epilepsy warning for the 'MAGIC' style :(
720 karma · joined October 14, 2016
But Excel is not a relational database: for simple cases, having an extra table for a non-scalar column is awfully confusing to present to an end-user. From that angle this makes total sense. There's still 1NF for more complex embeddings, anyhow
The killer feature is that you can put the two together for a rotation axis (~ vector) and angle (~ scalar).
With just three gimbals (rotating circles), if gimbals A and B are aligned, you only have two degrees of freedom (rotating A is the same as rotating B). Because of this, interpolating angles is unwieldy in vector space. 'Gimbal lock' confounds animation (in hilarious but unrealistic ways) but also aerospace (four hours before 'one small step for man', just after landing, Collins joked he would like a fourth gimbal for Christmas).
I'd love to see just how Lipschitz continuous, how smooth, the potential field can be; how adding features puts it closer to or further from solutions that fit the consecutive-no-duplicate constraint, say. Adding a smooth 'hill' is probably viable; is a 'river'?
One little critique I have is in the latter third. Using LLMs for proof is pretty standard now, but the way the text focuses on their tribulations was distracting. It might have been cleaner to use the mathematician's "we" after introducing the 'co-authors', so that the casual reader might sink their teeth into the math rather than be reminded LLMs can sometimes cost money and go around in loops.
But otherwise this is a really gorgeous article! The visualisation is really powerful, and something about how the symmetries impose a kind of conservation (which looks like hot soup but actually has a smooth potential) are very exciting, and curiously very physicsy.
Might be useful to
- add a wordle-style 'SHARE' button, and/or
- make the canonical URL that of the puzzle (and only the attempt on completing/abandoning it)
Confusingly, "more" means "please add more terms"...
As a current juror I really do not feel for how long jury selection might have taken here, and I can imagine deliberation will take longer yet.
Let the funding go to some actual charitable foundation which offsets the very real negative externalities of large-scale AI, I say. That seems most likely to benefit humanity.
[0] https://www.theguardian.com/technology/2026/apr/27/elon-musk...
I am also sick of handling port numbers - I end up allocating them on a schema to different services, so for testing I can spool any VM/service combination and avoid crossover. But if I want the same service twice, ah...
It always fascinated me that ports don't have any kind of textual resolver, so you can bind to `:1234` and also say "please also accept `:foobar`". But that would itself require some kind of "port resolver" on a device, and that's another service to break and fix :)
"OS-efficient cross-platform HTML-based UI toolkit" is a great technological thing, but neither PyWry and Tauri's sites make that clear, or meaningfully advertise what they do. Which is a shame, because there is myriad software which might benefit all to use this.
[0] Tauri is akin to Chromium, I think? https://tauri.app
[1] and also a rather large amount of LLM integration; the source for PyWry has a whole section for Claude bindings
[2] the Webview Rendering librarY (WRY) used in Tauri https://github.com/tauri-apps/wry
“Symmetric” user reporting is dearly needed in some websites; as you say something can be mass-reported with no real recourse.
> Why create yet another physics engine? Firstly, it has been a personal learning project.
which is really rather wonderful and inspiring to see.
Certainly a terrifying amount of responsibility and upkeep for each individual website. If the UK wishes to establish this and not want it to lead to an insane amount of privacy leaks, it should consider developing a technology that makes it work in a privacy-respecting way, like the European Age Verification Solution [0]'s Zero-Knowledge Proofs.
For... some in the comment section, please recall the HN guideline: "Comments should get more thoughtful and substantive, not less, as a topic gets more divisive."
The only link is the person -- that their acts inform their thoughts and habits, which informs future acts. In this case "good deed math" is likely a post-hoc rationalisation, predicted by the Franklin model but not exactly encouraged.
"Fantasy or faith? One company's AI-generated Bible content stirs controversy" https://www.npr.org/2025/09/07/nx-s1-5518263/ai-bible-christ...
It's ice that burns: cages of water trapping methane, and indeed the largest non-atmospheric store of it on earth. It forms interesting fractals under a microscope, has subtle and historical climate effects, fosters methanotroph communities. It has commercial interest for methane extraction and may work well for static methane storage.
A fascinating topic to stumble upon!
Its constant drain even when not 'in use' seems to imply it's classifying tabs as they change page (though it might be telemetry or uncommented testing). If so, it's an example of premature optimisation gone very wrong.
It's a shame, because it overshadows the fact that naming tab groups is a perfect use case for an LLM, alongside keyboard suggestions and reverse dictionaries [1]. I'm ardently distrustful of LLMs for many, many purposes, but for the tiny parameter and token usage needed it's hard to not like. Which is a shame it's (somehow) such a drain.
[0] https://github.com/mozilla-firefox/firefox/blob/7b42e629fdef... exports a SmartTabGroupingManager, though how or why that is used without being asked eludes me
[1] https://www.onelook.com/thesaurus/ Can be helpful in a pinch when a word's on the tip of your tongue, though its synonyms aren't always perfect.
Having pattern recognised and extrapolated to my perception of the wealth-happiness curve, it seems that when your wants are met by your current wage, wanting more money is paradoxical -- it requires either time or stress that take away from the many other richnesses of life.
A little ambition (and savings) is good -- you can't recline too far back into the comfort zone -- but wealth never struck me as a particularly important measure of a person.
I have only a _passing_ familiarity with VTubers (my friend is one) and this is obviously and patently wrong. A VTuber is a YouTuber with a virtual avatar; no more, no less.
The article goes onto correct itself, but it’s a bit disheartening to see obvious misuse of terms in the first sentence…
[0] https://www.congress.gov/bill/119th-congress/house-joint-res...
Speaking as someone not from the country, US federal politics has became alarmingly... gerontocratic. Which means the younger generations get fewer chances to grow political acumen and expertise (unless, of course, they have backing). Which only hurts the rest of the country in the long term.
I tend to liken them to very drunken scholars. They know things, usually at about a Wikipedia level, and they’re cheery too. But they lack a capacity to doubt themselves; they often are confidently wrong.
Often the greatest help is understanding natural language, but given its hiccups… time using it is probably best spent using it as a drunken librarian – to teach how to phrase and fetch information
As one telling recent example, I tried using an LLM to help with some jq (with which I’m rusty); it got a few basics and then repetitively tripped over a syntax hiccup on loop, “correcting” itself to the same answer each time. A StackOverflow search or two, for comparison, answered my questions and taught some new syntax too. Probably took less time, but more critical thought.
That, coupled with the fact LLMs tend to give an answer and then also an unnecessary verbose step-by-step, means I tend to dislike them.
I also have a huge bugbear about “AI” as a term because it tells you very little. Plenty of applied statistics (markov chains, clustering algos, deep learning eg computer vidion; even SearchRank) are used heavily in research and other cases to do a lot of good. Even for the layman: the Seek app by iNaturalist is awesome for identifying common plant species; Stockfish is (now) a NN that dominates in chess.
But these are classifiers, not generators. By their very nature it is just statistics to evaluate a classifier on a test dataset. Generators, however, are far, far thornier to test, and seem a lot more prone to overfitting.
While I’m not familiar with a typical trained generator tensor, I imagine the optimal one will be surprisingly sparse, though not in a structured way - corresponding to a more clustered “small world” network, which IRL seem the most productive.
That would be let alone the many other costs - including any fees to fight the uphill battle to prove the legality of this.
"Move fast and break things" is cute for a prototype when mistakes cost only time and pay dividends in experience. On a scale of government it's like taking a bulldozer to thousands of Chesterton's fences a day. Which is efficient, from a certain perspective...
(I very much agree with the sentiment...)
Of course this is another standard that tools would have to be incredibly careful to keep track of, but JSON is decently mature and a bunch of modern tools can operate in it, so at least there's slow progress away from the plaintext quagmire.
[0] https://www.theverge.com/2025/1/3/24335045/elon-musk-4chan-a...
(Based on the "vocalist has a seizure" line which I've seen before in similar situations, can it be assumed this is AI-generated?)