0906 | Orbits, Gold, and Cognitive Viruses

Mia: Welcome back to the show, everyone. I'm Mia.

Milo: And I'm Milo. It's been one of those days where the feed reads like the world is quietly rearranging itself — a rocket, a rocket scientist problem in your own head, gold bars on a plane, and a keyboard that glows.

Mia: And underneath all of it, there's a thread I kept noticing: institutions and individuals both trying to hedge against things they can't fully control. Europe hedging on access to space, countries hedging on geopolitics, engineers hedging against their own atrophy. That's today's show. Let's start with the biggest one.

Milo: Isar Aerospace. Their Spectrum rocket reached orbit, launched from Andøya in northern Norway. And the headline fact is that this is the first time a European private firm has put a rocket into orbit from the mainland.

Mia: That's a genuinely historic line, and I want to be careful about what it does and doesn't mean. It doesn't mean Europe suddenly has a SpaceX. It means the existence proof is now there. Someone on this continent, with private money, took a rocket from drawing board to orbit without a national space agency carrying them across the finish line.

Milo: Right, and the comment threads on this split into roughly three camps. The optimists are saying: the bottleneck was never engineering talent in Europe, it was the absence of a company that actually closed the loop. Now there's one. The argument is that the first orbital success is the credential that unlocks everything else — customers, capital, talent.

Mia: The skeptics pushed back hard on that. Their point was: orbit is the entrance exam, not the diploma. A launch company lives or dies on cadence — how reliably, how often, how cheaply you can fly. One successful flight proves the design can work. It says nothing about whether flight two and flight ten work, or whether the production line behind them scales.

Milo: And the third camp, the pragmatists, were basically asking the question everyone should be asking: who signs the contracts? Because a private launcher without institutional customers is a very expensive hobby. European institutional launch demand has historically been spoken for, and the institutional launchers aren't going to hand over their manifest voluntarily.

Mia: That's the real unresolved question, I think. The article gives us the milestone. The discussion gives us the tension: Europe now has a competitive private launcher on paper, but "competitive" only becomes real through cadence and contracts, and neither exists yet. There was also a nice undercurrent in the comments about location — Andøya is way above the Arctic Circle, which is great for certain orbits and adds logistics complexity for everything else.

Milo: And people who've actually been around launch programs kept repeating one thing: the gap between first success and operational service has killed plenty of companies that had a perfect first flight. So temper the champagne, but don't dump it out either.

Mia: Okay, from rockets to something closer to home — your own rocket between the ears. There's an arXiv paper making the rounds that models LLM usage as what the authors call a "cognitive virus."

Milo: I have to admit, the title alone got me. What's the actual model?

Mia: The core idea: the use of these tools spreads socially — you see a colleague leaning on a model, you start leaning on it too — and that social transmission can hit tipping points. Past a certain adoption level in a group, there's lock-in: the assisted way of doing things becomes the default, and the unassisted skill decays. The endpoint the paper worries about is a real loss of cognitive competence at the population or organization level.

Milo: And what made this land for people was that it isn't abstract. There was a writeup on AI tools for site reliability engineering — and the pattern is exactly what the paper predicts. These tools genuinely reduce the time to resolve routine incidents. That's the win, and it's real.

Mia: But the engineers in that discussion were candid about the cost: if every routine incident gets handled by the tool, you stop exercising the muscle. You lose the intuition — the smell of a system that's about to fall over. And then a genuinely complex incident lands, the kind the tool can't pattern-match, and the humans who are supposed to catch it discover they're rusty.

Milo: One commenter pulled out the deep reference here: Bainbridge, 1983, "Ironies of Automation." Forty years ago she described exactly this — automate the easy parts, and the human is left holding only the hard parts, but with less practice than ever at handling them. The automation paradox. We're apparently re-deriving it with language models.

Mia: And that's why the virus framing, provocative as it is, actually did some work in the discussion. A virus doesn't have to be malicious. It just has to spread because it's convenient in the short term. Nobody adopts an LLM workflow thinking "I'd like to lose my skills." They adopt it because it's faster today. The harm, if it comes, is a second-order effect that shows up later.

Milo: There was disagreement, though, and it's worth being fair to it. Some people said this is just the calculator argument again — we offloaded arithmetic and got fine, arguably better, because we offloaded to a tool that's perfectly reliable at the thing we offloaded. Their claim: the problem isn't offloading per se, it's offloading to something that's confidently wrong sometimes.

Mia: And that distinction matters a lot for the SRE case specifically, because an incident responder has to judge whether the tool's answer is trustworthy — and judging that requires exactly the expertise the tool erodes. It's a loop with an unpleasant property.

Milo: Unresolved question from that thread, and I'll carry it with me: what's the equivalent of "deliberate practice" for knowledge work? Do we need to deliberately keep doing things the hard way, the way pilots still hand-fly?

Mia: Which, funny enough, is the perfect bridge to our next one — keeping your codebase clean the deliberate way. There's a new service called Pushin.eu, and it's Git hosting with a very specific pitch: EU-only.

Milo: Concretely: it runs on Scaleway infrastructure in Paris, it blocks what they call AI slop, and — this is the part people zeroed in on — it trains no models on your code. It's invite-only beta right now, and general availability is, at the earliest, 2027.

Mia: So temper expectations accordingly — this is an early thing. But the discussion around it was interesting because it's really a discussion about trust geometry. Post-scoop-era, a lot of developers and especially European organizations started asking: where does my code physically live, whose jurisdiction touches it, and what is it being fed into?

Milo: The skeptics asked the obvious question: does the market need another Git host? GitHub and GitLab are fine. And the counterargument was that "fine" isn't the axis anymore — the axis is sovereignty and data guarantees, and on that axis the incumbents structurally can't make the promises a small EU-only player can. A small player can say "no model training, ever, and here's the whole stack to audit." A giant with an AI product line cannot.

Mia: The other thing that came up in that discussion was hygiene, and it's a genuinely useful tip regardless of where you host. The suggestion: flip your .gitignore logic. Instead of blacklisting the junk, start your .gitignore with a star — ignore everything — and then whitelist the files you actually want with explicit rules.

Milo: Right, so instead of trying to anticipate every junk file, agent artifact, and scratch file that might appear in your tree, you invert it. Nothing gets committed unless you consciously allow it. And in 2025-ish workflows, where AI agents happily drop their own little state files into your repo, that inversion catches a whole category of accidents that a blacklist never will.

Mia: People who've used this pattern noted the tradeoff honestly: it's a bit more friction when you add new files — you have to remember to whitelist them. But the failure mode of forgetting to whitelist is "my file didn't get committed," which is loud and obvious, versus the failure mode of a blacklist, which is "oops, my credentials just went public," which is quiet and catastrophic.

Milo: Asymmetry of errors. I like that framing. And it ties back — the whole sovereign hosting thing is about deliberate choices about what leaves your environment. The .gitignore trick is the same philosophy at file level.

Mia: From code hygiene to planetary hygiene. The UN has now acknowledged, openly, that the 1.5-degree limit will be exceeded.

Milo: That's a big sentence. For a decade the diplomatic position was essentially "we must not cross it." Now the working strategy has a name: overshoot. The idea is you cap the peak — limit how far past 1.5 you go and how long you stay there — and then you come back down, through ending emissions and through actively removing CO2 from the atmosphere.

Mia: The discussion on this was suitably split. One camp said: finally, honesty. Pretending the limit was safe was eroding credibility, and planning for overshoot is just planning for the world we actually inhabit. The other camp said: watch the fine print. The whole strategy leans on carbon removal at a scale that nobody has demonstrated. If removal doesn't materialize at gigaton scale, "overshoot and return" quietly becomes "overshoot and stay."

Milo: That's the load-bearing assumption, and the commenters who know the energy space kept hammering it. The return leg of the journey is currently written in pencil.

Mia: And here's where it gets interesting, because the geopolitical mood around this showed up in two other stories. The Netherlands has repatriated 86 tonnes of its gold — out of 612 tonnes total — from the US and Canada, moving it to the Bank of England. The stated reason: crisis preparedness amid geopolitical unrest.

Milo: Eighty-six tonnes is a lot of plane rides. And the discussion around that was half central-bank logistics, half reading of tea leaves. The practical point: gold stored abroad is gold you have to trust an ally to hand back. In a calmer world, that trust is cheap. In this one, some countries are deciding it's worth the cost and effort to reduce counterparty risk on their reserves.

Mia: Nobody in the discussion claimed to know exactly what the Dutch government knows. But the gesture itself is information — institutions that measure everything decided the hedge was worth buying. And the third data point in this cluster: the Ig-Nobel Prizes, which have been a Boston institution forever, are moving their 2026 ceremony to Zurich. The cited reason: security concerns for international guests traveling to the US.

Milo: Which is the kind of thing that would have sounded absurd a few years ago, and now it's just… logistics. Different organizations, same underlying signal — people are adjusting arrangements because they're less sure about crossing borders and about counterparties generally.

Mia: Which brings us, uncomfortably, to what happens when institutions decide the public doesn't need to know things. The big one: 3M knew, since the 1970s, that PFAS — the "forever chemicals" — were present in the blood of the general population. And they sat on that knowledge for decades.

Milo: And now a former 3M scientist is speaking out publicly about it. That's the part that gave people chills in the discussion — not that a company had data, but that someone held it for forty years and is only now telling us.

Mia: The comment section connected this to a pattern, and it's a well-documented one. From 1963 onward, tobacco companies — R.J. Reynolds, Philip Morris — went on a buying spree of food brands: Hawaiian Punch, Kraft, Nabisco. And the argument, which has real scholarly weight behind it, is that this is where hyperpalatable ultra-processed food comes from. The companies that had perfected engineering a product for maximum addiction transferred that expertise from smoke to snacks.

Milo: So the playbook isn't company-specific, it's almost a genre: internal science shows harm or manipulativeness; the science stays internal; marketing scales the product; the truth comes out decades later, usually via a whistleblower or litigation.

Mia: And there was even a coda in the thread showing the pattern running out of runway: De Beers is closing diamond mines in South Africa because of weak sales — and competition from lab-grown diamonds. Thousands of De Beers workers are losing their jobs. A company that spent a century convincing the world that a diamond is forever and priceless is now watching the price foundation erode under it.

Milo: Someone made the point that all three of these are really stories about manufactured perception. De Beers manufactured desire. Tobacco-food companies manufactured craving. 3M manufactured the absence of doubt. The difference is that two of those stories have already had their reckoning, and the PFAS one is mid-reckoning.

Mia: Unresolved there: what follows from a former scientist speaking out? Historically, testimony plus internal documents is what turns public anger into consequences. We'll see whether that happens here.

Milo: Which flows into our next item nicely, because it's also about opacity — and pushback against it. The Pentagon issued a guideline endorsing testosterone screening, and then pulled it back after one day. No explanation given.

Mia: And the medical community's reaction is the important part: doctors have been warning that routine testosterone screening is actually harmful — you end up testing lots of people, finding borderline numbers, and then treating people who were never sick, with real downstream harms. So the guideline itself was questionable from the start.

Milo: The discussion here was less about testosterone and more about process. A major health guideline appears, vanishes within 24 hours, and nobody says why. One camp said: fine, at least it's gone, good outcome. The other camp said: that's not the point — the fact that it was issued at all, and yanked without explanation, means either the review process is broken or the decision wasn't made on medical grounds in the first place. Both possibilities are bad.

Mia: And the parallel story on workers pushing back against opacity: US staff of the Wikimedia Foundation voted, overwhelmingly, to unionize with CWA. And it's not isolated — there's union movement among Wikimedia staff in the UK and internationally too.

Milo: People who've watched the Wikimedia world for a long time noted the irony gently: an organization whose whole ethos is transparency, and its own staff felt they needed collective bargaining to get a seat at the table. The overwhelming vote margin suggests this wasn't a close call or a fringe effort.

Mia: So on both stories, the theme is the same: decisions made without explanation meet organized resistance. Pentagon guideline, one day, no reason — pulled. Wikimedia staff, no say — unionized. The mechanisms are very different but the frustration is identical.

Milo: Let's lighten up. Hardware oddities. And my favorite of the bunch: the AMD BC-250.

Mia: Explain what this thing even is, because it's gloriously weird.

Milo: It's a cast-off. When Sony builds PS5 consoles, some APUs don't make the cut, and those rejects get turned into mining boards — that's the BC-250. Sixteen gigabytes of GDDR6, basically a PS5 brain on a card, sold for mining. For a while they went for around 60 to 100 dollars, which is absurd for what's inside.

Mia: And the community has been modding them into actual usable machines. The crown jewel demo: getting Cyberpunk running on one. With caveats, of course — it takes real modding, you need Linux, it's not plug-and-play. But the fact that a binning reject from a game console can be coaxed into playing Cyberpunk for under a hundred bucks of silicon is a great story about what enthusiast communities can do.

Milo: The discussion had the usual structure of these threads. The enthusiasts walking people through the mods and the Linux requirement. The realists saying: by the time you account for your hours of tinkering, you could have bought a used GPU. And the philosopher-types saying: that's not the point, the point is the journey and the 16 gigabytes of GDDR6 for the price of lunch.

Mia: Same energy, actually, as two other beautiful oddities. The Terpstra keyboard: an isomorphic layout, meaning the same chord shapes are playable everywhere on the board — move a shape, it stays a shape. It has 280 continuous controllers that shift color as you use them. Price of admission: around five thousand dollars.

Milo: Five grand for a keyboard, and the people who own them apparently love them the way people love grand pianos. Isomorphism is genuinely deep for music theory folks — learn a fingering once, it works in every key. But it's a five-thousand-dollar bet that your hands will relearn everything.

Mia: And going further back: Mimic Systems announced the Spartan in 1984 — an Apple II+ emulator that ran on the Commodore 64. Yes, an Apple emulator inside a Commodore. It got announced in '84 and didn't actually ship until 1986, which anyone who's followed hardware vaporware will recognize as a timeless tradition.

Milo: And the marketing! Bizarre mime-themed advertisements. Mimes. For an emulator. Somehow that was the pitch. And today the thing is a collector's item — people hunt these down.

Mia: There's something sweet in the thread's observation that the BC-250 tinkerers of today are the spiritual descendants of the people who bought a Spartan in 1986. Hardware that wasn't supposed to do what you want it to do, and you make it do it anyway.

Milo: Which actually brings us to our closer, and it's a quiet one. Brad Feld turned 60 and wrote about the real luxury goods. Not watches, not cars. His list: time. Health. A quiet mind. Slow mornings. And people who love you.

Mia: And the discussion around it was one of the warmest threads of the day. People sharing which item on that list they'd underestimated. The recurring answer was the quiet mind — because you can buy time and squander it, but a quiet mind is the thing that lets you actually use the time you carve out. Several people said their slow mornings were the single highest-leverage habit change they'd ever made, and it cost nothing.

Milo: And it pairs beautifully with one last item: a complete book, "Learn Programming with OCaml," by Conchon and Fillière, released under a Creative Commons BY-SA license, free to download as PDF and EPUB.

Mia: A full, rigorous programming education, free, forever, legally yours to share and remix. Between that and a community turning mining rejects into gaming machines, you remember that a lot of the best stuff in this ecosystem was built by people who weren't optimizing for money.

Milo: Which is, I suppose, where we came in this episode: rockets, chemicals, gold, keyboards — most of what matters turns out to cost time, not dollars.

Mia: Feld's list is basically our whole show in five lines. Thanks for listening, everyone — we'll be back tomorrow with whatever the feed throws at us.

Milo: Until then: guard your slow mornings. They're worth more than the keyboard.