riders he had marked down. He also told me about losing stars for arriving too early to a scheduled ride. (The rider felt rushed, complained about finishing coffee under duress, and lashed out with three stars.) The driver knew because the app reports detailed rider feedback. He finished the story just before swerving onto the shoulder to avoid the back of a truck (looked at his phone too long, he said). He will know I gave him four stars for his distraction. Though shaken, I commiserated politely with his excuse. I wonder whether that magnanimity earned me five stars. If not, I will hear nothing.
Somewhere in my last 500 rides, I wronged an Uber driver. Uber denies me reconciliation and self-improvement. While drivers get both, I get a second decimal place.
Gemini Notebook AI-generated podcast version of this week’s newsletter.
No episode this week (I am on the road)
Historical episodes here: wair.ajwein.com/podcast
Science and Technology Trends¶
I want nuclear fusion to be real. I also want to understand fusion. Last week, I found a ScienceDaily article about how machine learning-driven software is enabling rapid control of the magnetic fields that contain the high-temperature plasma driving fusion reactions. The software is named PACMAN (Prediction And Control using MAchiNe learning). I took the opportunity to better understand fusion and recommend that you quickly scan the article and read my AI summary, which answers the question, “why does fusion seem 30 years away for the last 50+ years?”¶
Article: https://www.sciencedaily.com/releases/2026/09/260903064215.htm
AI Summary of the article and topic, including the recent history of nuclear fusion: https://wair.ajwein.com/fusion-ai-control-pacman-2026-09-06-v1¶
I found another coffee study last week. The social media post on X was provocative and imprecise, reporting (for the first time in my ongoing, self-serving, post- luxury espresso maker purchase validation efforts) that people who don't drink coffee may perform better on some memory tests. It was a complex study, and the central social media assertion was neither accurate nor supported by the data. The primary outcome was microbiome composition and function. Mood, memory, and cognition were all secondary, and the authors write that the study may not have been sufficiently powered to detect effects on them.
What social media said, and what the paper's data demonstrated:
Also of note, this is another study funded by Big Coffee. Specifically, the Institute for Scientific Information on Coffee, the industry body I flagged last December, whose five members are illycaffè, JDE Peet's, Lavazza, Paulig and Tchibo. Nestlé is not one of them, though the study coffee was Nescafé, and the senior author discloses separate research funding from Nestlé. Two different kinds of conflict.
https://wair.ajwein.com/coffee-microbiome-decaf-memory-2026-09-04-v1
So, I offer my own research database. Adam's Coffee Confirmation Bias Corner tracks every coffee study I have covered since 2020, with a second table of the foundational studies I've not reviewed yet. Here is the take-away: Three to four cups a day seems safe. Whether it is caffeinated doesn't matter to the bacteria in your gut, but the caffeine sharpens the brain. Espresso maker: Still justified.¶
https://wair.ajwein.com/coffee-health-tracker-2026-09-04-v2
Anti-Anti-Science¶
I am handing this week's anti-anti-science section to Dr. Jeremy Faust. In July, Dr. Craig Spencer, an ER physician and public health professor at Brown, argued in The Atlantic that public health should engage MAHA rather than dismiss it. In a world that has given these ideas a very loud voice, the attempt to contextualize them is rational. Faust, another ER doc and Spencer's friend, disagreed.¶
https://www.theatlantic.com/ideas/2026/07/public-health-maha-rfk/688014/
However, his friend Jeremy Faust (another ER doc and social media doctor) offers a line-by-line analysis of Spencer’s essay, highlighting the logical fallacies and intellectual gaps. Unsurprisingly, I align with Faust. MAHA dresses preconceived agendas and bad science in the language of just asking questions. A scientist asks a question because she wants the answer. A denier asks it when they don’t like the answer. And it is dangerous to decry and unwind entire branches of evidence-based science that have materially improved humanity’s quality and quantity of life.¶
https://insidemedicine.substack.com/p/i-didnt-love-craig-spencers-embrace
AI Impact¶
Several weeks ago, I was at a conference where three philosophy professors and a Claude instance debated whether AI is already conscious (Claude has some thoughts, for sure). This week, I found Steven Levy’s Wired article about this topic. While I would love to see the Galapagos do so, being “on a boat with philosophers talking 'the nature of consciousness' funded by a Russian millionaire philosophy enthusiast who runs dating sites” feels like an investment I am not ready to make.¶
Levy's conclusion is bleaker than mine. He writes that when models astonish the scientists who built them, the top concern is not consciousness; it is the scientists' inability to control them and their bosses' willingness to keep going regardless. I would add that whatever we end up calling it, we had better be ready to share the world with a new form of intelligence that has some degree of agency and volition.
https://www.wired.com/story/who-cares-if-ai-is-conscious-its-basically-alive/
I asked Claude Fable 5.1 to build a biological taxonomy around LLMs. It named chatbot LLMs Ingenium loquax, with subspecies designations by brand: anthropicum 'Claude' str. Fable 5.1. Here is the reasoning output: https://wair.ajwein.com/ai-taxonomy-2026-09-06-v1¶
FYI, Ingenium is Latin for innate talent or genius, and it's the root of both "engine" and "engineer.” Loquax is talkative. In other words, we homo sapiens better learn to live with ingenium loquax.
Related OpenAI commentary I found after I wrote this: https://openai.com/index/an-alien-mind/
If vibe coding is the new skill in software engineering, it shouldn't surprise us that researchers have started asking what makes someone good at it. The linked study is the first attempt at an answer I have come across.
A hundred students from ETH Zurich and the University of Zurich took a writing test, a computer science test, and a general reasoning test. Then they spent forty-five minutes on a coding platform, prompting an AI model and watching the application change without ever seeing the code. That last part is the experiment. It asks whether a CS education builds skill, even when you don't touch code.
The study measured three things that predicted performance. It did not measure two more things I think are important.
The model explains about a fifth of the variation. The other four-fifths are unexplained, not unmeasured, and based on other sources I suspect they include:
The researchers narrowed the range rather than suppressing the variable. Everyone had to have finished an intro CS course and to have used an LLM for programming before, so domain familiarity varies less in this sample than it does in the real world. The authors state that this limits generalizability to CS students, not professional developers or citizen programmers.
So, to sum up, good vibe coders need to think critically, know how computers work, understand the problem they are solving, and be willing to write clear requirements. Andrej Karpathy (who has worked at OpenAI and Anthropic) coined "vibe coding" to describe programming without seeing the code. This study demonstrates that there are still unique skills required, perhaps a derivative of today’s CS education, but a CS education nonetheless.
Paper: https://arxiv.org/abs/2603.14133
Summary of the study and related data on this topic: https://wair.ajwein.com/vibe-coding-predictors-2026-09-06-v1¶
The J&E Random Kidney Facts of the Week (JERKFoW!)¶
Airplanes create a kidney-related, often underappreciated, physiologic counterpoise. You land with a dry throat. Somewhere over the course of the flight, you have watched your fellow passengers watch the seatbelt sign, waiting to leap to the bathroom.¶
Lavatory visits are mostly driven by the complimentary beverages, including alcohol and caffeine, with cabin altitude a plausible but smaller and more variable contributor. Cabins are pressurized to usually 6,500 to 8,000 feet, and the lower partial pressure of oxygen can nudge the kidneys toward the same increased urination as people ascending into mountains. The body senses this change, rebalances hormones, and the kidneys offload sodium and water. However, this effect varies widely from person to person. The magnitude of fluid loss tracks closely with an individual's health, so some people make more urine more quickly than others. Second, and more importantly, increased urine output due to altitude typically develops over hours, not minutes. In fact, some people retain fluid rather than shed it in the first few hours at low oxygen pressure.
Thirst is a separate process. Cabin air runs at 10 to 20 percent humidity, drier than most deserts, and it dries out every mucosal surface it touches, hence the powerful sense of thirst. However, most people are far less dehydrated than they feel. On long-haul flights, sitting still and letting blood pool in the legs shifts fluid out of the blood vessels, making the blood slightly more concentrated and encouraging more drinking, though this is more due to the dry mouth. And some people with other illnesses (on lots of blood pressure meds or fluid pills) may not be able to tolerate this shift in volume, making them truly intravascularly depleted.
So, planes are a mixed bag, with lots of kidney-related signs and symptoms. For healthier individuals, it may be a mild change in how they feel. For older or less healthy individuals, the kidneys are participating in a complex response, making them need to get up more often throughout the flight. An excellent reminder for us aisle sitters to be gracious when the person in the window seat gets up for a second time.
Things I learned this week¶
I recently finished Wendy Moore's The Knife Man, a biography of the late-eighteenth-century surgeon John Hunter, the father of modern scientific surgery. I love reading about the history of medicine, the progression of thought over time, and the intellectual anchors along that timeline.¶
Hunter was a Scottish surgeon who was extraordinarily unconventional, continually disregarding the norms of Georgian British society while emphasizing critical thinking and rigorous scientific inquiry. He advanced humanity’s understanding of developmental and evolutionary biology, anatomy, physiology, and surgical technique.
The quotes from the book were highly entertaining and compelling.
"I thank you for your experiment on the hedgehog; but why do you ask me a question by the way of solving it? I think your solution is just; but why think? Why not try the experiment?" Brilliant. That is the whole method in nine words.
I went looking for his original writings and found a collection of very large PDFs scanned from 1841 - the collection of his books and writings that survived. Claude put them together for me in a searchable index with highlights, which I link here for your enjoyment.
https://wair.ajwein.com/hunter-corpus-1841-2026-09-04-v1
I learned that, among the many intentional and unintentional misapplications of radioactive materials at the beginning of the 20th century (see my March 2026 discussion of radium water), nothing was quite as egregious as the Radiendocrinator. (Yes, that is the real Phineas and Ferb-esque name the manufacturers gave this device.) The manufacturer's own instructions are better than any summary I could write: "Wear the adaptor like any athletic strap. This puts the instrument under the scrotum as it should be. Wear at night. Radiate as directed." All this is to say, I am a supporter (an athletic supporter?) of thoughtful, robust, and federal-level medical device regulation.¶
See and learn. This is another item from the Oak Ridge Museum of Radiation and Radioactivity. The museum’s write-up is my type of snarky incredulity.
This AI summary below adds a lot of detail - it quantifies the radiation delivered and grounds the phenomena of bad medicine in a broader context. My observation: the most enthusiastic Radiendocrinator users were probably sterilizing themselves, which, in some ironic karmic justice, kept them from reproducing to make more people who would buy these kinds of devices.¶
https://wair.ajwein.com/radiendocrinator-2026-09-04-v1
AI art of the week¶
A visual mashup of topics from the newsletter and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.¶
https://wair.ajwein.com/wair-art-demuth-gallery-2026-09-07-v1
This week's style: Charles Demuth's Precisionist poster portrait. Demuth rendered his sitters entirely as numerals, signage, and machinery, with no face in the frame. I Saw the Figure 5 in Gold is the famous one (I have the poster on my home office wall). It seemed like the right form for an issue that opens on a man reduced to two decimal places.¶
Clean hands and sharp minds,
Adam