Nietzsche said when you gaze long into an abyss, the abyss gazes back into you. He just couldn't predict that the abyss would be five inches wide, made of glass, and refreshing every three seconds.

For those of you railing about the price of gas: The oil companies make between ten and fifteen cents per gallon of gas. That means if they decided to get zero profits, you'd save on average twelve cents per gallon. State and federal governments make about 80 cents per gallon in taxes on gasoline. If they gave a temporary gas tax moratorium, you'd see a significant reduction in the price of gas and diesel.

I'm seeing a lot of euphoria about how Opus 5.5 is good at TLA+, and this means that all software will soon be formally verified. As a person who loves TLA+ so much he wrote a book on it, I want to throw a particular cold shower on people's enthusiasm by talking about the limits of what you can actually verified with it. The high level simplification is that TLA+ sees a system as a set of "behaviors", or possible sequences of states. For example, the pseudocode "pick a random number from 1-3 and decrement it to 1" has three behaviors: `{3 -> 2 -> 1, 2 -> 1, 1}`. From here, there are two basic kinds of TLA+ properties: - `[]P` means that `P` is true in *all states* of *every behavior*. - `<>P` means that `P` is true in *at least one state* of *every behavior*. `[]P` is immediately useful as an **invariant**, or something that always be true of your system. This is things like "your data is never corrupt" or "there's always at least one server online." `<>P` is a little more abstract, but for technical math reasons I won't get into here, can be stacked with `[]` to create really complex and useful properties. `<>[]P` represents things like "the algorithm eventually converges on the right answer", `[]<>P` things like "if two data stores desync, they will eventually resync", and `[](P => <>Q)` things like "If a message is put on the queue, it's eventually processed by a worker". Really cool stuff! These primitives were chosen to make a wide array of properties useful. And if we're clever, we can do all sorts of more complex properties, like bounded time constraints and history properties. But we're always constrained to 1) define a logical formula 2) over individual behaviors, and 3) check that all behaviors satisfy that formula. So some things that we *cannot* express in TLA+: - Possibility and reachability properties: that it's always possible to *make* P true, even if you don't actually decide to. Things like "I can always shut down the computer" or "A user can always change their password". These can't be expressed with `<>P` because that's "for all behaviors, P happens at least once", we actually want "for all behavior prefixes, there is at least one behavior where P happens at least once". - Hyperproperties: properties that are defined over two or more traces. These are things like "painting a car red doesn't make it faster" or "users cannot infer secret data by observing public data". We can't do these because TLA+ only looks at one behavior at a time. - Statistical properties: 95% latency is 1ms. Impossible because most of these are hyperproperties. - Properties about if a system is robust against code changes. Impossible because, uh, you have new behaviors now. Some of these are solvable in different logical formalisms. CTL can do reachability, PRISM can do statistical properties, etc. Those have their own tradeoffs and limitations, though, and no system can do everything. Others are solvable with a lot of cleverness tailored to the specific spec, like lifting a model into a hypermodel. But these are insanely inefficient and make your "clever spec" diverge significantly from the real world system, so introduce a lot more opportunity for things to go wrong. The core problem, though, is (1): properties are logical formula. If we don't know how to express a system property as a logical formula, we can't verify it. 99% of the properties we care about fall under this. The information on the site is easy for a user to find. Our LLMs behave as we expect them to. Our application can't be used to break the law. TLA+ (and Quint and Lean and Rocq) are near-useless here, no matter how clever you are. Don't get me wrong: `[]P` and `<>P` represent a huge range of useful properties and TLA+ is incredible at finding awful concurrency bugs. But there's a lot it fundamentally can't do and we shouldn't believe that it will solve all our worries about software bugs. And the same goes for all other formal verification languages, too.

I was one of those nerds. You've missed the way I'm adapting to AI, which is by moving *up* the stack rather than down. My job is design and specification now, not coding. I was a hand-coder for nearly 50 years. I find I miss it surprisingly little. I guess I was a designer after all.

“Gas is expensive, so I’m going to vote for the party that shut down Keystone XL, opposes combustion vehicles, and pushed to ban fracking.” Did you people eat glue in school?

I've been a Steely Dan fan for 50 years and I am left utterly without words. It's a beer jingle, and yet...the voices, rhe instrumental work, the Dan style is undeniably recognizable. Weird, weird, weird!

I expected the Trump tariffs to be a disaster. I'm puzzled by what I'm not seeing. They don't seem to have harmed the US economy in any way I can notice with my eyeballs or in the statistics. Business activity is booming, unemployment seems to be low, household consumption is up, stock market seems to be doing fine. Is my whole model of trade and economics broken? How is he getting away with this?

Unfortunately there isn't a better alternative yet. But @gmkurtzer is working on building one. Support him any way you can.

There's a lot of this going around lately. It's a study in incentives. In the short term Linux UI development mostly stumbles around like a drunk chasing shiny things, because most of the people doing it don't have a freaking clue about user-experience design. Still, over the long term, there's selective pressure to get better, mostly in the form of competition among Linux distributions. Omarchy and COSMIC are two recent examples of how this can produce startlingly good results. Windows has the opposite problem - all its incentives push in the direction of worse. What Microsoft wants more than anything else is increased lock-in and increased revenue capture; neither of these is good for users. An underappreciated problem is that there isn't any competition in Microsoft's UI space. When its one UX team makes a bad decision, there isn't any alternate Windows UI users can defect to to express their displeasure. Not and stay current with security fixes anyway. Over the last 30 years, most of us hoping for a Year of the Linux Desktop thought it would eventually happen because Linux got better at UI design and stability. That did happen, but it turned out to be insufficient. We're seeing posts like this now not so much because Linux has been improving, but because Windows has been rapidly enshittifying itself. Ads in the start menu. Account requirements. Constant OneDrive upsells. Intrusive AI crammed in everywhere. Sneaky telemetry and data theft. Mandatory updates that interrupt your work or your game. Linux will never have these problems. Microsoft can never be rid of them because line must go up. And that's what's going to turn out to be the dealbreaker.

This is a detailed exposition, by somebody who understands corporate finance and debt leverage much better than I do, of the argument I've been making on here that the current AI build-out is unsustainable and guarantees a fairly near-term crash. I saw this coming because when you estimate the training costs for frontier models it immediately becomes obvious why the big AI providers only claim to be profitable relative to inference costs, not training and build-out costs. The technology is sound, but the investment assumptions are obviously crazy. This guy supplies the detailed mechanics and the dates. groundbrkr.com/p/the-teaser-…

Why we KNOW fossil fuels played a completely negligible role in the rise of atmospheric CO₂ in plain English 1. Carbon has a fingerprint. Carbon atoms come in two stable kinds. One is a little heavier than the other. Every source of carbon on Earth has its own mix of heavy and light atoms, and scientists measure that mix as a single number. Think of it as a fingerprint. Coal, oil, and gas have a fingerprint of minus 28. Carbon dioxide bubbling out of the ocean has a fingerprint of minus 8. Volcanoes are minus 5. Dissolving limestone is 0. A lower number means lighter carbon. These fingerprints are measured facts, and the IPCC uses the same numbers. 2. You can read the fingerprint of whatever is being added to the air. When carbon is added to the atmosphere, the atmosphere's fingerprint drifts toward the fingerprint of what was added. In 1958 Charles Keeling showed a simple way to read off the fingerprint of the added carbon directly from the air measurements. It is a graph. Plot the air's fingerprint against the amount of CO₂ in the air in a certain way and you get a straight line. Where that line meets the edge of the graph is the fingerprint of the net addition. This is not a computer model. It is bookkeeping. You count the heavy atoms, you count the light atoms, and the answer follows. Nothing about oceans, forests, or weather goes into it. 3. The fingerprint of the added carbon is minus 13. Koutsoyiannis (2024) did this with the four Scripps monitoring stations in Alaska, California, Hawaii, and the South Pole, using records from about 1978 to today. Every station gives the same answer: minus 13.2, give or take 0.2. These are the same records the IPCC relies on. Two completely separate measurements say the same thing. A Caribbean sponge that grows in yearly layers, like tree rings, records ocean carbon back to about the year 1500. Its fingerprint is about minus 13 for all five centuries, with no drift. Air trapped in Antarctic ice at Law Dome, analyzed by Köhler et al. (2006), also gives minus 13. Three different materials, three different laboratories, five hundred years, one number. 4. Simple mixing math sets the range from the modern data alone. Every carbon atom added to the air came either from fossil-fuel combustion or from some non-fossil-fuel source: the ocean, volcanoes, rock weathering, plants and soil. The fingerprint of the mix tells you how much came from each. It works like mixing paint. If you mix black and white and get a very light gray, you know there was very little black. Fossil-fuel carbon is the black, at minus 28. The ocean is the white, at minus 8. The measured mix is minus 13, much closer to the ocean than to fossil fuel. The modern record from 1978 to today does not by itself pin down the exact fingerprint of the non-fossil-fuel carbon. It could be anywhere from minus 8 (pure ocean) down to minus 13 (matching the mix exactly). Run the mixing math across that whole range and the fossil-fuel share comes out somewhere between zero and 26 percent. That is what we know from the modern data alone: not more than 26 percent, possibly zero. The exact answer comes in step 6. 5. The official claim is impossible under any assumption. The IPCC, NASA, NOAA, EPA, and Engelbeen, Hannon, and Burton (2024) say fossil-fuel combustion caused essentially all of the rise. That is close to 100 percent. "Dominant" by any normal meaning is at least 80 percent. Run the mixing math backwards and ask what the non-fossil-fuel carbon would have to look like for each of those claims to be true: - For fossil fuel to be even a bare 50 percent, the non-fossil-fuel carbon would need a fingerprint of plus 1.6. No carbon source on Earth is that heavy. - For 80 percent, it would need a fingerprint of plus 46. That is more than 20 times heavier than the heaviest source that exists. - For 90 percent, it would need plus 120, about 60 times heavier than anything on Earth. - For 100 percent, it would need to be infinitely heavy. The official claims do not fall a little short. They require carbon sources that do not exist. 6. Five hundred years of no change gives the exact answer: essentially ZERO (0% fossil-fuel sources). The modern data left one unknown: the fingerprint of the non-fossil-fuel carbon. The sponge and ice-core records supply it. In the year 1500 there was essentially no fossil-fuel burning, so the fingerprint of the added carbon then, minus 13, came entirely from non-fossil-fuel sources. That fixes the unknown at minus 13. Since 1500, fossil-fuel burning has grown about ten thousand times over. If fossil-fuel carbon were any real part of the added carbon, the fingerprint would have slid toward minus 28 as burning grew. It did not move. It is still minus 13 today. Put minus 13 for the non-fossil-fuel carbon and minus 13.2 for the modern mix into the mixing math and fossil fuel's share of the added carbon comes out at about 1 percent, which is zero within the measurement error. The range of zero to 26 percent from step 4 collapses to a single answer: essentially zero. 7. Why this is knowledge, not opinion. The conclusion rests on three things only: - counting atoms, - the standard measured fingerprints of fossil-fuel carbon and the other sources, and - the Scripps, sponge, and ice-core records. The IPCC uses all three. Anyone who accepts all three and rejects the conclusion is refusing to do arithmetic. The only way out is to throw away one of the three, and each one is a foundation of the IPCC's own work, and nobody has done so anywhere in the peer-reviewed literature. 8. What is actually happening. Burning fossil fuel does put CO₂ into the air. That is measured at the smokestack. But a CO₂ molecule stays in the air only about four years on average before the ocean or plants take it back. So fossil-fuel CO₂ never piles up. Add up all fossil-fuel emissions since 1850 with that four-year turnover and only about 4 percent of the carbon in today's air came from fossil fuel. The growth in atmospheric CO₂ since the industrial revolution came from non-fossil-fuel sources. Fossil fuel's share of that growth is essentially zero.

My AI doom scenario: AI is successfully captured by governments. It is good enough to be extremely useful to them in maintaining control of their populations, but not good enough to bring about radical innovation and transformative growth. It is kept in this state intentionally, since radical change is a threat to the incumbent power structure. All compute is centrally monitored so no one can advance the field further. Dissent becomes functionally impossible due to AI-powered censorship. Given the politics of the people who want to capture AI, this means further mass migration, degrowth climate policies, an ever-increasing regulatory burden, and ubiquitous anti-White discrimination in the name of equity. To a doomer, even a right wing doomer, this would be a relatively good outcome because we haven’t all been gunned down by terminators. I don’t think it’s a good outcome and I want to prevent it. The world is, and has always been, a race between the chaotic forces of growth and the centralizing forces of control and stability. I want growth to win. AI has the potential to shatter existing global power structures and allow something new to grow in their place. Speed of change is the only defense against parasitism, control, decay. Change so fast that they can’t keep up. AI will either end the evil dream of centralized control once and for all, or, if the controllers manage to capture AI, cement it for all time. Hence, acceleration. I want more open source, more competition, more diffusion. I want us to print so much compute that everyone can run frontier models on their desktops. I want power too cheap to meter. I want the scales to finally and decisively tip towards explosive chaotic growth and end the possibility of closure forever.

I'm shocked, *shocked* to discover gambling going on going on in this establishment.

An Mi-8 is a Soviet-era helicopter (NATO code name "Hip") used both as a gunship and for troop insertions. Everybody knows how this game is played. You keep doing these little feints to train your opponent into under-reacting when the actual punch comes. The accelerating tempo of Russian airspace incursions is another warning sign that they may be about to bust a move against Narva, or possibly in the Suwalki Gap.

It would take a heart of stone not to laugh.

Well. Everything old is new again, including false-flag operations by thug regimes. The Baltic states have now made public intelligence reports indicating that they think the Russians are planning a false-flag operation, by Russian special forces infiltrated across the river Narva and wearing Estonian uniforms, against Russian conventional troops on the other side. The more historically literate among you may remember that World War II actually started with the Gleiwitz incident, Nazis faking a Polish attack on a German radio station. This is a clever move on Putin's port. It's going to greatly increase the pressure on the Estonians not to respond with military force when the conventional Russian military comes over the border and does whatever it's ordered to do, which is most likely seizing the city of Narva.

I see that the Institute for the Study of War is thinking along the same lines I was yesterday about a Russian seizure of the city of Narva. Not surprising. It's a pretty obvious scenario. understandingwar.org/researc…

The first and second are best friends because the first gets us the truth and the second defends it.

We've seen the playbook that Vladimir Putin is running on the Estonian border before. If there is military action there - and there might not be, he might still be successfully restrained by allies inside the Russian government - here is how it will go... First, a Russian separatist movement will magically appear in the 87% ethnic Russian city of Narva. It will of course be run by FSB agents, attracting mainly criminals and the local underclass. There will be street demonstrations and minor riots. There's an old Communist trick where you plant agents among peaceful demonstrators to fire handguns at the police, who then "overreact", firing into crowds and killing civilians. If you are lucky, this will cause genuine popular outrage among the Russophones. But you don't need that, you just need to be able to organize a massive propaganda blitz about how Russophones are being oppressed by the fascist Estonian government. Russia will get some help here from history. There was significant Estonian collaboration with Nazis during World War II, including active participation in Einsatzgruppe A and formation of a first Estonian SS division. Putin's propaganda apparatus will honk about Estonians being Nazis and NATO being a neo-Nazi puppet organization. Then they will stage troops across the River Narva and occupy the city. Not a lot of troops, because they don't expect to have to fight an army. After which they will not move any further into Estonia. Not yet. Instead, Russia will present the world with a fait accompli. Do you really want to go to war with a nuclear superpower over one border city full of ethnic Russians? I can't tell you what will happen after that. Putin clearly believes NATO will crawfish, and in doing so destroy itself as an effective military alliance. He may not be wrong.