August 2026 links #1
...
Faced with probability, AI can become a predictable parrot by Adam Kucharski
A week or so ago, a new pre-print asked 165 LLMs to name a random number between 1 and 100, as well as some other questions that have single token outputs, and showed that it’s possible to predict with 60% accuracy which model family was used based on these set of tokens.
If the work is good why does it need a story? by Claus Wilke
Regardless of how interesting and logically sound your work is, you will have to craft it into an engaging story if you want to reach a larger audience. The purpose of storytelling is not to manipulate. It is to overcome your audience’s inertia, most importantly boredom and disinterest.
LLM-enabled pandemics are fiction, for now. Don’t let them become fact. by Egan Peltan
Let me be clear: we are not yet at the point where anyone can vibe-code a novel, stealthy high-risk Potential Pandemic Pathogen (PPP). The headline AI-assisted synthetic viruses are innocuous phages, and in fact quite similar by sequence to the parental virus. In the future, novel high-risk pathogens might be a prompt away. For now, and in my estimation for the near future, it’s not. Perhaps it’s never possible.
Giving the AI models a biology exam by Matt Lubin
And this is the final line and the one most relevant to questions of biosecurity: Anthropic’s models may be doing lots of refusals, but if you want to get around them, you have lots of other options. Your accuracy may be reduced by some 10-20 percentage points, but you may pay as little as a fortieth of the cost to still get some pretty smart AI biologists.
Improving Fable 5’s biology safeguards by Anthropic
In our testing, this update reduced biology-related fallbacks by about 85% across our product surfaces.
[…]
It’s often difficult to tell apart beneficial and harmful uses of AI in biology. For example, in some cases researching a treatment for a disease requires scientists to produce the dangerous compounds that cause that disease in the first place. This is most obvious for live vaccines, which require scientists to grow the same pathogen they’re aiming to prevent.
Why AI-assisted bioweapons won’t kill us all by Abi Olvera
Since the 1980s, there has been exactly one fatal bioterror attack.
[…]
AI can supply biology and chemistry knowledge, but not hands-on skill.
[…]
Reading every surgery textbook doesn’t make you a good surgeon. The biology laboratory works in a similar apprenticeship model. You learn by failing repeatedly under a mentor who has also failed repeatedly.
Bioweapons as the Optimist’s Exception by O.H. Scharfman
She is right that bioweapons are hard today. She is wrong to infer that they will stay hard, or that this justifies being, as she puts it, “no longer worried” about bioweapons.
[…]
Unlike cybersecurity, biosecurity has a profound offense-defense asymmetry. An attacker does not have to go through clinical trials or pay for manufacturing. An attacker has time to design a weapon and can choose when to release, but the defender can mostly only start responding after the attack.
Benchmarking is not validation by Nils Homer
Benchmarking asks whether a tool performed well on a defined dataset at a point in time. Verification asks whether a tool continues to do what it claims to do, across versions, inputs, and edge cases. Clinical validation asks whether the result is fit for the medical use in front of it.
On Fermi Problems in Human Genetics by Jonathan Pritchard
These skills are now more important than ever, in this new era of vibe-coding. Let’s say Claude Code confidently gives you some answer. How would you know if its answer is even in the right ballpark?
The only reason you’ll ever need not to write with AI by Colin Carlson
Academic writing is a collaborative venture, and in my experience, very few researchers have a comprehensive conversation about GenAI use before they collaborate. [...] That means that all of us have the ability to end each other’s career.
LLMs reward expertise by Sean Goedecke
For many tasks, the human is the bottleneck, not the model, because the difficult part is in communicating to the model exactly what kind of solution the human wants. The information is “in the model” already, but it takes a very smart human to pull it out.
On Biological Data Generation 2/n: The Economics of Metrology by Michael Koeris
The unit I want to reason about is therefore the cost per task-relevant, identifiable observation. An observation is identifiable when it preserves enough of the relevant state and context that the latent variable, transition, or mechanism can be inferred without a cascade of heroic assumptions.
“Code was never the hard part” is an insult to all programmers by Senko Rašić
If coding is easy, how come programmers were in high demand, and have demanded large salaries for years (even before ZIRP)? Why was there so much stress, overwork and burnout even before AI started churning out 5000-line PRs?
If coding is easy, why is software so damn buggy?
Whoever you are, don’t outsource your understanding, judgement, empathy and taste to AI. Don’t abdicate your responsibility. Don’t be a meat proxy.
Don’t be a meat proxy by Niklas Gruhn
This is not adding value. I can talk to Claude myself. It’s going to be faster and I get to control the context. I don’t need a meat proxy in between.
Why Is Everyone In Tech So Sad? by Aaron Horwath
What’s different about this moment is that the questions are not just economic but existential, the kind of questions that cause high-earning technical professionals to contemplate throwing it all away to start a goat farm in Washington or become a surf instructor in Costa Rica.
The Quest For Caffeine You Can Have At Night by Scott Alexander
Since the beginning of time, mankind has yearned to drink a cup of coffee at 5 PM and go to sleep at 10. But now there’s an entire scientific subfield and several companies pandering to this absurd fantasy.
If there’s a war on cars in the USA then the cars have won, by Max Mautner
Drivers struck and killed 7,080 people walking in the United States in 2024, 72% more than in 2009, when pedestrian deaths were at their lowest.
I’m Begging You: Never Write With A.I., by Bret Stephens
What, uniquely, does writing do? It compels thought. It compels thinking in ways that silent contemplation or spoken language rarely can. It compels us to subject our thinking to the effort of articulation, the rigor of grammar, the tests of intelligibility and coherence, the inspection of others. In doing so, it also enjoins us to be clear, logical, accurate — and accountable.
Even Elite Colleges Have Become Trade Schools (gift link), by Jessica Grose
Young people who don’t have the resources to pay out of pocket for their college degrees are going to be pushed into jobs that they might be ill suited for, rather than getting to explore their intellectual potential in fields that are not immediately remunerative.
My own thoughts on this one, starting with a deliberately provocative response:
Maybe elite colleges should be trade schools.
The author argues that elite colleges already are trade schools and she treats this as a loss. Her worry is that students get pushed toward jobs they might not be suited for instead of being able to “explore their intellectual potential in fields that are not immediately remunerative.”
For y’all who don’t know me well. I’m a first generation college student from the hollers of a rural county in southern Virginia. We had one stoplight. There are now more MAGA signs than intersections. Epithets like Redneck and Hillbilly were terms of endearment (unless you were an outsider, then them’s fightin’ words). My mother worked in payroll and billing at a local clinic. My father was a machinist and then an HVAC maintenance mechanic at a Westinghouse (later Hitachi) factory that built electrical transformers (a trade whose demand and pay have climbed hard in the few years since he retired). Some days I envy it actually. An honest day’s work with your hands. The job stays at the plant when you clock out. No inbox at 9pm. I love fixing stuff, from cars to bikes to all the weird stuff in the 100-year house I now call home.
Now we weren’t poor by any means. My parents poured every bit of savings they had into me, but it was nowhere close to being able to pay for my college tuition outright. I took out student loans to attend James Madison University (a state public school here in Virginia) and I was saddled with debt that I had to pay down for over a decade after finishing. College was a path to earn a better living than my parents, and their parents before them. And it had to be. I didn’t come from that Old Southern Money that people think about when they think of old Virginia.
Some back of the envelope arithmetic, using public information. The public state university where I work publishes an estimated total cost of attendance for 2026-2027 that runs from roughly $43,000 to $55,000 for in-state students, depending on entering year and school (scroll down and expand any of the Schools). Let’s make it easy and round it to $50,000. Four years comes to about $200,000 for a public state school in-state, and closer to $400,000 from out of state.
Studying whatever interests you, finding your way, exploring your intellectual potential, etc., with no regard for what the labor market will pay… this is a luxury good at this price. An 18 year old kid has no way to price the bet they’ll have to take to finance this with borrowed money, and I’d consider it borderline unethical if I were to advise a student to take on $200-400k in debt for a credential with no plausible repayment path, unless they come from Family Money. Maybe it’d be different back in nineteen-diggity-something when you could buy a house for a nickel and gas was free. But alas, 2026 is different and college ain’t cheap.
None of this is an argument against the humanities and arts. In fact I’d highly encourage anyone headed off to university to spend as much time as you can outside your major. My most memorable and impactful classes I took as an undergraduate weren’t in my STEM major. One was an Eastern Religions course I took in the Department of Philosophy & Religion department. Another was a Human Sexuality and Intimacy course in the Psychology department, taught by one of JMU’s most beloved professors, Uncle Bijan. A course that incidentally I took with my now wife. I had actually planned to go to JMU to double major in music, until I got bitten by a dog in the summer before my freshman year, tearing my lower lip in half, making trumpet a thing I’d never be able to play again.
I wrote in December about the false choice between meaning and accountability in higher education, and I still stand by it.
If you’re going to pay $200-400k for college, that college owes its students both serious intellectual formation and serious vocational preparation. Asking which of the two a given program actually delivers is just basic accountability. You’d prefer to get both, but if you don’t come out with employable skills enabling you to recoup those costs, you’d better come with a trust fund or an athletics scholarship behind you. If finding your way or self-discovery is the only thing that matters to you, there are many other options. Enlist in the military. Walk the Camino de Santiago. Do a few years in the Peace Corps. Hike the AT with the library of congress on your 200-gram e-reader. Coming of age and finding yourself doesn’t have to cost a small fortune.
The author of this essay ends by rooting for the Lehigh graduate she profiles, who described job hunting as riding a bicycle blindfolded. I’m rooting for him too. I’m also rooting for the kid from a backwoods county with one high school who will be the first in her family to hold a bachelor’s degree. She should get the art history seminar and the marketable skill. Both-and. What she can’t afford is a quarter million in student loans and a shrug about what comes after.