Dynomight

Quite a fun read more often than not: https://dynomight.net/ Takes all kinds of lifestyle and tech topics and nerds out about them thoroughly. If you've ever wanted to see mundane things overanalyzed and backed with solid facts, I recommend. I don't necessarily agree with all their views, but I've always enjoyed an article and it's rarely if ever confidently wrong.

OP has another post on the definition of heritability, which I really liked: https://dynomight.net/heritable/ . I'm a layman, though, so since you seem knowledgeable, I would love to hear your thoughts on that article! For instance, OP's definition H = Var[G] / Var[P] seems to bypass the issues you mentioned: > For example, the way it is used also usually has a big problem in that the standard formula assumes that Cov(G, E) = 0 (or at least is negligible), whereas in practice that is not actually true [3, 4]. > This definition of heritability is also mathematically flawed in that it assumes (without evidence) that P = G + E, or at least can be reasonably approximated this way.

I highly recommend the author's other post https://dynomight.net/heritable/

I'm surprised that this article doesn't appear to mention the RCT on semaglutide and alcohol use disorder by Hendershot et al. that was published in JAMA Psychiatry in early 2025 (though it's possible I missed it) https://doi.org/10.1001/jamapsychiatry.2024.4789 This was largely portrayed as a great result in the popular press although personally I think it was a bit of a disappointment given all the amazing anecdotes https://dynomight.net/glp-1/

Smaller doesn't mean harder to filter, beyond a certain size. HN favorite (and pre-covid particulate filtrering proponent) dynomight explained this well in an article on a cheap air purifier: https://dynomight.net/ikea-purifier/ Also, I'm pretty sure respiratory viruses need aerosol droplets to get anywhere they can harm you, and those droplets are much bigger than viruses anyway.

Technically yes, but... it's moderately tricky to get an LLM to play good chess even though it can. https://dynomight.net/more-chess/ This is significant in general because I personally would love to get these things to code-switch into "hackernews poster" or "writer for the Economist" or "academic philosopher", but I think the "chat" format makes it impossible. The inaccessibility of this makes me want to host my own LLM...

There's a complex Numpy indexing codes riddle in the section of "I don’t like NumPy indexing" and Gemini Pro 2.5 came on top (DeepSeek R1, only get the first time right but not later) [1],[2]. > For fun, I tried asking a bunch of AI models to figure out what shapes those arrays have. Here were the results: Based on the results from the top 8 state-of-the- art AI models, Gemini is the best and consistently got the right results: [1] I don't like NumPy (204 comments): https://news.ycombinator.com/item?id=43996431 [2] I don't like NumPy: I don’t like NumPy indexing: https://dynomight.net/numpy/

Gradient descent is a dumb optimizer. LLM training is not at all like a human reading a book and more like evolution tuning adaptations over centuries. You would not expect either process to be aware of anything they are converging towards. So having lots of books that talk about chess in training will predictably just return a model that knows how to talk about chess really well. I'm not surprised they may know how to talk about the rules but play them poorly. And that post had a follow-up. Post-training messing things up could well be the issue seeing the impact even a little more examples and/or regurgitation made. https://dynomight.net/more-chess/

If it's studies you want with quantifiable impacts, I'd recommend you read (part of) this investigation (because it's so long)[0]. A couple of the top recommendations are to extinguish candles with a lid and to ban incense. [0] https://dynomight.net/air/

Note: the possibility is not mentioned in the article but rather in the comments [1]. I had to click a bit to see it. The fact that the one closed source model is the only one that plays well seems to me like a clear case of the interface doing some of the work. If you ask ChatGPT to count until 10000 (something that most LLMs can't do for known reasons) you get an answer that's clearly pre-programmed. I'm sure the same is happening here (and with many, many other tasks) - the author argues against it by saying "but why isn't it better?", which doesn't seem like the best argument: I can imagine that typical ChatGPT users enjoy the product more if they have a chance to win once in a while. [1] https://dynomight.substack.com/p/chess/comment/77190852

I'd like to work on my writing skills. Some mix of nerdiness and personal Journaling. https://dynomight.net/ is a website that I'm quite inspired by, as it's a interesting mix of personal commentary (https://dynomight.net/advice/) and deep data dives into e.g. homelessness and seed oils. My main barriers are 1) writing discipline and 2) some obsessive need to link/cite every claim I make. This results in me taking months to write an article, be it due to laziness or constant rabbitholes/google scholar searches about an idea. Any clue on dedicated disciplined writing time or even considering "reducing" the rigor of writing?

I just want to say that dynomight also has a great article on air purification in general: https://dynomight.net/air/.

I agreed that I thought SU wouldn't work in today's internet, and I clicked on the cloudhiker link thinking I'd be met with SEO trash, but I ended up on this post: https://dynomight.net/ikea-purifier Which was a great post and now I understand more than I did about how air filters work...more complicated than I'd thought. Maybe I should give cloudhiker a try.

It’s good of you make this correction, we need to be accurate when discussing the research at hand. That said, it’s worth pointing out that “adjusting for” a factor isn’t quite as effective as it sounds: https://dynomight.net/control/ Having built some regression models in my time, throwing a ton of extra variables in doesn’t fill me with confidence that I am “controlling” much of anything.

Dynomight was fairly critical of Wirecutter’s coverage of air purifiers as well (specifically their negative review of the inexpensive and decent IKEA air purifier). I used to love Wirecutter, but it seems to have fallen prey to the inexorable enshittification of the web. https://dynomight.net/ikea-purifier/

Cleaning dryer lint does massively spike PM 2.5. As far as we know, any PM 2.5 probably is bad for you and could reduce your lifespan. https://dynomight.net/humidifiers/ That being said I have no idea if it's worth wearing a mask. Edit: this joke comment that struck a nerve haha https://news.ycombinator.com/item?id=34428710

The intention of the OP/post might be good, but the comparison (https://dynomight.net/img/cuboid_purifier/boxfan.jpg) makes no sense whatsoever. They have just thrown 3 filters on top of the box fan with lots of gaps on the side. Air will mostly follow the path of lease resistance and not go through the filters. You need a filter that is the same size, 20x20, as the box fan and then the filter(s) will need to be sealed, via duct tape or so, around the fan for it to be effective.

The author's last-years underrated reasons to be thankful was also excellent: https://dynomight.net/thanks/

Q2 of 2020 was one of the weirdest quarters in American history. From covid lockdowns to race riots, to heightened police-civilian tensions; there was nothing normal about it. I would love to see these exact same graphs juxtaposed against a normal year such as 2019. Anecdotally, the crime rates in west coast blue city downtowns have skyrocketed. It seems impossible to park your car in Seattle/Portland/SF downtowns and not have it be broken in. The average junkie seems to have gotten more aggressive and bashful too. I wonder if the rise of p2p meth has something to do with it. (https://dynomight.net/p2p-meth/)

I've built dynomight's DIY filter[^1], and I love it. No DRM issues either. [^1]: https://dynomight.net/better-DIY-air-purifier.html

Agreed. There was an interesting post here recently that had this theory why churches work: "Eventually, you pick a relatively relaxed “normal” religion, complete with all the “God” stuff. You don’t believe everything they say is literally true, but our reality is strange and inexplicable, so who knows? There’s a great stable community and you talk about the broader themes of life and you’re healthier and happier. You secretly guess that the group is sustained by a minority of true believers who make up critical mass for a larger group of people like you, but you never talk about this and neither does anyone else. When you have kids, you feel weird about explaining your thoughts about all this to them, but you don’t parrot back what the religion says when you’re talking about the meaning of life either." https://dynomight.net/plans/

> Robert Lustig videos on Youtube are a great way to start learning about this. Or off YT: https://videos.lukesmith.xyz/w/feioCyaQEZ1ogHK1oNJM4K > Thus, the people taking fiber out of and putting sugar into everything are the biggest current mass killers. Stringing them up may be an over-reaction. Maybe convicting them would be, too? Anyway stopping them seems like a good idea. Think about them next time the news is full of somebody killing a dozen people. Is a dozen bad, but a million A-OK? IMO simply not compelling millions of people to pay them for it would be a good start. Defund all the agencies that promote poison, whether by subsidizing corn syrup, villianizing natural saturated fats to sell trans fats and sugar, or conspiring with drug companies and corrupt researchers to sell addictive substances (as discussed this week: https://dynomight.net/alcohol-trial/).

Great site! I love the "What happens if you drink acetone?" article: https://dynomight.net/2020/09/14/what-happens-if-you-drink-acetone/

I really like https://dynomight.net/

> My real worry about seed oil theory is that it’s a distraction. If you want to be healthier, we know ways you can change your diet that will help: Increase your overall diet “quality”. Eat lots of fruits and vegetables. Avoid processed food. Especially avoid processed meats. Eat food with low caloric density. Avoid added sugar. Avoid alcohol. Avoid processed food. > > I know this is hard. You could even argue it’s unrealistic. That wouldn’t make it wrong. > > Look, I wish strong seed oil theory were true. That would be great. All we’d have to do is reformulate our Cheetos with different oil, and then we could go on merrily eating Cheetos. Western diet without Western disease! Sadly, I think this is very unlikely. Dynomight has written an excellent article on seed oils. We have examples in our families of people with very bad diets, favouring tons of butter because ‘at least it’s not processed’. The research doesn't back this up. As the quote above suggests, we need to stop thinking of ‘magic bullets’ that’ll get us out of eating better. Added sugar and trans-fats might be the only true bullets. https://dynomight.net/seed-oil/

It is shockingly difficult to use LLMs in chess study. I don't need it to be a better (or worse) Stockfish; an LLM should be great at taking a FEN or multiple lines from Stockfish via MCP or tool call and explain why positions are evaluated the way they are, typical plans in the position (drawing from pretraining knowledge of a vast archive of games), and how to explain to a human to study these positions. I suspect that the large amount of chess pretraining data is not well synchronized with positions, because in books and articles the text is typically accompanied by pictures of the positions, NOT FENs / PGNs. So the training on the text is decoupled from the representation of the position. Regarding your tool call thing with stockfish/maia, I made a tool like this for myself called Blunder Sniper which iteratively feeds positions - that I'm likely to get given my openings - and recursively calls the lichess DB and finds the first time in the top 80% played moves in each chain where the opponents in the rating range blunder as the most common move in the chain. It was a fun way to use an alternative to engine-based preparation that many strong players use, which is something like Nibbler + lc0 and using contempt high values to find higher variance lines rather than game theory optimal ones. Some day I'll expand on the gpt-chess articles [0] that I found super interesting, fine-tune models... well, I keep telling myself that, anyway... [0]:…

This is a nice follow-up: https://dynomight.net/car-trouble/

You can indeed take this route [0] and even DIY something with a box fan. But I think the jury is a bit out on this. Quote from that link below. You also need to move that much more air to get the multiple passes required and that can get pretty noisy. > They claim that around 90% of particles sized 0.3 microns are larger [for a MERV] are eliminated in a single pass. That’s good, but not totally reassuring. The question is, does it remove 99% in two passes? If 90% of the particles in the ambient air were large and the filter only catches large particles, then additional passes would never get rid of the most dangerous small particles. This is why I trust HEPA filters a bit more: since they remove almost all particles in one pass, I’m confident they should remove almost all particles eventually. Also, there is the idea that HEPA filters act like a sieve. They don't. This article from the same guy is a great read [1]. So basically, multiple passes for the win. > Air filters easily catch both large and small particles. It’s the intermediate regime where things are hard. The size where the filter performs worst is called the Most Penetrating Particle Size (MPPS). Typically this is around 0.15 microns. [0] https://dynomight.net/2020/12/15/some-real-data-on-a-DIY-box-fan-air-p… [1] https://dynomight.net/ikea-purifier/

It is an interesting article and some of the results are unexpected, but the layout is way too long and could be greatly condensed. At first glance and speaking from how I like to lay things out in a paper: - You do not really need to show the prompt interface. You can show it once and thereafter use a bulleted list format, or simply show the input image if it responded correctly. - Your figures should be about half of their size, they don't need to fit the width of the body. - For comparative results against other models you can use a table with colored cells, with the test name on the row and model names on the columns. - For the dog, show a side-by-side figure with the raw image on the left and the box on the right, and include the coordinates it gave you in the body. - In your conclusion show the full matrix table of comparative results and summarize the relative strengths of the model against the others. In terms of the writing and methods your conclusion says little and your tests do not go into significant depth. For example with the tire image you could show that it succeeds when cropped but as the photo gets wider it begins to fail to correctly identify the text in the image's center. For example see the methodology and presentation this article used: https://dynomight.net/ducks/ Also, the OCR test is too simple, even a 20-year-old OCR algorithm would probably recognize that. Experimenting with progressive degradation of the image could show its strengths, and…

My phrasing was perhaps too strong. I learned this from here https://dynomight.net/air/ This is a great read. Here is a previous HN discussion on this: https://news.ycombinator.com/item?id=26964722 There are some disagreements but I find the argument that air quality is greatly underappreciated as a health risk convincing. And that candles cause bad air quality. The article talks a lot about healthspan which I think is a great measure and idea. Your referenced article talks about at-risk people and that they should avoid candles. It also states that candles should be burned in well ventilated spaces, which I believe most people don't do as they don't consider it. A healthy persons healthspan is negatively affected by candles and such. They just have more health to spare. The article talks about cooking as well and they are likely talking about cooking with gas which is very unhealthy. It should be illegal to sell indoor gas stoves for cooking in my opinion.