HN user
In 1955, Mary Tsingou helped run a computational experiment at Los Alamos that revealed unexpected behavior in nonlinear systems—work that later became foundational to chaos theory and computational science.
This piece traces that history from early supercomputers and the Fermi-Pasta-Ulam-Tsingou problem to modern AI systems used to explore uncertainty in scientific research.
Neutrinos were originally proposed as a “desperate remedy” to fix missing energy in nuclear decay, but turned out to be real—just incredibly hard to detect. Tens of quadrillions can pass through matter before one interacts. Los Alamos physicists (Reines and Cowan) confirmed their existence in 1956 using a nuclear reactor as a source. Since then, neutrino experiments have repeatedly exposed gaps in the Standard Model—like neutrino oscillations, which proved they have mass, and ongoing hints at possible “sterile” neutrinos. What’s interesting is how detection capability drove the science: better detectors led to unexpected anomalies which led to new physics. Today, neutrinos are used as probes of everything from stellar processes to matter–antimatter asymmetry, and experiments are still chasing open questions like whether neutrinos are their own antiparticles.
Story about how Los Alamos is reinventing neutron imaging — think “x-rays with neutrons” — using new event-mode cameras that record each neutron interaction with nanosecond-level precision.
Traditional neutron imaging works like a long-exposure photo: useful, but blurry. The new system (based on a camera called LumaCam) timestamps every photon from neutron events and uses techniques like event centroiding and pulse-shape discrimination to clean up noise and sharpen images dramatically.
Why neutrons? Because they reveal things x-rays can’t — like light elements, isotopic composition, and crystal structure. This is especially powerful for materials science and nuclear diagnostics, where understanding what something is made of (not just where it is) really matters.
Article from Los Alamos is about the ICE House, a facility where they test electronics by blasting them with neutrons that mimic what you’d get flying at 35,000 feet for decades. One hour of testing = 100 years of cosmic radiation.
It’s part of a larger effort to make electronics rad-hard — so that microchips don’t randomly glitch or die in flight (or in orbit). Especially relevant as chips shrink and transistor counts hit hundreds of billions (i.e. more chances for failure).
Some highlights:
Neutrons from space can flip bits or cause “latch-ups” (think: permanent short circuits).
These upsets can lead to weird bugs, BSODs, or worse — especially at altitude.
The ICE House runs ~24/7 and still can’t keep up with demand from avionics and chip companies.
They’re now planning a third beamline to expand testing capacity, and even working on proton-based testing for space use cases.
If you’re into hardware reliability, aerospace, or just cosmic-ray horror stories for computers, this is worth the read.
LANL researchers are cataloguing the molecules in healthy human breath to create a baseline profile that could enable new non-invasive diagnostics.
Some details:
Method: Using tandem mass spectrometry, the team has identified 227 distinct compounds across 31 volunteers, narrowing to 48 common features that may define “healthy breath.”
Patterns: Certain metabolites correlated with sex and time of day; others trace back to environmental contaminants or microbiome interactions.
Partnerships: Collaborating with the University of New Mexico to expand sampling (including both breath and blood data).
Goal: Easy-to-use diagnostic tests where a patient might one day “just breathe” to screen for illness, fatigue, or impairment.
The approach echoes the original breathalyzer’s leap in the 1950s but applies it to a far wider range of health conditions.
Los Alamos's proton radiography (pRad) facility uses beams of high-energy protons to capture images of dynamic, high-density events — think explosions. A clear, animated explainer of how it works.
Chaos theory gave us the butterfly effect: tiny changes that balloon into massive consequences (Lorenz’s weather simulations, or Bradbury’s time-travel butterfly). But Los Alamos researchers just showed that quantum systems don’t always play by those rules.
Using theory, simulations, and IBM’s quantum processors, physicists explored whether small quantum-level disruptions would spiral out of control over time. The result? At the quantum scale, entanglement actually heals damage. A particle “sent back in time” and deliberately altered can return to the present nearly unchanged.
In other words:
Lorenz-style chaos does exist at the quantum level (slight variations can diverge wildly). But there’s also a quantum anti-butterfly effect: in sufficiently entangled systems, information “damaged” in the past can be restored in the present. This has direct implications for quantum computing (a new way to measure “how quantum” a computer really is) and potential applications in information security and error correction. As lead scientist Bin Yan put it: “At the quantum scale, reality is self-healing.”
If the writing does the job it needs to do--in this case, a deft summary of an article--why is it better if it comes from a human vs. AI? Analysis, sure. But summary? This is the whole point of the article...do you actually prefer to read bad writing because it was written by a real person?
The class of 2026 has had generative AI for their entire college career. What started as a novelty in 2022 has become second nature: surveys show >90% of undergrads now use AI for schoolwork, from drafting essays to summarizing readings.
For students, the motivation is pragmatic: AI saves time, reduces stress, and helps balance overwhelming academic and extracurricular demands. It’s less about “cheating” and more about survival in a system that prizes productivity and credentials. Professors, meanwhile, are scrambling—reverting to handwritten exams, shifting grading toward tests, or trying moral appeals. Yet many remain unaware of just how normalized AI has become on campus.
The result: higher ed has been fundamentally reshaped in just three years. Students expect project-based, real-world assignments that resist AI shortcuts. But with faculty stretched thin by budget cuts, research demands, and political headwinds, systemic redesign feels unlikely. For now, both students and professors face the same reality: a college education is what you make of it—AI included.
If you're wondering--yes, I used AI for the synopsis. Big question for me, is what does the future of education look like? How do kids get the skills they need to use AI, while still getting the skills they need to be skeptical of it?
Original discussion here: https://news.ycombinator.com/item?id=44798695
Discussion continues, with more on dynamic imaging and the labs, here: https://news.ycombinator.com/item?id=44876919
Following last week’s discussion about LANSCE and dynamic imaging at the national labs, here’s a deeper look at a sibling system: pRad (proton radiography).
Los Alamos just ran Pagoda, an experiment probing why some detonations fail, using pRad—one of the few facilities anywhere that scientists can image high explosives in billionths of a second using near-light-speed protons (not x-rays).
Built after nuclear testing ended, pRad feeds critical data into stockpile certification models. But the system is aging. After 25 years and nearly 1000 experiments, it’s finally getting a major upgrade—Cold War hardware out, throughput doubling, and plutonium capability returning.
This is the kind of highly specialized, high-accountability work that likely helps keep DOGE out of the weapons side of the national labs.
Full story (great visuals, including images of the explosion at the bottom): https://www.lanl.gov/media/publications/1663/prad-future-sto...
I just don’t get this. Anybody want to take a shot at explaining how this kind of private/public profit-sharing fits into the vision of the capitalist-in-chief?
Nvidia and AMD are reportedly handing over 15% of their China-bound chip revenues (H20 and MI308, respectively) to the U.S. government in exchange for export licenses that had previously been denied on national security grounds. No word yet on what the government plans to do with the money.
The deal effectively clears the way for billions in chip sales to China, despite earlier restrictions—and sets a pretty wild precedent for direct federal revenue participation in corporate exports. Markets didn’t exactly cheer: NVDA and AMD both dipped slightly in premarket.
The executive order, titled "Improving Oversight of Federal Grantmaking", grants political appointees sweeping authority over all federal grant funding.
If implemented, political appointees — not scientists — would take control over decisions about research grants. It would, among other changes, allow political appointees to overrule advice from scientists on award decisions, and let them terminate ongoing grants based on political criteria.
Researchers may have discovered a biochemical "vector" for aging. A new study in Metabolism shows that a DNA-binding protein called HMGB1, when released by aging or stressed cells, can induce senescence in nearby healthy cells—essentially spreading aging. The reduced (low-oxygen) form of HMGB1 triggered this effect in both cell cultures and live mice; oxidized HMGB1 did not. Mice injected with the reduced protein began to show signs of aging within a week. The findings suggest that aging may be transmissible at a cellular level—at least in part—through circulating proteins in the blood.
Makes me wonder if this is responsible for the effect of rapid aging at (around) 44 and 60.
I guess my point is that it's hard to simply cut research that's essential for certifying that the stockpile is safe and works. I'll avoid making any predictions, because who the hell knows what's going to happen, but I think dynamic imaging work may prove a tough target for DOGE.
For sure. But depending on what Congress does, think defense budgets could grow, which would mean more money for defense-positioned Labs like Los Alamos.
I worry about this, but these capabilities are hard to replace. This kind of research hasn’t historically been something you can outsource to private companies. Or—at least—it hasn’t been until now. Even if this administration wants to open that door, the infrastructure investment required for the accelerators alone is staggering: easily in the multiple billions.
depending on the experiment, could be unnatural causes.
Fascinating look into the dynamic imaging capabilities at Los Alamos National Lab—essentially, how the U.S. is able to analyze nuclear-level explosive events without actually conducting nuclear tests.
The Lab uses multiple systems to image these high-speed events:
• pRad uses proton radiography to get 20–40 frames of a detonation, with material-level resolution based on density.
• DARHT uses dual-axis x-ray imaging to create 3D snapshots from two angles, ideal for testing whether the computational models built from pRad hold up.
• Scorpius (in development) will take this a step further by using subcritical plutonium in a new accelerator at NNSS, capturing multiple high-resolution frames just nanoseconds apart.
The fact that they can tailor experiments based on frame-by-frame behavior of individual materials under explosive stress feels like the real-world version of “bullet time” physics modeling. The margins of error come down to billionths of a second.
So does helium hydride just not exist naturally anymore? Wondering why they had to recreate it.
Is it worth it?
the sad truth
My bad--thank you
Doesn't include Microsoft's latest announcement, but still useful
AI: like infecting the workplace with ADD.
"The work offloaded to AI in each task tended to be the most cognitively demanding aspects of those tasks, which are also normally the most rewarding. When critical thinking becomes automated, the quality of the outputs might noticeably improve, but the sudden return to critical thinking produces a kind of emotional whiplash that leaves workers feeling sapped of motivation."
Seems like the wide-spread adoption of AI will demand that workers find value and motivation in other aspects of their work. What's that gonna be? The sheer volume of stuff you can get done? Feels like the rat race just keeps getting faster...
Country specific AI in action.