0xcde4c3db

I don't think "plateaued" is the right word, but I do feel like there's been something like a logistic curve compression in the difference between smaller and larger models as the field evolves. For inference at least, the scale of practical difference between a single high-VRAM GPU or SFF UMA box, a whole rack, and a whole data center seems to be falling far short of what we might have imagined just a few years ago. The conversations I've heard have largely turned away from breathless anticipation of the next frontier model and toward attempts at hard-nosed evaluation of which tokens are worth the cost.

It would absolutely make sense for someone who enjoys cycling to buy a pedal-assist e-bike, especially if they e.g. moved to a hillier area, sustained an injury, or just started having their strength/stamina decline with age.

(incidentally, the proliferation of unregistered/unlicensed motorcycles marketed as "e-bikes" has done enormous damage to the viability of e-bikes as a general category)

I doubt that's the only factor, but I agree that it's almost certainly a factor. The modern "tech" industry has filled our waking hours with unceasing demands to do something. Buy this product, watch this video, listen to this podcast, react to this social media post; even "practice self-care" has somehow turned into an obligatory virtue-cultivating task. We obviously had advertising before, but this is a different beast, something akin to living in a casino.

As Bo Burnham put it, "apathy's a tragedy and boredom is a crime".

True, but it's also the case that in some places and subcultures we're already at a point where meat with certain qualifications in some contexts is a Veblen good, and the rhetoric around it is very much a quasi-mystical blend of notions of purity and vitalism (see e.g. paleo/primal/ancestral diet influencers). That edge, almost by definition, can't be produced in a lab, and if we were really being purely rational/utilitarian about these things, I doubt we'd be eating as much meat as we are today.

What really matters is that there's a coherent policy for assigning addresses, which is absolutely compatible with having multiple DHCP servers on the same subnet (e.g. each server configured to hand out non-overlapping IP ranges).

I'm not a copyright lawyer, but my understanding is that you do not own a license when you buy a CD or cassette, you own a copy (strictly speaking, a "phonorecord" in US law). Copyright doesn't regulate the sorts of typical uses that the owner of a copy would engage in, so a license is not required. When you see those warnings on VHS/DVD/Blu-ray that they're licensed only for private home viewing, what they're telling you is that you don't have a license to do anything controlled by copyright such as broadcasting, publicly exhibiting, or reproducing it.

Software is significantly more complicated and gets into issues of physical permanence/transience and specific exemptions/limitations to normal copyright law (see e.g. 17 USC 117).

> Worse, in the context of DHHs rants, it comes off as racist - I don't get the sense he wants to throw out Norwegians from Denmark, for example.

I think "primarily" is doing some lifting in "Denmark is primarily a country for the Danes", along with his vagueness about what it means to be of "the Japanese" (Yamato/Zainichi/Ainu/Ryukyuan/Bonin) or "the Brits" (English/Scots/Welsh/Cornish/Irish/Manx/Jamaican/...).

Right; it's not political in the sense of "vote for candidate/party/policy XYZ", it's political in the sense that Omarchy is distinguished by being a vehicle for DHH's opinions, and by promoting Omarchy one can promote the general idea that DHH is a respectable guy with good opinions while still remaining firmly situated in "technical" subject matter.

Also, the book that popularized "hello world" was very consciously advancing a programming-first pedagogical agenda, as opposed to building up a theoretical foundation first:

> The only way to learn a new programming language is by writing programs in it.

The problem is that once you draw "the line", then what? Warehouse a bunch of helpless people in overcrowded prisons? Bus them to some other less-draconian municipality? Forcibly drug them into quasi-submission and then turn them loose again until the depot injection wears off? Or perhaps we should "just kill 'em", as Fox & Friends cohost Brian Kilmeade once suggested?

Yeah, it's almost impressive how often hypotheses for what's causing rising obesity rates simply ignore actual data. I haven't been able to find it again, but a while back I stumbled on a journal article that was basically some epidemiologists ranting about this. IIRC there are whole categories of proposed causes that you can eliminate just based on the fact that that the rise is correlated across age groups.

The implementation is superficially silly, but I think this can also be read as a meditation on exactly what distinctions we're making when we use terms like "program" vs. "process" or "input" vs. "dependency". Perhaps a cousin to the idea that compression formats represent specialized instruction sets for bitstream generation, or to static single assignment. It makes explicit what is usually implicit in emulator "movies", that each frame has the state of the prior frame as a dependency, even though that relationship is normally thought of as a side effect of the passage of time.

The underlying principle of translating those implicit temporal dependencies into explicit data dependencies very well might have applications in optimizing high-accuracy computer/console emulators. The usual implementations iterate over some variable that represents the passage of time and compute all possible side effects, which is expensive. Many compilers already do a related static optimization by eliminating intermediate variable mutations that can be proven to have no side effects.

(also, if you're not familiar with the "tool-assisted speedrun/superplay" tradition of optimizing video game input sequences as a flavor of metaprogramming, see https://tasvideos.org/NewcomerCorner)

I doubt there are many dyed-in-the-wool Republicans as such, but I'll bet there are many who still identify strongly enough with the old-school Republican messaging around rugged individualism and supply-side economics that they view the median Democrat as actively destructive by comparison to the merely flawed median Republican.

I'm far from being one of those myself, but I imagine the argument would be that any regulation one could propose to address this is bound to be worse than the problem itself, in terms of chilling effects on legitimate speech and commercial activity.

That might run the risk of backfiring, by teaching people that hallucinations are obvious and thus easily avoided. Perhaps one of those things that only happens to other people who aren't so smart/skilled/savvy (car crashes, scams, mental illness, etc.).

We've been doing variations of "educate the general public about healthy dietary choices" for roughly as long as the obesity epidemic has been around. Shouldn't it have worked by now, if it were going to? As purportedly modern, scientific, data-driven societies, why do we keep engaging in this impotent self-flagellation?

As far as I understand it in retrospect (having grown up in a strictly PC household and later investigating "retro" platforms where SCSI was standard), SCSI was not inherently expensive. It's more that for PC platforms, ATA provided a hack that was cheaper at the time, which led to a bifucation of the market. ATA roughly turned HDDs into a subset of the ISA bus, which in turn originated as a buffered version of the CPU bus. That led to SCSI being at a premium on PC platforms because the target market was limited to customers for whom ATA wasn't good enough (mainly because they wanted more drives or longer cables than the thin translations of ATA could support).

> The COVID vaccine was an example of very heavy-handed enforcement

Do you have some concrete examples of this? I recall a lot of hyperventilating about the possibility of "vaccine passports" (complete with "mark of the beast" references), and a handful of instances where people got angry about COVID vaccination being folded into (decades-old) policies covering mandatory vaccinations for school/work, but not much beyond that. I carried a mask in case places had mask policies, but it never even occurred to me that I might need to prove that I was vaccinated.

And even supposing that an administration doesn't fully trust Congress (or its ability to execute through parliamentary gridlock), there are a bunch of people in the executive branch whose whole-ass job it is to investigate this stuff, and could legitimately be directed to shift their priorities without Congressional approval. Oops, scratch that; there were a bunch of people with that job. [1]

[1] https://en.wikipedia.org/wiki/2025_dismissals_of_inspectors_...

Probably because it's implausible, trivial to fake, and there's no way to conclusively prove it's real without the original tweet still being up.

Being in bed with tax preparation companies is probably the main thing, but I also vaguely recall a statement by someone years ago (perhaps Grover Norquist or Dick Armey) that filing tax returns should be kept annoying simply for the sake of keeping people angry about taxes in general.

I was sure this was wrong, but it's true: according to official state counts, Donald Trump won 49.80% of the popular vote to Kamala Harris's 48.32%. The top 5 was rounded out by Jill Stein (0.56%), Robert F. Kennedy Jr. (0.49%), and Chase Oliver (0.42%) [1].

[1] https://www.fec.gov/resources/cms-content/documents/2024pres...

(yes, I'm aware of the irony of linking to federal agency data in this thread)

And Larry Ellison wants us all under the eye of AI cameras so that "citizens will be on their best behavior". I almost used the word "panopticon" there, but Ellison is proposing something strictly worse, in that there's no hope of the cameras not being watched.

> Or do devs around the world just have to bite the bullet and learn enough English to be able to use the majority of tools?

I'm a native English speaker, but from looking at various code bases written by people who aren't, I gather that it's basically this. It wasn't too long ago that one couldn't even reliably feed non-ASCII comments to a lot of compilers, let alone variable and function names.

It's easy for any one agency or police department to cook the books, but I don't think that scales to the kind of huge shift across multiple categories and data sources shown in the article.

Yeah; it's not strictly accurate, but in a similar sense that it's not strictly accurate to call the typical copper Ethernet connector "RJ45", to say that a UDP "connection" occurred, or to say that a modem connects to a "DB-9" serial port.

I suspect the root of the problem is that over time, "proxy" has become strongly associated with application-layer protocols like HTTP, and after that shift it wasn't obvious what to use for something lower-level that encompassed a wider range of protocols/endpoints/conversations. In principle, "tunnel" would probably have been better (and a legible metaphor to boot), but that's just not how things shook out in practice.

It's been argued that similar dynamics also inflate executive pay, although I'm not well-versed enough in the overall economic policy debate to know how well-established this actually is [1].

[1] https://www.epi.org/publication/reining-in-ceo-compensation-...

There's also "Free File Fillable Forms", which is part of the weird compromise that tax return prep companies like Intuit and H&R Block negotiated to prevent individuals being allowed to file directly with the IRS. It's actually operated by one of the companies in the "alliance", but it's set up like a digital version of the IRS forms (up to and including the fact that you need to make your own copies of everything because they delete all accounts/data every year).

> Hmm. I'm ready to believe you, but 2257 doesn't seem totally infeasible to comply with in a VHS world.

I'm not saying that it was supposed to be impossible to comply with per se, but rather that it was supposed to be impossible specifically for the porn industry as anti-porn crusaders imagined/alleged it to be at the time. There was a bunch of drug-war-like rhetoric about how porn had become more violent, exploitative, and outright criminal since 1970, when the previous government commission on the subject reported that porn wasn't an important social problem and should not be restricted for adults.

The commercial incentives around this are also terrible, given that the prevailing assumption is that a private third party will do the attestation as a for-profit service. It's practically guaranteed to add cost and liability for the porn vendor as well as risk of leaks, tracking, and data brokering for the consumer. It may as well have been designed to shut down free porn altogether.

And maybe it was, considering the history of porn regulation in the US. If you look at the requirements and contemporary rhetoric around 18 USC 2257 (yes, the thing that basically all legit porn sites with US operations have a disclaimer for), it was pretty blatantly intended to render the porn industry unable to operate. Porn producers were never actually intended to come into compliance, but rather presumed to be unable to (due to the ridiculous procedures and the supposed omnipresent use of underage, undocumented, and coerced performers).

Lowe is known and (probably rightly, on average) lauded for his frankness in light of his expertise. He's sort of the chemistry equivalent of a Raymond Chen or John Carmack. I like him a lot overall, but I think he badly failed to "read the room" in this particular case.