rmunn

Firstly, they didn't spend trillions on training.

Then please argue with the guy who said they did, the guy who goes by richardw and whose comment I was responding to in the first place. He's the one who said "If the US companies need trillions to barely beat Chinese companies spending billions, despite a multi year head start..." and, to the best of my knowledge, you haven't replied to him at all. Instead, you have consistently replied to me, arguing points I'm not discussing, not TRYING to discuss, and arguing with me about points that somebody else made. As I told you in a different comment, PLEASE read more carefully in the future.

I am sick and tired of talking about this by now. I no longer care if you have a good point or not, you have exhausted my supply of good will. Good bye. Feel free to have the last word, you're the only one who cares by now.

In Codeberg's case, they have chosen not to distinguish between those. Their bright line is "mostly", as in, if your project "mostly" consists of LLM output then it is not allowed. The definition of "mostly" is, of course, vague, and people are going to argue about where the line should be drawn. But there's at least something vaguely approaching an objective measure: they might look through commits and count how many have an LLM listed as a co-author, or they might count the number of lines of code changed by those commits. Who knows, at this point, what they will end up doing. But they're not trying to judge "how well did you curate the output of the LLM"? They have, instead, decided to judge "how many lines/commits/something created by LLMs have ended up in your code?"

Which is true, but also completely irrelevant to the point of this discussion. Which was about the U.S. companies spending trillions and the Chinese companies spending billions and matching them in quality. I made the point that the U.S. companies and Chinese companies were doing different things: training on raw data vs. distilling the model that someone else had trained. That difference explains the difference in spending. (The secondary point is that the Chinese companies would not have been able to get to the point they have, while spending as little as they have, without Claude, GPT, et al to copy from).

This is probably the last reply I'll make to you. I'm getting a little tired of repeating myself. Whether you're just not getting it, or refusing to get it, either way it's starting to feel like a waste of time to try to rephrase the same thing again and again. Please read more carefully in the future.

And, as I pointed out elsewhere, the difference I was pointing out was an economic difference. You keep on ignoring that fact and thinking I'm talking about ethics, but I'm not. I'm saying the difference is between spending trillions on training from raw data vs. spending billions on distilling that trained-from-raw-data model.

It takes more computing power, and money, to pay for training an LLM, compared to distilling an LLM that someone else has already trained. That's the economic difference.

The difference I was referring to was economic; I was not making an ethical judgment. So many people seem to be reading my comment as making an ethical judgment (based on their reactions); maybe I should edit it to clarify.

Nope, seems I'm past the edit window. Oh well.

It quotes figures from Q1 of 2026, so it was definitely published this year.

The first time I encountered VTubing, I thought the point was just to hide your face, for various reasons. Being sensitive about your looks, not wanting to attract stalkers, whatever. But there's apparently a lot more to it; I was surprised.

Huh? What exactly do you think I'm trying to "cope" with? That American AI companies are overleveraged and some of them are going to go bankrupt? I've been predicting that for months. Why do you think I would need to cope with the idea?

Not saying the American AI companies are ethically in the right, just that the work they did is a lot more expensive to do than what the companies distilling their work are doing.

Good UIs look alike, yes. But I'm not sure that comment is a good rebuttal in this particular case, because the model calling itself Claude is pretty strong evidence of distilling: https://x.com/denisewu/status/2077984660211269870

That's the difference between innovating and copying/distilling someone else's innovation. https://wccftech.com/chinas-kimi-k3-identifies-itself-as-ant...

And null termination is guaranteed to work, because the only two characters forbidden in Unix filenames (for most varieties of Unix, I won't guarantee there aren't some weird variants out there) are / and null.

The only times I've needed something more than `find -print0 | xargs -0` has been when I need to apply logic to decide whether to process one of the files, in a way that's not easy to express in a `find` command. Then I write a small script with a for loop and if statements inside it.

But more people should know about `-print0`. It's the answer to 95% of the problems with `find | xargs`.

"Torched"? Nah, not at all. But the donut tutorial is famous for a good reason. A donut is basically a torus, which is one of the basic shapes that Blender gives you, so you're starting from a shape that's nearly right — and then the tutorial teaches you how to deform it slightly so that it looks less perfect and therefore more "real". Which means that you're learning the process of modeling. That same tutorial then teaches you the process of texturing, of lighting... By going through the process of crafting an object you don't actually care about, you learn the steps involved in crafting the ones you do care about.

I prefer larger screens on laptops, so every laptop I've ever owned has had a numpad. Naturally you can't fit a numpad onto a 12" laptop, but USB (or Bluetooth) numpads are available; a quick Amazon search says $10-20 for most of them.

Plus, Blender has an "Emulate Numpad" option that lets you map the numpad controls onto your regular numeric keys (making it harder to use them for their normal purpose, so I wouldn't recommend it unless an external numpad is not an option for some reason). And they have an "Emulate 3 Button Mouse" option, where the middle mouse button can be emulated by holding down Alt while clicking the left mouse button. (Which means that some Alt+LMB shortcuts become unavailable, so again, the best option is to plug in a USB or Bluetooth mouse).

They've thought of it. Their defaults remain the default for a good reason, because the hardware to properly use the default shortcuts is cheap. But they've also accommodated the people who don't want to buy an external numpad or mouse.

Yes. https://www.raspberrypi.com/products/power-supply/

I have always bought the official power supply, yet of the two Pi 3s that I have connected to my TV at home, the one that runs Kodi/OSMC to serve up movies has never suffered any filesystem corruption, despite it suffering repeated power outages without warning (it's low enough to the ground that the toddler has often pulled the power cable out, and I haven't been able to find a better place to keep it). The other one, bought at the same time from the same source, is meant to be the one the kids use for programming in Snap!, a variant of MIT Scratch that doesn't have the artificial limitations that Scratch has. (Snap! functions — the exclamation mark is part of the name — can take lists as input, whereas Scratch functions can only take numbers or strings, giving it an artificial ceiling; also, Snap! functions can take functions as input and return functions, making higher-order programming possible, while Scratch deliberately does not allow that). However, that Raspberry Pi has suffered two SD card failures in the past year; it's been down more often than it's been up, and I've ended up using either my laptop or my wife's laptop for my kids' programming lessons.

I don't mind at all.

The main benefit that I see (not being a linguist myself and seeing it second-hand, as it were) is saving time; it's quite a lot of effort for humans to listen to an audio recording and write it down in IPA. I didn't realize just how much effort was involved until I watched my wife researching her thesis, listening to the same recording over and over and over.

Also, there are possibilities for automated translation that the IPA transcription opens up. Speech-to-text models need different training than text-to-text translation models. The other day one of the linguists I know sent me a message, saying "I have a lot of recordings in language X" (I forget which language, but it doesn't really matter which language it is) "and I want to transcribe them. Is there a way to use AI to do that?" I thought it through, and realized that recording-to-IPA followed by IPA-to-correct-alphabet would be a way of dealing with the combinatorial explosion (according to https://www.ethnologue.com/insights/how-many-languages/ there are close to 7,200 languages spoken in the world, though "spoken" may be a misnomer since that number includes sign languages). There's no way anyone can train a model to convert about 7,000 languages from spoken form to written form; that's just way too many possibilities, and I suspect that audio models are necessarily large because audio is more complex information than written text. (If I'm wrong on that, someone please inform me). But although the human mouth (and throat) can produce dozens and dozens of different sounds, that's a finite number. (I think it approaches a couple hundred, but that's still a lot less than 7,000). So the number of distinct sounds and their IPA equivalents that the model would have to distinguish is smallish. Then another model could be trained on converting IPA to the alphabet (and spelling) of the world's 7,000-ish languages — and that would be a lot simpler than sound-to-spelling.

You're right that the recording is "more ideal" than any representation, and linguists will always archive the recording (in a lossless format, whenever possible, though if the original recording was done by a non-linguist who recorded it in MP3 then there's not much point in storing a WAV) alongside their transcription. This allows anyone else who might be doubtful on a point to go listen to the original recording and say "Yeah, Smith got the transcription wrong on that word. That's not an alveolar click, it's a velar click." But when you're writing a paper, you don't stick audio recordings into your document, you write down transcriptions. An audio recording without a transcription is not nearly as useful to linguists as having both. (EDIT to add: And such transcriptions, when you're writing papers about the sounds of a language, are pretty much always in IPA as well as in the language's standard alphabet. IPA isn't just a means to an end, it's also an end in itself. So even if a speech-to-text model existed for the language my linguist friend wanted to transcribe, he would still also want an IPA transcription as well, 90% of the time.)

The linguists I know are not necessarily a representative sample... but they're mostly interested in just that: maximally faithful IPA representation. They want to know if the speaker switches back and forth between aspirated p and unaspirated p on the same word, because that tells them something about the language — that aspirated consonants are not meaningful.

Linguists studying the sounds of a language, its phonology, often want to find "minimal pairs", words that differ by only a single sound. For example, din and tin in English. You record a native speaker saying both words, and telling you their meaning, and then you play back either recording A or recording B to other native speakers and ask them which word it is. If they can identify the word every time, then you've found two sounds that are meaningfully distinct in this language. (Some languages don't distinguish the d and t sounds, but English does). But if the native speakers go 50/50 on which word it is, or ask to hear it in a sentence for clarification because it could be two or three different words, then you've found a pair of sounds that this language does not distinguish. (Note that you're playing the words in isolation, because sentence context might make it obvious which one it is, e.g. you can't tell if an English speaker is saying their or there until you hear more words of the sentence).

So yes, the linguists that I know (who, again, are not necessarily a representative sample) are interested in as faithful an IPA representation as they can get, because that inconsistent transcription will give them many clues about the language. It still all has to be checked, because that switching back and forth between aspirated and unaspirated p (for example) could have been an artifact of a poor-quality microphone not picking up the aspiration, or a windy day causing aspiration sounds that the speaker never said, or the speech-to-text model making a mistake. But I watched my wife listen to the same two-second recording on loop over and over, trying to be certain of which sound she was hearing in the middle of the word. Double-checking the output of the model would (in most cases) only require listening to the audio once or twice, not half-a-dozen times like she typically did while researching her thesis. At the time, LLMs were not really a thing yet, but if she were doing her thesis today I bet a speech-to-IPA model would have saved her quite a lot of time — but only if it output every distinction, even the ones not meaningful in the target language. The "maximally faithful" representation, as you put it.

... very-rarely-spoken languages probably don't have distinct dialects ...

That's true if you mean "very rarely spoken" literally, as in even the native speakers don't get to use it very often. But many languages aren't widely spoken (such as only in a certain geographical area, which sometimes is only a single village, or other times a small number of villages). But inside that area, they are frequently spoken. And you might be surprised how many of those small-geographic-area languages still have distinct dialects.

For example: my wife (a linguist) did her master's thesis on the pronunciation of a language with about 7,000 speakers, and identified how many distinct dialects there were. (Which is why I know a little bit about this). She recorded native speakers from all 13 (I think it was 13, but it might have been 14) villages where the language was spoken, and found five different dialects, which she grouped into two "main" dialects. (Think American vs British in the English language, with subdivisions into Midwest, New York, and New England accents and so on, and you'll have the right general idea — though these dialects were closer to each other in sound than Midwest vs New York). I'd have to go reread her thesis to give you any more details. But this was a language that was only spoken in a small geographic area, but it was frequently spoken, because that was the main language of those villages. (The country's official national language is what the kids learned in school, but some of the people, mostly those 60 years old or older, hadn't gone to school, because the first government school in their area was only built 60 years ago -- so they only spoke their minority language, and not the country's language, and their kids had to translate for them if they had to leave their village and go shopping in a major town).

I've seen many analysts say (I'm no expert myself so I'd have a hard time judging) that the official government figures for things like GDP and so on cannot possibly be accurate. (And before you ask "Which figures, which years, what degree of inaccuracy?" — you'll need to ask someone who actually knows). Which does make it kind of hard to figure anything out about China's economy from outside.

There are, broadly, two kinds of audio recordings that linguists want to transcribe. One is native speakers telling traditional stories, where they're speaking naturally and taking the natural shortcuts (such as "wanna" and "gonna" in English). The other is native speakers reading words (or short example sentences) very carefully and distinctly, so that the linguist can listen to the recording over and over to learn how to pronounce the word right. In those recording, they'll say "want to" and "going to" rather than "wanna" and "gonna".

Thanks for the pointer; I'll check out that model and see if it handles the "slowly and carefully" type of recording better than the "natural speaking" type. (And depending on what kinds of errors the model makes, even the recordings where it makes errors can prove useful: for example, a linguist studying regional variations in speech would want the model to produce the IPA for "gonna" rather than "going to").

Looks very cool. One thing I have been looking for, which this doesn't seem to cover (at least I didn't see any mention of IPA in the model documentation), is a way to transcribe unknown languages phonetically, using the International Phonetic Alphabet to spell them (sound-based spelling rather than meaning-based spelling). I know several linguists doing research on minority languages (fewer than 10,000 speakers in some cases), which are small enough that they will never have enough effort made towards training language-specific models in that language.

Are there models I'm not aware of that are trained for this task? Taking audio in an unknown language, and rather than identifying the language, just transcribing the sounds to IPA? That would not be useful to most people, but it would be a Godsend to many, many linguists working with minority languages around the world.

I'm not going to predict the moment, because I thought the dotcom bubble would pop back in 1999. I could see how overvalued things were, what I couldn't predict was how long the market would remain irrational.

If you can wait until one of the major companies goes bankrupt (assuming that that happens, of course) before buying a new PC, then it should be a lot cheaper than today. Because when a company goes bankrupt, a lot (or possibly all depending on how bankrupt they are) the hardware they bought will be sold off to pay their debts, meaning a LOT of now-scare RAM will go back on the market, bringing prices back down as supply suddenly catches up with demand.

It's not guaranteed, and the pent-up demand for RAM from all the other people who had been holding off on buying new hardware will mean that supply won't exceed demand for quite some time. But that's my strategy: wait for one of the companies who are currently outspending their resources to go bankrupt, then buy hardware once the supply is no longer quite so ridiculously tight as it is now.

The Arch wiki is one of the best Linux-related resources out there. I used to look things up there all the time back when I was running Mint (which is basically Ubuntu under the hood). Ironically, now that I'm actually running Arch I think I'm looking things up on the Arch wiki less often than when I was running Mint.

Have you read https://www.sqlite.org/howtocorrupt.html? Section 1.2 addresses the exact scenario you wrote about in "Backups are a file copy". You got lucky with your testing, and didn't manage to copy the database in the middle of writing a transaction to the WAL file. Backing up an SQLite database with `cp` can produce corrupt backups if you lose the race condition, an unlikely but possible scenario,

Your later advice about "VACUUM INTO (backupfile)" is good, though: the SQLite manual guarantees that that's safe. But it's not safe to back up with `cp` if there are transactions currently writing to the DB: that has a chance of copying files in an inconsistent state, resulting in a corrupt DB and data loss you don't realize has happened until you restore the backup and find out there are some rows missing, or some rows have an invalid mix of old and new data.

There's some bad advice in the article:

"Backups are a file copy. [...] you can back up a live SQLite database, under write load, without stopping anything."

This is straight out of section 1.2 of https://www.sqlite.org/howtocorrupt.html. Yes, you can do that, and sometimes you will end up with a valid, non-corrupt backup. But it's timing-dependent: lose the race and you'll end up backing up a partially written transaction, making the backup corrupt. They didn't end up losing that race when they wrote the article, but that doesn't mean it is safe 100% of the time.

The section later on about running "VACUUM INTO backup-$(date +%F).db" is 100% safe, though: SQLite guarantees that you'll get consistent state if you do that.

"use loose; footloose; kick off your Sunday shoes"

Found what appears to be a minor mistake you may want to fix: "This means that a dangling reference easily results in a reference to the wrong column" should probably be "... a reference to the wrong row". (In the paragraph about SQLite's tendency to re-use ROWIDs).

Ah. When that one commenter said "Wealth is having economic power over the people around you", I ignored him, because that's a ridiculous definition, not to be taken seriously. Any argument that arises out of that definition is inherently flawed, since one of its premises is false; it would be like saying "Since the moon is made of green cheese, therefore X". The definition of wealth that nearly everyone agrees with is, well, I'll just cite the opening line of https://en.wikipedia.org/wiki/Wealth because it's a pretty good summary: "Wealth is the abundance of valuable financial assets or physical possessions which can be converted into a form that can be used for transactions." Or, to put it in simpler terms, wealth is the stuff you have that you can use to satisfy your desires. Anyone trying to make it about economic power over people is trying to redefine the standard definition for his own purposes... and that's usually a big red flag warning you of a bad argument.