Vaguely technical

academia is Important. having a structure that allows people to pursue knowledge creation outside the normal pressures of capitalism is beneficial and we should be willing to subsidize it, as a country and a civilization.

the humanities are an important part of this, not an afterthought - history, philosophy, and yes even sociology are very important subjects to study.

academic dishonesty and a culture that tolerates it and advances people for reasons other than the quality of their work actively undermines both the function of academia (knowledge production) and the public backing for academia at a time when it is under threat by right wing forces that wish to destroy it.

You don't need data centers to serve Tumblr. Or your gps maps. And for sure not to use a gig jobs app.

You need a server, indeed. That server is basically a computer that needs to be somewhere with a good internet connection. If the site or app has a lot of usage, you may need several of them, rarely more than the lower two figures. That's it.

"the internet we know" doesn't rely on data centers. Data centers, as we know nowadays, basically were invented by Amazon, who started selling them "as a service" maybe 15 years ago. Before that, companies used to, well, have their own servers. Often in their own office buildings. Sometimes in MUCH smaller facilities dedicated to host servers.

But the modern concept of data server? That is needed mostly for data mining and selling processing power as a service, so shitty startups can try to invent the next torment Nexus without needing to hire people who actually know how to work with physical servers nor spending big money on hardware.

But most of the apps and websites you use everyday? Any shitty laptop could very well be used as their server and support a couple of hundred concurrent users without breaking a sweat.

Read "gig jobs using cloud apps" and thought I meant Uber Eats

Unironic Torment Nexus reference in a serious discourse post in 2026

Implication that companies running their own data servers locally is a more energy efficient model than renting cloud VMs

That said! My perception is there is a material uptick in the goal of new data-center construction as an investment in LLM compute (of course, this may be nothing more than a perception), which is the specific use-case that I don’t personally agree comes out on the right side of the cost-benefit analysis environmentally speaking*because* of the datacenters which are built as a response to VC-based demand, as opposed to consumer-based: ie AWS saying wow our ec2 reserved instances are such a high percentage of our total actual compute usage, it would really behoove us to have some more physical servers to avoid Insufficient Capacity errors.

Allegedly - according to McKinsey, in 2025 the workloads for data centers were pretty split; approx 50/50 between standard web-hosting (predominantly video delivery, I assume) tasks and whatever they define as an “AI Workload.” My understanding is they mean both inference and training task loads. They do predict this will swing in favor of AI workloads in the next few years. That seems true to me, but maybe they’re wrong, who knows. But again, this trend isn’t fundamentally usage-focused. This is compute spent that people *believe* will get them a solid ROI.

My counterpoints are:

The reasons companies are spending a big chunk of GDP building new data centers is because demand for serving existing LLMs and training new ones is very high. This is because LLMs are a phenomenally useful technology already that continually get more useful with each new generation (which harnesses more and more compute), and continuing this trend is good. Describing this as "VC-demand and not consumer demand" is not accurate at this point.

tl;dr: There's no evidence the interview that this was mentioned in even exists, no evidence the route of the cable has ever changed, and the whole story seems to originate from one dude's podcast.

Our first look in the comic of a map of OZ. Baume was one of the first, if not the first, fantasy writer to include a map of his world in his books; but his original map has one very interesting detail which eagle eyed viewers might see

Hint, look at Munchkin Country, ruled by the Witch of the East, and Winkie Country, ruled by the Witch of the West. They are inverted! This detail has been debated OZ fans and scholars for years with some thinking its just a typo, while others thinking it was intentional to show how odd OZ is. Most modern (and even later original books) publications correct this, so here's a better look at both the map of OZ as well as a map of the greater world OZ is in

Hall's Lie groups, Lie algebras and representations is good as an elementary introduction. However, it focuses (understandably) on matrix Lie groups which means that the theory is presented in a slightly less general way ar first, although towards the end of the book this is rectified I think. The book does treat sl2 and sl3 quite nicely iirc. Would recommend for a first meeting, but not if you know more.

Duistermaat and Kolk Lie groups is very geometric, too geometric for me, but depends on your background and purpose. It definitely talks about representation theory, but not always in a clear way if you do not already know the concepts. I distinctly remember associated vector bundles being called something different which confused me at first, skill issue though.

Fulton and Harris' Representation theory is a classic, and very good, book on representation theory. Especially if you care about the Lie algebra side of things more it is great (together with Humphrey's, although I havent studied that book much). However, since it deals with quite a lot of general theory, it can be hard to distinguish the essential features of the theory of cpt Lie groups. For example, I don't think the Peter-Weyl theorem is even mentioned? Even though for my applications that is one of the main reasons to care about representations.

Bump's Lie groups is also a good book, but it assumes basic knowledge of representation theory of cpt Lie groups in all but name. This can be supplemented by Hall though. For example, I think Peter-Weyl is proven around page 20 or so lol. So, better if you already know a bit imo.

To give some more context, I have done some representation theory of Lie groups and Lie algebras before and it did mostly stick to the representation theory if sl₂ℂ but we did do a bit about sl₃ℂ (though I intend to refresh that stuff cause it's been over a year now since I looked at this stuff).

Also I'm approaching this from the view of equivariant homotopy theory. Here, knowledge of the closed subgroups of the (compact) Lie group is important to things (there's actually a post I'm working on where I explain one such instance of this actually). So if there's some sort of general theory about understanding what subgroups can show up in a compact Lie group would be helpful to know (if there even is such a theory?).

Also the reason I'm interested in representations is again they play an important role in the structure of stable equivariant homotopy theory. So having a general framework for understanding the representation theory of compact Lie groups would be really helpful, cause we didn't really get to that point in my undegrad course. (Explanation for why they're important will be under the cut for anyone interested)

The point of this is to make clearer my reasons for wanting to learn more about Lie groups in the hopes that it might make the suggestions more narrow if you have the time?

the biggest hurdle honestly is probably just that at a certain point the rate of a device results in qualitatively different capabilities as you ascend the power-law curve. babbage could envision a general purpose computation device in the early 19th century, and the biggest conceptual leap from that to a cell phone is orders-of-magnitude improvement, but this gets you from "a way to speed up tedious by-hand computations for mathematics" to "a complete modern information-technology ecology," which is qualitatively fundamentally different. there are people who do grasp the Implications of that sort of thing intuitively (ray kurzweil predicting AI in the 2020s just based on increasing computational power) but there are a lot of smart people who have trouble with that sort of thing. and then of course "at a certain point of computing sophistication you can use statistical methods to train a machine on the whole corpus of human language and get a computer program that has enormous general-purpose capacities starting from nothing but blobs of random numbers" is also deeply counterintuitive. i still have trouble with it!

but i wonder how far back you can go and still find somebody smart you could Explain The Smartphone To. you would have to explain a lot more about the basic operation of electricity and magnetism to newton, hooke, and huygens; but they seem like they'd get there pretty quick. certainly newton would. and you don't have to delve into relativity here or anything that's really going to cause a philosophical headache: newton might be quite relieved that you can point to specific physical waves as your medium for electromagnetism, especially if you mention as an aside that that model works for gravity, too. i guess if you go back too far and you lose symbolic algebra--and concepts like "the square of the distance" become principally about geometry, and not a free-moving mathematical abstraction, you lose a major conceptual tool for the formalisms. and once you hit the 15th century you might lose inertia, momentum, and the idea that terrestrial and celestial matter all obey the same rules.