Let’s say you wanted an offroad vehicle which rather than being a car-shaped cowboy hat was actually useful for camping. How would it be configured?

The way people really into camping approach the process is very strange to normal people and does a negative job of marketing it. You drive to the campground in a perfectly good piece of shelter and then pitch a tent. Normal people aren’t there to rough it, they’re there to enjoy nature, and sleeping in one’s car is a much more reasonable approach.

To that end a camper vehicle should have built-in insulation, motorized roll-up window covers, and fold-up rear seats. You drive to the campground, press the button for privacy on the windows, fold up the seats, and bam, you’re all set.

It should have a big electric battery with range extender optimized for charging overnight. The waste heat during the charging process can keep the vehicle warm while you sleep in it.

Roughly 8 inch elevation off the ground and a compliant suspension designed for comfort on poorly maintained roads rather than feeling sporty.

Compact hatchback form with boxy styling. Hatchbacks are already boxy to begin with and a flat front windshield works well with window covers so it’s both functional and matches the aesthetics.

Available modular fridge, induction plate, and water heater. With custom connectors to the car’s battery the electric cooking elements could ironically be vastly better than the ones in your kitchen.

Unfortunately having a built-in shower or toilet is impossible in a compact but the above features might be enough to make it qualify as a camper van which you’re allowed to live in. They’d at least make it practical to inconspicuously live in one’s car and shower at a gym.

Subscribe now

Leela Odds is a superhuman chess AI designed to beat humans despite ludicrous odds. I’m a decent player and struggle to beat it with two extra rooks. It’s fun doing this for sheer entertainment value. Leela odds plays like the most obnoxious troll club player you’ve ever run into, more like a street hustler than something superhuman. Obviously getting beaten in this way is also humiliating, but it also seems to teach a lot about playing principled chess, in a way which raises questions about objectivity, free will, and teaching pedagogy.

Most computer chess evaluations suffer from being deeply irrelevant to human play. When decently strong humans review games with computer evaluation as reference they talk about ‘computer lines’, meaning insane tactics which no human would ever see and probably wouldn’t be a good idea for you to play in that position even after having been told the tactics work out for you in the end, much less apply to your more general chess understanding. There’s also the problem that the only truly objective evaluation of a chess position is one of three values: win, lose, or draw. One move is only truly better or worse than another when it crosses one of those thresholds. If a chess engine is strong enough it can tell that a bunch of different moves are all the same and plays one of them at random. Current engines already do that for what appear to be highly tactical positions which are objectively dead drawn. The only reason their play bears any resemblance to normal in those positions is they follow the tiebreak rule of playing whichever move looked best before they searched deeply enough to realize all moves are equivalent

Subscribe now

So there’s the issue: When a computer gives an evaluation, it isn’t something truly objective or useful, it’s an evaluation of its chances of winning in the given position against an opponent of equal superhuman strength. But what you care about is something more nuanced: What is the best move for me, at my current playing strength, to play against my opponent, with their playing strength? That is a question which has a more objective answer. Both you and your opponent have a probability distribution of what moves you’ll play in each position, so across many playouts of the same position you have some chance of winning.

This is the reality which Leela Odds already acknowledges. Technically it’s only looking at ‘perfect’ play for its own side, but in heavy odds situations like it’s playing the objectively best moves are barely affected by the disadvantaged side’s strength anyway because the only way a weaker player can win is to get lucky by happening to play nearly perfect moves. And here we’re led to what I think is the best heuristic anyone has ever come up with for how to play good, principled, practically winning chess: You should play the move which Leela Odds thinks makes its chances against you the worst. The version of you playing right now has free will can look ahead and work out tactics but the version of you playing in the future cannot and is limited to working out tactics with only some probability of success. You can learn from advice from the bot about what are the most principled chess moves which give you the best practical chances assuming the rest of the game will be played out by your own not free will having self. Everybody has free will but nobody can prove it to anybody else, not even themselves in the past or the future. The realization that your own mental processes are simply a probability distribution does not give you license to sit around having a diet of nothing but chocolate cake and scrolling on your phone all day while you wait for your own brain to kick in and change your behavior.

Philosophical rant aside, this suggests a very actionable thing for making a better chess tutor: You should be told Leela Odds’s evaluation of all available moves so you can pick out the best one. The scale here is a bit weird. In an even position it will say things like your chances of winning in this position are one in ten quadrillion but if you play this particular move it improves to one in a quadrillion. But the relative values do mean a lot and greater ratios mean more so some reasonable interface could be put on it. I haven’t worked out what that interface might be. This approach may break down in a situation where you’re in an objectively lost position instead of an objectively won one and you should be playing tricky troll moves yourself. That seems to matter less than you might think, and could be counteracted by reverting to a weaker version of Leela Odds which can’t work out the entire rest of the game once it gets into such a position.

So far no one is building this. Everybody uses Stockfish for evaluation, which suggests a lot of lines you could have played if you were it, but of course you’re not, and is overly dismissive of alternative lines that would have been perfectly fine against your actual non-superhuman opponent. Somebody should build this. In the meantime if you want to improve your chess you’re stuck getting humiliated by Leela Odds even when you’re in what seem to be impossible to lose situations.

Subscribe now

SR-17018 is a novel drug which is getting increasing underground usage for quitting opioids. It is technically an opioid itself but produces an amount of euphoria which is somewhere between barely noticeable and completely nonexistent. While taking it people don’t get withdrawal symptoms from Fentanyl but their Fentanyl tolerance fades at about the same rate as if they were going cold turkey without the SR-17018. People have been successfully using it to quit opioid addictions and even keeping a stash of it around in case they relapse, which is bizarre behavior for usually addicts. Usually if there are any opioids around they’ll take them and it will cause a relapse, so this stuff must really not be much fun or addictive. Opioids for opioid quitting has a bad reputation because of Methadone, but swapping Buprenorphine for Fentanyl is a big improvement and SR-17018 seems to be truly good for cessation. Unfortunately because it’s technically an opioid and there hasn’t been any movement on getting it approved for cessation purposes (it was originally studied as a painkiller which it’s unsurprisingly not very good at) most likely it will get shoved into schedule 1 at some point, sanity and reality be damned.

Subscribe now

Varenilicline is a good smoking cessation drug but causes nausea in some people. The obvious fix would be to give patients Ondansetron with it. This has been suggested but doesn’t seem to have been tried, not even a case study. There seem to be two problems here: The drugs in question are generic and there’s no incentive to develop treatment improvements which are very cheap, and there’s a general view that any treatment of addiction is super scary and the patients should have to suffer, even for fairly safe drugs with no reason to think they’ll have a bad interaction.

Sodium Oxybate is about to get orphan drug status, for the second time, for the same drug, which is already making more than a billion dollars a year and was neither discovered nor characterized by the company which got the orphan drug status the first time. Pharma has the deeply broken structure that exclusivity periods are the only form of reward for research but a start to fixing it would be to make it that formulation changes are both much easier to get through and give much less exclusivity. A bare minimum start to that would be to clarify that orphan drug status was never meant to apply to formulation changes. It would also be good to make sectors which are already making massive profits not qualify as orphan any more and to reduce the exclusivity period for formulation patents in general, with time release formulas and salt changes handled as specific special cases.

Subscribe now

*Criticism of Benjamin Netanyahu may be an offense under Australia’s new hate speech laws, Greens warn.*

An expert said, *It seems that the implication is criticism of Israel, and the Israeli government, and suggesting it is engaged in genocide or something of that kind, would be enough to at least trigger the start of the process, by satisfying the provisions about inciting racial hatred.* But whether a court would see that as permitted

I oppose the prohibition of so-called "hate speech" because it tends to result in censorship of legitimate political condemnation of politicians that practice of advocate cruel and violent policies — in effect, defending that cruelty and violence, as in the case of Netanyahu.

ChatGPT is worse than lacking in intelligence — its store of text comes from Grok, which spouts Nazi propaganda.

One of Gavin Newsom's flaws is that he opposes the billionaire wealth tax.

While the design of California's one-time wealth tax proposal is clever in that a billionaire can't save any money by moving out of the state now, I don't think a once-time wealth tax will do what is needed to make the US start working again. Billionaires' under-taxation has made many aspects of the US work badly.

The autopsy on Geraldo Lunas Campos ruled that he was killed by the thugs while a prisoner in a deportation prison.

The Department of Hatred and Suffering publicly claimed that Mahmoud Khalil and other foreign students were "supporting terrorism", but internally admitted that they were only advocating Palestinians' human rights.

Sweden, pressured by a right-wing extremist party, is kicking out asylum seekers who have been in Sweden for over a decade and have fully integrated into Swedish society.

I wonder how serious a crime must be to serve as grounds for expulsion.

Syrian government and mostly-Kurdish Northeast Syria have agreed on a one-month cease fire for sending PISSI prisoners to Iraqi prisons.

This is good as far as it goes, but the Syrian government should allow people in Northeast Syria should not be stripped of the rights and social advances that they have fought for.

The terror spread by deportation thugs in Minneapolis is traumatizing schoolchildren. Many don't dare go to school.

*Texas Black man exonerated 70 years after execution in case marked by racial bias.*

A thug fabricated the claim that the murder victim had identified her killer to him, though other witnesses said she was beyond speaking by then. Then other thugs beat a false confession out of him.

The article linked to above displays symbolic bigotry by capitalizing "black" but not "white". (To avoid endorsing bigotry, capitalize both words or neither one.) I denounce bigotry, and normally I will not link to articles that practice it. But I make exceptions for some articles because I consider them important — and I present this comment about them.

*European leaders who know their continent's history must now see that the US president is siding with the forces of tyranny.*

People are unsure of what the inevitable huge disruptions AI will bring to software will eventually be, but one thing which is clear is that enterprise software as a service will be hard hit. The industry is producing products which are too awful, and is too bottlenecked on software development costs, to not be completely upended.

The way that industry works currently is that there’s generally a single dominant player in each niche which has a codebase with a million features ten of which are important. The problem is that every one of their customers uses twenty features: The ten which are important to everyone, and ten others which are important to them specifically. And which ten long tail features each customer cares about have very little correlation to each other.

It’s clear that million dollar a year saas contracts are going away. It’s becoming way too practical for customers that large to write their own bespoke solutions from scratch and wind up with something which sucks less. But that doesn’t mean everybody is going to write everything completely from scratch. Most likely there will be open source solutions for most problems which only have the ten big features and everybody vibe codes customizations for their their own deployment.

The open source business model for this is time honored and straightforward: The company maintaining the open source version also has a service where you pay for deployment. But now it’s even better, because they’ll have a vibe coding interface which is super trained on ten thousand other customizations of their codebase. They’ll likely even sneak in some human intervention in the background to help with rebasing when a new release of the base product comes out. And they’ll have a license which allows and all customizations to be upstreamed if the maintainers want them to be. There will probably be niche consultancies which specialize in helping companies do customizations of specific products but that won’t be done in house by the maintaining company because saas shops will still try to maintain high capital efficiency.

The whole saas industry is much more vulnerable than people realize. You could get me to switch off Jira just by making a comparable product which had page load times out of this century. And vibe coding will absolutely be at the core of the new way of doing things.

This video caught my fancy so here are my thoughts on improving the design. There seems to be a lot of things which can be done which should make big improvements but consider everything in this post speculative spitballing. Anyone who wants to improve on this mechanism is free to try my ideas.

Technically it’s a bit wrong to say this mechanism has ‘no moving parts’. It does have moving parts, they’re just air bubbles which are being captured on the fly and hence aren’t subject to wear. The problem is that air bubbles don’t like behaving.

Starting with where the water comes in:

The mechanism in the above video is cheating a bit because the pump getting the water into the top is aerating it. A proper mechanism should have a way of getting air into the water when it’s coming in slowly and steadily. In particular it should have a mechanism for being able to recover if the mechanism as a whole ever gets overflowed so it isn’t stuck with no bubbles in it forever. The simplest mechanism for this is to have a section of the pipe going down which has holes in the sides. As long as water is flowing fast it will pull air bubbles in through the holes. If it gets backlogged water will escape through the holes and can be directed to the exit, making room for air to be let in. The ideal size and spacing of the holes is unclear. If the mechanism were big enough it would probably improve things a lot to split across multiple pipes which have air intake holes to pull more bubbles in. It might also be a good idea to make a whirlpool and stick a pipe in the middle to help the air go down but that gets complicated.

Once bubbles are captured the downward pipe should be split into a bundle of straws to keep the bubbles from coalescing and forcing their way upwards. The ideal diameter of the straws is probably somewhat dependent on their length but should be small enough that surface tension makes water form plugs. The length of the downward pipe in the above model seems to be way too long. It appears to be that this is being done to make the pulsing effect happen but there’s a better way of doing that which I’ll get to.

The intake for the air bubbles should come from the bottom of the chamber where the pumping upwards happens. That should lead upwards to a manifold which is a short pipe with a horizontal cap at the top with holes in it, all kept under water. Air will then build up in the pipe and result in a steady stream of bubbles coming out of the holes. The size and depth of the holes as well as the material they’re made out of and the width of the pipe relative to the rate of air coming in all affect the nucleation of bubbles. What should happen is that bubbles of a reasonably consistent size come up at a reasonably consistent rate in a nice steady stream instead of the chaos you see above. There’s probably a range of possible sizes and rates of bubbles which are possible and that needs to be studied.

Instead of a single pipe going upwards there should be a bundle of straws. The bottoms of the straws should splay out and have tapered inlets with a one to one correlation with the holes in the manifold so the bubbles from that hole go directly into that straw and push the water upwards. The ideal number and diameter of the straws is very dependent on how far the water is being pumped, how quickly the air is coming in, and what they’re made out of. They should be thin enough that surface tension causes water in them to form a plug and makes bubbles force the water upwards. The idea is to make the water flow up slowly and steadily, with the upwards force of the bubbles just barely able to force it to the height it’s being pumped to, without wasting any energy on the momentum from those pulses. Maybe this shift in emphasis makes the whole thing technically a different mechanism.

At the top the straws should flare away from each other so the water going out of one straw doesn’t fall into its neighbors.

Hopefully these changes can improve the efficiency of the system from awful to merely bad. You’d still only use it when you care less about efficiency than low maintenance or quiet or specifically want aeration. Using all those straws will reduce how well it works on water containing particulates.

AI slop has been doing real damage to bug bounty programs. My company has expended significant engineering resources wading through the garbage. The bug bounty platforms do a decent job of filtering but some things are simply outside their expertise and need to be forwarded along.

This is an unfortunate turn of events. A few years ago false security reports were not terribly onerous and even when they happened they were usually someone earnestly thinking they’d found something. Even when the person was horribly confused they were usually serious enough that it felt right to try to encourage them.

To be clear, there’s nothing wrong with using AI as a tool for searching for bugs. If someone finds a completely legitimate security problem using AI as part or all of their toolchain and submits a properly formatted report they are free to claim it. I would give the benefit of the doubt and think that even the mostly bogus reports we’re getting are from people who are doing nontrivial amounts of work to train models specifically for bug finding with their own filters and processing to maximize chances of success. They must be submitting because they have some real hit rate.

The problem is that the burdens of evaluating false positives are borne entirely by the entity handing out the rewards. This wasn’t a problem back when submissions were done manually because back then having an instance of a report which was probably wrong but having a 1% chance of success was rare, and the costs of validating such things properly were small compared to the costs of coming up with the possible attack in the first place, and if you did submit and got a follow-up question answering it was a real burden on the submitter. Now none of those things apply so there’s a flood of low probability but worth a shot reports.

The solution to this I’d like to propose is something which would have been completely verboten a few years ago but now unfortunately may be necessary: Anyone submitting for a bug bounty should have to put down a deposit. Even a relatively low amount like $100 would probably make a huge difference. Ideally there’s a policy in place that there’s a generous refund program that submissions which are at all earnest get their deposit back even if they’re mistaken. If that causes too much arguing about what’s ‘earnest’ it may be necessary to make it a fee rather than a deposit, but I think it’s always legally okay to have a policy of returning such fees as long as it’s made clear up front that it’s completely discretionary on the part of the evaluator.

No doubt this suggestion will make some people very upset because it completely violates the traditional ethos of how bug bounties work. It would also create an opportunity for scammers to set up bug bounties for fake projects with lots of security holes which they then pocket the fees for submissions on and refuse to pay out any owed bug bounties. These are real problems and there are mitigations but rather than diving into the weeds I’d just like to say I know and I’m sorry but the situation is sufficiently out of control that this is probably necessary. I’m suggesting this publicly so I can be the bad guy who other people point to when they suggest it as well.

Before diving into this it would be helpful for you to read my basic music theory post and possibly more music theory.

Counterpoint is often described vaguely as ‘two voices playing off each other’. There are different phenomena which are referred to as counterpoint, but they all have two things in common: There’s more than one melodic line, and they aren’t playing in parallel. The most basic form of this is caused by an internal contradiction within the diatonic scale, best illustrated by labelling every other key like so:

You might notice that the white and yellow keys each form their own nice sub-scales where every pair of adjacent notes is a (major or minor) third apart and with one exception every pair notes two apart forms a fifth. This make the scale have similar properties to the pentatonic in that you can bang on notes roughly at random and it will all sound consonant, but it has two weaknesses: It has even fewer notes per scale than the pentatonic at 3.5 per octave, and it doesn’t contain the octave, which is the strongest consonance on the piano. To illustrate this the C notes above have been given asterisks, and you can see they alternate between white and yellow. This is because there are seven notes in the octave and seven is an odd number. In case you’re wondering why it’s called an ‘octave’ which indicates eight that’s because the vernacular predates people believing zero was a real number so all the interval names are off by one.

The result of the above is that a lot of melodic lines clash if they’re played in parallel with another instrument one octave off. If they’re two octaves off it returns to the same half-scale and it’s safe to play in parallel again. When there are two instruments about an octave apart they tend to dance around each other, playing in the same half-scale or hitting the octave on the off-beat or otherwise doing whatever they can to avoid playing a second against the other other instrument when one of them hits a seventh. This dancing around is referred to as ‘counterpoint’.

This phenomenon doesn’t apply to the pentatonic scale where it’s always consonant to play the same melodic line in parallel an octave apart.

Lots of the CVE world seems to focus on “security bugs” but I’ve found that it is not all that well known exactly how the Linux kernel security process works. I gave a talk about this back in 2023 and at other conferences since then, attempting to explain how it works, but I also thought it would be good to explain this all in writing as it is required to know this when trying to understand how the Linux kernel CNA issues CVEs.

The problem with learning to juggle one more ball is that it’s such a big leap. Not only is it one more object, it’s a completely different pattern. In order to have a gentler lead-up to a certain number of objects you need stepping stone patterns which are similar to the final pattern you’re trying to achieve but much easier and help you practice different aspects of it. If you work on patterns which are too easy you simply cruise and don’t improve. If you work on patterns which are too hard you don’t get anywhere and also don’t improve. Improvement comes not from the feedback being all positive or all negative but enough of a mix that you can train off of it.

Ideally you’d practice a certain number of balls by having a low gravity chamber where you start at a gravity level low enough that you can do it and gradually increase it to earth normal. Maybe some zillionaire will rent out enough time on the vomit comet to do that but for most people it’s impractical.

Short of that one could rig up a system where balls are held up on strings and counterbalanced on a pulley or pulled by a spring so their downward acceleration is lower. That would have strings in the way of things so you’re stuck with columns type patterns but it would be something. But I’m not aware of anybody trying it out.

You could also rig up a plank at an angle where you roll the balls up it to juggle them and gradually increase the slant until it’s vertical and you remove the board completely. I’m not aware of anyone attempting to learn numbers juggling this way but it’s an experiment worth trying.

Much more practical is to use siteswaps. This is the approach I used for learning 5 and have successfully used to teach others. Sadly I still can’t run 6 but have made progress. Some people find the later siteswaps harder than the final pattern but I think this has a lot to do with whether your goal is qualifying or running. Qualifying is getting twice as many catches as objects, running is keeping it going indefinitely. This approach is much more focused on running than qualifying.1

Here are the patterns for learning 4, 5, and 6 in rough order of difficulty. You should practice all the patterns which you don’t find too easy or too hard at the current moment and do the asymmetric patterns both ways.

Patterns leading to a given number contain a lot of throws of that height. To avoid needing a lot of high and low throws these patterns mostly mix that with 0, 1, and 2. You should clap on every 00 to keep the pattern from collapsing. In some cases of a 0 or 2 by itself you should touch your empty hand or the ball to your thigh but you don’t need to do that for all of them.

40, 441, 4

501, 52512 (baby juggling), 5511, 51, 55500 (clap), 552 (touch), 55550 (touch), 5551, 55514, 5

600 (clap!), 1601600 (clap), 66111, 61611, 66611, 66161, 60, 1616160, 666060, 660 (touch), 66661, 6662, 666660, 6

Before reading this post you should read through my last post on music theory. Here are a few slightly more advanced things.

The Major Chord

The notes of the major chord all come from the overtones of one note, like this, with the overtone keys labeled with red numbers for their ratio from G1:

The combination of all these notes is literally the major chord, as in all chords called major are subsets of it. The most common canonical major chord is the one on 4, 5, and 6, labelled gray in this picture. Adding in 7 or 9 gets spicy, especially if you keep in 8 or 10 or move them to 7/2 or 9/2. Music with that spiciness added is generally referred to as ‘jazz’. 11 and 13 aren’t things at all which is why I didn’t include them. You might notice that the tonic here is placed on the G instead of C. That’s to make the 7 land on a white note. A strong case can be made that in jazz the default major mode should be Wednesday instead of Tuesday.

The Minor Chord

The notes of the minor chord are all undertones of one note. Here they are with the blue labels being the undertone ratios from A6:

Like with the major chord this is the source of all minor chords. The most common subset is the ones labelled in gray, which is usually what people mean when they simply say ‘a minor chord’. This is slightly awkwardly placed with the tonic on D even though the note at the center of everything is A. The source of the asymmetry with major is that the human ear perceives the defining part of a chord as being its lowest note. Undertone series tend to sound less consonant than overtone series. Also as with the major chord 7 and 9 are spicy especially with 8 or 10 or moved to 7/2 or 9/2. This diagram is positioned so 7 and 9 land on the diatonic scale. It illustrates that a strong case can be made that the default mode for jazz minor pieces should be Thursday and not Friday.

The Blues Scale

The blues scale is the pentatonic scale with one more ‘blue’ note added. It’s best thought of as being midway between two different just intonations, like so:

With that aggressive use of 7-limit intervals and multiple things to bend to blues works well with guitar.

Well Tuning

With just intonation the second, tritone, and minor seventh can all be reached either by going up or down resulting in them having two natural values which vary by a syntonic comma, like so:

Tuning a piano to the midpoint of these value is an example of what’s called ‘well tuning’ where all the keys sound good. There are many different types of well tuning. This is the easiest one to tune a real piano if you don’t have special equipment and is also the most just flavored. The standard equal temperament tuning is the most smoothed over.