Various notes

Today was kind of scattershot and weird. I got a lot of work done, but it was interspersed with naps, reading sessions and all sorts of stuff. As a result, my “work notes” today are just a few unrelated notes.

On Focus

I’ve finally gotten a copy of Johan Hari’s “Stolen Focus”, and it’s really good. In particular, it is touching on a bunch of different topical areas that I’ve been thinking about myself, but weaves them together in a nice and coherent way. I’m less than half way through so far, but suffice to say that the takeaway is good so far.

In particular, I am trying to be careful to let my mind wander a bit. Take the brain out for a walk, so to speak. It’s already paying off in terms of background processing turning up creative ideas while I’m not paying attention.

As Simple as Possible, but no Simpler.

I was stumbling around the Internet last night looking for a solution to a problem when I encountered the Quite Okay Image format, by way of Mim Hufford’s Jai module for it. Gotta say, I love stuff like this. The process that led to the creation of QOI is described in a quite snarky, but spot on way in the original blog post about it, but generalized, it appears to be something like:

  1. Observe the state of the art.
  2. Decide it is unnecessarily complex.
  3. Make a stripped down, minimal replacement.
  4. Be astonished by how close to optimal it is, while also being substantially cheaper.

I wonder if this recipe has a name. It should. It’s like the technical flipside of what Venkatesh Rao observed about legibility – things you cannot do to socieities, but you can do to technologies.

It’s deeper than “do the simplest thing” or even “do the cheapest thing”, because it involves a meditation on the marginal utility of complexity. QOI is not the simplest possible format – that would be a raw pixel list, probably – but rather, it’s just a very simple format that includes exactly the amount of complexity required to be able to remain competitive with PNG without becoming so complex that it suffers.

In some ways, this is similar to the process that Patrick J. G. Stiennon suggested in his book “The Rocket Company”, and SpaceX have been following, in a certain way, in that industry. There are other examples in other industries.

Being able to make things as simple as they can be while still remaining technically competitive, and realizing benefits elsewhere in the process, is a really important idea.

There are some technical problems with QOI though. In particular, it bothers me how hard it looks like it would be to do SIMD processing on it. You could probably do some inside a given chunk, but because the previous pixel value is an important factor it might get in the way of parallelizing. The other technical problems I saw mentioned on a cursory browse around the Internet yesterday are less convincing. For example, complaint that it doesn’t compress quite as well as the most advanced formats in the world is in particular uninteresting and misguided: the goal is not to be the most advanced format in the world.

Anyway, the world needs more of this kind of thing.

Taking down NFTs

I have been meaning to write up a proper takedown of the endless procession of bullshit that is NFTs and blockchain culture in general, but I never get around to it because doing so is almost as much of a waste of time as NFTs and blockchain themselves. However, I do appreciate it how many good people have taken it upon themselves to be the bearers of bad news to the cryptobros, and in particular I really like Folding Ideas' very clear, methodical takedown of the entire scene: Line Goes Up – The Problem With NFTs.

I have been hounded for a few weeks by some people who want me to show up on their podcast about bitcoin stuff, and I have agreed to go, against my better judgement. I tried, while I was in parliament, to be even-handed in my treatment of cryptoassets, drawing attention to predictable and obvious problems while also remaining open to the virtues of innovating with the technology. Unfortunately, even then, the writing was on the wall: the incentives to use cryptoassets for scams, fraud and criminality are so much stronger than the relatively few and far between legitimate uses of blockchain technology. And four things are true of every single reasonable-sounding use of blockchain I have ever seen:

  • They are based on claims of legal backing or other merits that are either demonstrably false in practice, or, on closer inspection, rely on actual legal contracts that could be executed just as well, if not better, without involving a cryptoasset at all.
  • They are in all meaningful terms orders of magnitude more expensive to hold, exchange and manage than an equivalent use case that doesn’t involve a cryptoasset.
  • Their non-blockchain based equivalent systems would have many merits beyond the blockchain-based solution, such as the ability to fix data errors, roll back erroneous translations, prevent abuse, and not waste hundreds of terrawatthours of energy each year.
  • And finally, the blockchain itself could be replaced with a regular transactional database, to gain on the order of six orders of magnitude more throughput capacity while losing zero of the actual security benefits in practice.

And while I do agree with my old friend Vinay Gupta that the idea of having computers do things automatically instead of humans is frequently (but not always) a good idea, I disagree with the straw-man involved in that as a counterargument to my criticisms, specially when it shows up with the sneakily implicit assumption that “doing this with a computer” somehow must imply the use of cryptoassets.

(And while I will grant that the Avalanche blockchain does in fact appear to have a demonstrated throughput capacity of 4500 transactions per second, I will also note that that is several orders of magnitude slower than SQLLite3, which is not a particularly fast database.)

Anyway, whenever I point out things like this, I am faced with cryptobros claiming that I don’t know what I’m talking about, don’t understand, or even that “the matrix has me”. One thing I am never faced with is evidence to the contrary.

Because of this – but not out of any personal need to flex – there are a few things I find myself having to point out to people who try to lecture me on why cryptoassets are in fact the future. First, I was on the Metzdowd Cryptography mailing list back in 2008, saw the original email when it came in, read the whitepaper, thought it was interesting. But I quickly saw a bunch of technical problems with it, most of which have not been resolved. Some have. The original version relied on IRC as a poor mans peer-to-peer messaging, for instance. What worried me more was that it could not scale. I didn’t care much back then because virtually nobody was using it and I was doubtful such a poor technology would take off, even though the idea was fascinating. Unfortunately, I have a poor track record of predicting what technologies people actually end up getting excited by. There are many reasons I don’t work in marketing, I guess.

Secondly, I understand the underlying cryptographic technology – in particular, I have made my own implementations of chunks of it. In fact, earlier today, I was updating my old SHA256 implementation because I wanted to implement a HMAC version of it for something I’m doing. Am I a cryptographer? No. But I have a pretty solid understanding and a pretty substantial amount of real-world experience. This is only important to note because, in general, when I talk with cryptobros, they tend not to have the foggiest idea about how the underlying technology actually works, and are therefore largely incapable of understanding why I giggle a bit at the idea of staking the aggregate future value of an economy on there not being any exploitable collisions on SHA256.

And thirdly, I also understand enough economics to know why collisions on SHA256 are the least of their problems. But rather than belabor all the obvious points about deflationary systems, the misplaced theory of value, the liquidity traps and so on, let’s jsut say that the pure scale of scams and fraud on the system is all the evidence needed on that point.

Point is: people who have actually bought into the libertarian techno-utopian version of a pyramid scheme are in absolutely no position to accuse me of being uninformed.

Anyway, watch the Folding Ideas video.

A Bathtub of Books

I find it oddly difficult to find books written for a post-intermediary, pre-expert level of expertise on most subjects. I’ve been looking for a good book covering modern techniques in meteorological simulations. There tends to be a bathtub distribution of books where there are lots of simple (and often simplistic) introductory books that assume little to no pre-knowledge, and a reasonable number of (shockingly expensive) books that assume deep expertise.

I would like a book on meteorological simulation techniques that assumes that I know things like how linear regression and matrix multiplication work, but do not assume that I have a good mental model of the relative merits of different data structures for spatio-temporal clustering.

That last bit is just a random topic I was thinking about, not something I actually saw in a book. But the point still stands. Where is the medium difficulty level?