woopsn

JavaScript is a colossal solution to the demands of product, design, business, users and other stakeholders. Crockford was one of the first to effectively sell it to engineers. It has developed into a nice language with modern features and an incredible debugger, and objectively (to the people we serve, whether users or other clients) the browser is a much richer, more capable and reliable platform today as a result.

NPM has by its nature and on many individual occasions done lasting damage to the project, the reputation of the language and professionalism of the JavaScript community. There is a lot of frustration over this, as well as the "10,000 frameworks" issues (which seemed overwhelming for a time), these and other experiences justifiably put a lot of people off who decided never to get into JavaScript because it seemed like a shitshow.

How much criticism of the web platform today is about its vestigial language features? Very little it seems, compared to 10-20 years ago when the language actually sucked and was considered a joke. It doesn't make sense to stop using it because it's mature and in place, we invested in this thing (as far as the web goes). No, JavaScript's biggest issues are apparently cultural.

Well, the sequence 0, 2, 4, ... , 2k, ... is indeed simple, can be recovered starting from the value at an arbitrary index (eg the last one announced). As can (3k), (5k), etc...

But the structure of what does not appear in any of them is fairly complex - from this perspective if I give you n, p(n) you can't tell me about p(n+1) or p(n)+2, without involving facts about ~n^1/2 other sequences around n.

Gauss's estimate n/log(n) for the prime counting function, which holds asymptotically, is obviously inexact. As is the logarithmic integral. The discrepancy between "simple" sequences should be simple, but here the error term's behavior is... hardly that.

With respect, this is an epic undertaking. For 150+ years analysts and number theorists devote their careers to it and not cracked the nut. Although there has been great progress.

Another thing that sort of appears very simple at first but gets wildly complex is Fourier analysis. It's just a way of writing functions with a trigonometric basis. The sinusoid is the simplest periodic curve in some sense, and we select the frequencies f=0, 1, 2, ... Okay but this is a basis for... what? It's not simple. Another 200 years.

The two are connected. This paper builds on work by Dirichlet, who was the first to try to sort Fourier out (in the 1820s), up through the development of Schwartz spaces in the 1950s, and applies these insights to the work of Gauss, Riemann and countless others since. And we still don't understand the structure (eg bounds) of the error term!

> Getting answers on the web can take a lot of effort, often requiring multiple attempts to get relevant results.

Don't get me wrong, search has become extremely problematic, ... but how much effort does it take really? Compared to writing a letter, reading a map, walking half a mile etc?

The google results for "music festivals in boone north carolina august" are completely adequate. If not then you search again, sure. What makes that "a lot of effort" compared to "asking follow-up questions like you would in a conversation"?

According to Sequoia Capitol there is a $600B hole in this sector at the moment, which continues to grow.[1] They need to invent something akin to the global smartphone market, over the next 2-4 years. Some thing new that solves significant problems for people that aren't already solved for free.

[1] https://www.sequoiacap.com/article/ais-600b-question/

Mozilla is up in a bunch of shit at this point. They have a half a BILLION dollar nut to raise annually, and money from Google covers the whole thing. It's ridiculous.

https://en.m.wikipedia.org/wiki/Mozilla_Corporation#Finances

The discipline informed by science is computer engineering. They have to worry about, e.g., things melting.

I'm critical of computability as discussed here to some degree, in that many of the algorithms in effect do not compute anything but instead melt a computer, or the earth, or would otherwise deplete the environment of negentropy without answering the question.

In standard mathematical analysis there is no behavior "at" infinity, rather we may have some decent picture of what happens at each n (a series' remainder, truncation error, a bound on the value of certain terms, etc.), and talk about what happens to it for greater and greater n, as you mention. This is that other science, numerical methods and analysis.

Is it fair to compare the two? If I recall this right every register machine is associated with a system of diophantine equations, with the number of variables about equal to the register count, and the size of the system equal to the length of the program. This is a precise representation of the machine at any step n for a given starting state.

To do computer science really analytically you have to solve huge systems of dio equations or something equivalent. Beyond sounding hard, this was Hilbert's 10th problem and has long been answered as not possible. We've been puzzled by these systems all along since antiquity.

I am pretty ambivalent towards thinking of BB numbers as scientific results, for the basic assumption of every working model we have (that erasure of a bit is possible) would break before they could compute a result. But at the same time, we are not the first constructivists -- the situation is really puzzling still. I think we no longer realize how shocking some early-mid 20th century results were to mathematicians and logicians of the time. This still hasn't been resolved, nor on the "other" side in physics. We'd love to have the tools we have in applied math in computer science, but that's what all the fuss is about.

Keep in mind, the "hole" takes ChatGPT and Copilot/Gemini/etc into account at a rate of $5-10B per product. The remainder of $500B is about the size of the global cell phone market. Where will it come from?

I tried a few techniques, in terms of accessibility I think "old" ways are the best here.

    <div class=flowchart>... crazy modern shit ...</div>

    <ol class=sr-only>
      <li id=step1> Water on stove.
        <ol>
          <li>Boiling? Proceed to <a href=#step2>Step 2</a></li>
          <li>Otherwise return to <a href=#step1>Step 1</a></li>
        </ol>
      </li>

      <li id=step2> Cook dumplings in water. Wait a minute.
        Floating? Proceed to <a href=#step4>Step 4</a>.</li>
        Otherwise proceed to <a href=#step3>Step 3</a>.</li>
      </li>

      <li id=step3> Stir dumplings. Return to <a href=#step2>Step 2</a>.</li>

      ...
   </ol>

(or start with the anchors/list, and progressively enhance)

What is the concern? These businesses didn't exist two years ago. There's nothing to reckon. Google's had a wildly profitable model forever, their "existential threat" is LLMs not petering out.

Personality seems to rule the product and venture capital side of things, and at the end of the day their side is in charge.

There is also a theory that as you approach the horizon you'll be incinerated by a kind of hidden firewall. This apparently contradicts GR - but on the other hand if you're allowed to fall in it seemingly contradicts QM.

A big part of the problem is illustrated by the fact we are now trained to google e.g. even basic information about our own podcast. Big tech gets so much more out of us than we get out of them - their products are entrenched largely by perception. You can use another search engine. Open a spreadsheet from your Documents directory. Run 7-13B models yourself with llama.cpp. Save a bookmark. Wait three seconds. Whatever you need to do you can do. The data brokers, ad men, media corporations etc. now identified as "the technology industry" are (explicitly) in the business of wasting precious time, not saving it.

You might look up the sagas of Rocket Lab and other small launch providers if you're interested in what it takes to put ~200lb into LEO today. It's way beyond dad and son stuff, still incredibly hard, but not so much anymore that you need to be Boeing/Lockheed/SpaceX/etc if not a national agency. This is a recent achievement.

"Our primary fiduciary duty is to humanity."

No... A good number of folks will even go so far as to say that "all we do" is token prediction too. It's worth noting -- OpenAI founder Elon Musk claims in a lawsuit that the company has achieved AGI. Make of that what you will, but certainly there are many people on this site who believe in the general potential of LLMs.

Clarifying, the specter of a hidden animal will usually take the form of a diffuse sparkle or blur, typically hovering off to the person's side and somewhat above them, and as a result when carried through to the "other side" cannot possess what remains of the person in that domain (because they are returned to the origin in turn).

In most photos with a recognizable subject, spectral energy will be concentrated around the origin (the upper left corner) as it is here

https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_pr...

The same is true for the DCT of the woman. Meanwhile, the subject of a photo is typically located towards the frame's center. This helps minimize interference between the space and frequency domain data in the composite, thus preserving kitty's expression when the transform is inverted

https://substackcdn.com/image/fetch/w_1456,c_limit,f_webp,q_...

(and vice versa for the woman)

Yep, that's it.

Well the paper shows that looking for signatures of photovoltaics is an impractical strategy if not impossible.

The only other games in town, if we're talking huge amounts of power, and ignoring the speculative "beyond", are fission and chemical combustion. Unlike PVs these can be expected to alter the planet's atmosphere distinctively, and that is the thing we can actually observe during transit events, in contrast to panels (and dams, wind turbines, geothermal etc) down on the surface.

Apparently it's worth bothering just to remain inconspicuous. Aliens are clearly doing everything possible just to hide themselves from Earth.

Fascinating mission. I don't understand how microlensing events resulting from primordial black holes can be differentiated even statistically, given that they "can't be formed by any known physical process", but in any case it's exciting to discover there's much more of something up there than we knew. Even if they are "just" rogue planets.

From behind the paywall -- NYT sued them, and so OAI will pursue for now licensing arrangements with some major media outlets at least.

Yes significantly. But take for example the stock market, which is obviously of interest, to OAI in particular, but which cannot over the medium-long term be extrapolated accurately from historical data.

Mostly games on Oculus. I haven't played online socially, just with friends IRL - one has a really sweet room all set up for VR. Also back in the 90s-00s, on the more primitive arcade systems. I haven't looked at the "metaverse" yet.

It's not like I'm against fun, and if there is more emphasis in the "metaverse" on meaningful social connection, that's good. Games and activities that get you moving around are also a very positive development. Wii was one of my favorite consoles (but was hardly immersive).

My brother is a former pilot and sprung for a full setup so he could play aviation sims. I haven't done a movie yet.

As I said I have nothing against headsets or immersive digital experiences. I do think in this day and age it's worth being more protective of the word "reality", compared to the 90s. Not just in a political sense. The core business of Facebook is advertising and data brokering. I know I might sound like a quack but they really do want to control perception of reality and for us to stare at their screen for as long as possible. Partly because there are ads there, and partly because as we do they stare back at us and take measurements. So that's why I don't accept their "Reality Lab", "The Metaverse", etc. marketing language, and to the extent they control VR it is not VR. They don't want to build the metaverse a la Snow Crash, it will be more like the social -> news feed -> analytica history, plus raking app developers.

> to avoid the debris reassembling into a satellite would require an amount of energy larger than the Moon gravitational binding energy, estimated to be 1.2 × 10^29 J.[2][3][4] This equals a bit less than 600 billion 50-megaton nuclear bombs, such as the Tsar Bomba,[5][4][2][6] roughly equivalent to the full energy output by the Sun in six minutes

That's 33.33 quadrillion GWh. Or about two trillion times the annual energy consumption of Earth.

If we somehow harnessed windfall energy on that scale, hmmm what to do, what to do... let's blow up the moon!

There is no way to store or transport the energy though, it corresponds to hundreds of millions of tons of mass fused every second. Instead you would have to somehow beam or guide the full power of the sun towards the moon for several minutes. Pretty cool light show, for those who dare to look up... but don't miss!

Cool idea. What does the expression/spec language look like? I would guess it has to be mapped... not straightforwardly. It seems like your goal is maximize throughput, but then the datapath would have to be planar, no?

In many cases practitioners already have appropriately non-magical language, for clarity in communication.

*-reality interfaces are "immersive", they don't alter or recreate reality.

What artificial intelligence is is (statistical) "inference".

The cloud are various *-providers, "hosting providers", "storage services", etc. In aggregate, "external" or "3rd party" or "outsourced" (computing) infrastructure.

I said it elsewhere -- what frustrates me is that the rest is marketing, but some engineers are willing to accept and even reproduce it for some reason. When Facebook says they're releasing a new reality development platform for the metaverse though, we should roll our eyes, even if the technology is exciting.

We outsourced our CRM system last year to a certain company. Or you could say we migrated it to the cloud. Either way we had to audit data policies, establish contracts with them, transfer data to them, ensure the same functionality and workflows were replicated, look at the quality and reliability of their system, assign points of contact, make redundant some roles within our company, etc... Many of the things you allude to are implied by either term.

Time-sharing and remote job entry were also reasonable back in the day. You bought time on a computer to run workloads. I don't think they fit so well with the "cloud computing" model in a modern sense technically, but I do envy them for being descriptive, logical terms that describe what is going on. We got the cloud from Google CEO Eric Schmidt who said to a conference in 2006

> What’s interesting is that there is an emergent new model. I don’t think people have really understood how big this opportunity really is. It starts with the premise that the data services and architecture should be on servers. We call it cloud computing — they should be in a “cloud” somewhere. . . . And so what's interesting is that the two – "cloud computing and advertising – go hand-in-hand. There is a new business model that's funding all of the software innovation to allow people to have platform choice, client choice, data architectures that are interesting, solutions that are new – and that's being driven by advertising.

- https://www.google.com/press/podium/ses2006.html

It's strange to me that as engineers we accept and even promote marketing language such as this.

Reconfigurable logic may be used to implement fairly small models, in applications where they're already employed and adding a coprocessor specifically for ML is either infeasible or doesn't make sense. As the paper mentions, you need hard logic blocks for arithmetic (if not floating point), and these are always in short supply. In DSP applications I've worked on we used the fpga for timing and i/o, to control jitter and sample from many ADCs in parallel - but apart from some filtering then ran the numbers on an adjacent non-reconfigurable core. You can get huge chips with a lot of hard logic built in, but they're relatively expensive compared to fpga + traditional coprocessor with shared memory. The high end application specific chips are shockingly expensive - worse than GPUs because there is comparatively little market for them. We had one evaluation board that was like $100K iirc.

I'm not concerned about trademarks. I'm concerned about what isn't trademarked -- language which instead we are all supposed to adopt and throw around as much as possible. As I said: * reality, AI, cloud, etc.

That's the point though. Manufacturing, logistics, payroll, research, advertising, legal services, ... -- outsourcing certain functions to specialist, often (but not always) cheaper or more efficient companies is a typical business practice. The term is broadly applicable and used where ever this is the case. Since there are well-known business and organizational risks associated with outsourcing, it is conspicuous that we "migrate to the cloud" instead.