jjoonathan

Haha, I remember making a "dust cover" to read GPU Gems 2 in class without controversy. RIP Nvidia demo girls, who appear to have been scrubbed from history for similar reasons at corporate scale.

...and today, ChatGPT can one-shot the meme and many of the shaders. I miss those books too.

https://chatgpt.com/s/m_6aa819fddec08191a90f2f84b63159ca

"Similar" is doing substantial work. If this is your only clue, it is likely to mislead you for at least 50% of the game, and I strongly suspect you will have fun anyway :)

Outer Wilds, the video game, does a brilliant job expanding on this theme if you're hungry for more. "There's more to explore here."

Warning: progression is gated behind knowledge so spoilers are worse than usual and The Algorithm will aggressively try to spoil you if you start poking too deep into "outer wilds" searches. If you like The Last Question and can fit a game in your life, Outer Wilds is a solid bet.

Outer Wilds vibes! I love it!

(It's a video game that does a brilliant job touching on similar themes to The Last Question. If you liked The Last Question and can fit a video game into your life, you will probably like Outer Wilds. Warning: if you start searching for "outer wilds," the algorithm will aggressively try to spoil you. Progression in the game is gated behind knowledge, so this is worse than usual. If you have trouble resisting the temptation to google past a rough description, it's a sign you should just jump in and play it. End recommendation.)

It is, but he got the macroeconomics backwards so enjoying it on an aesthetic level rather than a mechanical level is still the right choice.

The internet itself went through a similar growth pattern without astroturf. The original users were all researchers, which served as a strong implicit filter, and then the new users were students who had to be taught Netiquette every September, and eventually the floodgates opened to the public and the academics lost the ability to steer the culture in what was called The Eternal September (1993).

The same "initial implicit filter followed by gradual but inevitable reversion to the mean" dynamic explains your observations of early reddit without implying fraud, although it certainly doesn't imply the absence of fraud either. That said, "fraud" is probably a strong word for reddit astroturf in this present day and age where we have a (comparatively) planet-sized Dead Internet built on geological quantities of ads and slop.

No, that was 4o. Agreed about factual prompts showing less sycophancy in general. Less-factual prompts give it much more of an opening to produce flattery, of course, and since these models tend to deliver bad news in the time-honored "shit sandwich" I can't help but wonder if some people also get in the habit of consuming only the "slice of bread" to amplify the effect even further. Scary stuff!

Yeah, the heavily distilled models are very bad with hallucinations. I think they use them to cover for decreased capacity. A 1B model will happily attempt the same complex coding tasks as a 1T model but the hard parts will be pushed into an API call that doesn't exist, lol.

ChatGPT opened with a "Nope" the other day. I'm so proud of it.

https://chatgpt.com/share/6896258f-2cac-800c-b235-c433648bf4...

Good point. Still, I have to imagine that the engines themselves are dual use in some regard. GNSS or spy satellites maybe? These days it seems like everyone and their dog has a GNSS constellation, but it wasn't always that way.

Yes. The idea was to keep the engines (and engineers) out of the hands of the other likely buyers. You've seen how soviet military surplus gets around: the same channels work for rocket engines, and those engines work in ICBMs just as well as they work in orbital launch platforms.

I don't know how effective this was. Did it backfire by promoting economies of scale in a program that went on to sell to adversaries anyway? Did it murder the domestic engine programs and did that have knock-on consequences? I don't know if the policy was effective, but I do know that stopping "engine proliferation" was a widely given and accepted reason for those programs.

Offices revalued due to increase in WfH? Those dastardly Democrats must be at it again!

If it were 400BCE they'd blame writing

    this discovery of yours will create forgetfulness
    in the learners' souls, because they will not use
    their memories; they will trust to the external written
    characters and not remember of themselves

- Plato, quoting Socrates

Academics? Almost all of the free speech complaints I see these days come from the right, from people who would feel insulted if you called them an academic.

> There's no other scalable solution given existing copyright law.

(Informercial hands slip on screwdriver) Call Today for our $29.99 special grip that solves all those slippery screwdriver problems you definitely have! There is no other solution to the screwdriver grip conundrum!

No, practicality does not demand "binding shitty algorithmic decisions for thee, extreme latitude for egregious errors from me." Determinations don't need to be scalable to backstop a system of back-and-forth escalating claims that keeps the incentives correct for everyone at all stages: human beats algorithm, identified human beats unidentified human (note that at this point and all subsequent points rights holders have an enormous, automatic scalability advantage), identified human with legal commitment to consequences for being incorrect beats uncommitted human, and finally bump it to the legal system if all else fails, but by now everyone has skin in the game committed to their claims so none of the disagreements will be spammy.

This is all possible, it's not even particularly difficult, but it wouldn't create a cozy relationship with big rights holders which is what youtube actually wants, so instead we get "binding shitty decisions for thee, extreme latitude for egregious errors from me."

The feature lag wasn't the problem, the bugs were the problem: the only reliable OpenCL implementation was the one from Nvidia, but this meant it tended to drive people towards Nvidia rather than steal them away.

She turned the company around and got it on the right path, but in interviews I get the feeling that she might also be responsible for the "Hardware 1st, 2nd, 3rd, 4th.... eh, maybe software can be 5th" culture and AMD's deep denial that it has a problem.

https://news.ycombinator.com/item?id=40790924

That was OK for the CPU turnaround, but on the GPU front it completely shut them out of the first rounds of the AI party and maybe a trillion in market cap.

They can absolutely be recycled, lol.

Sounds like a big step back from the car charges at home and is "always full" without ever "filling up", except on roadtrips, where it schedules periodic rest stops.

Nobody cares about preachy commercials, except in that doing the opposite of what they preach makes you edgy and cool.

A big celeb getting caught doing something gauche and getting dogpiled in the press? That is what effective deterrent looks like.

It should be. "Allowed in the US" is a bar so low that you can get oil out the end of it.

Oh no. Ohhhh nooooo. No, no, no!

Xilinx dev tools are awful. They are the ones who had Windows XP as the only supported dev environment for a product with guaranteed shipments through 2030. I saw Xilinx defend this state of affairs for over a decade. My entire FPGA-programming career was born, lived, and died, long after XP became irrelevant but before Xilinx moved past it, although I think they finally gave in some time around 2022. Still, Windows XP through 2030, and if you think that's bad wait until you hear about the actual software. These are not role models of dev experience.

In my, err, uncle? post I said that I was confused about where AMD was in the AI arms race. Now I know. They really are just this dysfunctional. Yikes.

So I knew that AMD's compute stack was a buggy mess -- nobody starts out wanting to pay more for less and I had to learn the hard way how big of a gap there was between AMD's paper specs and their actual offerings -- and I also knew that Nvidia had a huge edge at the cutting edge of things, if you need gigashaders or execution reordering or whatever, but ML isn't any of that. The calculations are "just" matrix multiplication, or not far off.

I would have thought AMD could have scrambled to fix their bugs, at least the matmul related ones, scrambled to shore up torch compatibility or whatever was needed for LLM training, and pushed something out the door that might not have been top-of-market but could at least have taken advantage of the opportunity provided by 80% margins from team green. I thought the green moat was maybe a year wide and tens of millions deep (enough for a team to test the bugs, a team to fix the bugs, time to ramp, and time to make it happen). But here we are, multiple years and trillions in market cap delta later, and AMD still seems to be completely non-viable. What happened? Did they go into denial about the bugs? Did they fix the bugs but the industry still doesn't trust them?

Yeah, if this article wanted to be remotely even-handed, examples are plentiful. How many times has the Laffer Curve been trotted out to claim that reducing taxes on rich people will increase receipts, only for it to not happen that way? Or social cuts rationalized for the purpose of balancing the books, only for the savings to be spent on reducing taxes for rich people, leaving the books unbalanced? The purpose of a system is what it does.

Of course, the article didn't want to be even handed. Its primary intent wasn't really to explore POSWID as a general concept that applies nearly everywhere in politics. Its purpose is what it did.

That's like calling Arnold Schwarzenegger a weakling because Hafþór Júlíus Björnsson was stronger.

> given their protection is imperfect and not like actual vaccines

Since when were "actual vaccines" anywhere near perfect? The COVID vaccines were unusually good for vaccines, not unusually bad.

Even a low-efficacy vaccine can still be high-efficacy in a population due to herd immunity (you don't need 100% protection, you just need to push R below 1.0 so that the infected subpopulation sees exponential decay rather than exponential growth), but we don't even need to consider this nuance in the case of the COVID vaccines because they surprised so strongly to the upside.

They were able to assess the efficacy by looking for illness, not death. It takes some verrrry "special" framing to twist this into anything but an enormous success:

> BNT162b2 was 95% effective in preventing Covid-19 (95% credible interval, 90.3 to 97.6).

Yet the deaths clearly mattered when multiplied up to population scale, because COVID was able to fill up the hospitals.

The vaccines were so good at their job that the hospitals were full of unvaccinated people even after most people got vaccinated, and this astonishing feat of efficacy persisted for quite some time.

Normally I'm sympathetic to a bit of healthy cynicism, but when the official narrative is backed by extremely strong extremely public corroborating evidence, sorry, I'm not going to sacrifice the safety of myself and those I care about to indulge someone else's politically-driven blindness towards said evidence.

We have millions of devices in the field older than that on the same D-cell LiSOCl2 battery they shipped with. Some of them even take magnetic measurements with powered excitation!

In my corner of the world (electric/water/gas utilities), we had been using proprietary techniques that matched the "NB-IoT" description for 20 years, and they weren't new then. It was fun to watch the public dialogue walk the same path for a while and rediscover some of the main tricks, but there were other important tricks left untouched (broadcast firmware downloads, highly compressed debug logs, low power clock synchronization so that you can do timeslots for ultra low power, mesh heuristics) probably because they were a massive pain in the rear and the people who would have risen to the occasion saw that the major ecological niches were already occupied by players who got in 30 years ago when this all became viable. (Disclaimer: I didn't actively chase down NB-IoT talks so I might just have been missing these things.)

If your market is large enough that managing your own infrastructure is attractive (vs cellular), you were also able to afford a few EEs and RF guys 30 years ago, and your needs are already met by a competitive ecosystem of players who have been walking this path for a long time. It doesn't leave much to compete on except price, and you don't want to compete on price.