hyperthesis

He's just saying it's more like an "I" than his pharmacy-bot example. His concern is the future: what seemed "far off" now possibly might be in 5 years.

Humans have limited "working memory". We manage to cram more into it via hierarchical decomposition into "chunks", a single concept that is more complex inside.

I submit that not everything can be hierarchically decomposed in a way that's useful - i.e. any "abstraction" you try to force on it is more leaky than non-leaky; in that it doesn't simplify its interactions with other chunks. You might say it's the wrong abstraction - but there's no guarantee there is a right abstraction. Some things are just complex. (This is hypothetical, since I don't think we can conceive of any concepts we can't understand.)

An AI could have an arbitrarily large working memory.

Note: I'm talking about intuitive understanding. We could use it mechanically, just never "get it", cowering before icons, being the one in Searle's Chinese Room https://wikipedia.org/wiki/Chinese_room

PSA: This is much longer than pg's other essays. There are multiline gaps between sections that can seem like the end.

Finding the search terms is half the search. If you already know "Sparta", you're past that stage.

Or, by Douglas Adams, knowing the question is pretty important.

Misconceptions are insidious because you don't know what you don't know (e.g. boneappletea, france is bacon).

A private tutor, coach-like, can observe and correct, and fill in gaps. I think this applies to all fields of endeavour. It won't make you smarter, but gives you a foundation for the next level - where, maybe, some intuitive smarts of yours will shine.

Now, the only problem is simply finding a good tutor. You just need to know if they know what you don't know you don't know.

Linear time is fine for small n, and console commands aren't time critical anyway.

People famous for optimization also know where it matters. Optimization obscures.

Industries mature.

I kinda thought coding might be immune, because it's about information. But of course, you build on lower levels. Today, you can get outsized returns by applying LLMs to real problems. Also about information.

> fund’s purpose is ... demonstrating that prescribed and cultural burns are low risk, so private insurance companies will enter the market.

How does a fund demonstrate that? The linked legislation (https://wildfiretaskforce.org/prescribed-fire-liability-clai...) claims "rates of escape and loss are very low" already. Insurance companies' actuaries are experts at quantifying loss, so if it is low already...?

From the date of lost-contact (18 June 2023), there have been 13 deaths from mass shootings in the US. https://www.gunviolencearchive.org/reports/mass-shooting

I had a hardtime remembering if I didn't understand.

But I tried an experiment at uni (once...) to learn a subject solely by rote. I could typically recall about 7 out of 10 items. I was astonished I could do it at all. Unfortunately, I lost all my usual strengths in that subject: unable to generalize, unable to justify, unable to adapt.

Re: software technologies: non-leaky abstractions are the way to understand without details. Algebras are a great example: arithmetic, concatenation, boolean, Kleene, relational. Although they still leak (overflow, PCRE, etc), an idealized core plus ad-hoc crap beats all-crap. jq has an algebra of , and | operators (though it doesn't call it that)

\aside git: I wonder if git internals can be interpreted as an algebra (absent in UI)? TBF I tried to design a programming language around git internal operations (tree navigation and construction), but it was tedious to use. Maybe something analogous to a declarative SQL over relational algbra would solve this?

owner != founder

> Are humans limited to low-risk applications like that?

No, but arguably civilization consists of mechanisms to manage human fallibility (separation of powers, bicameralism, "democracy", bureaucracy, regulations, etc). We might not fully understand why, but we've found methods that sorta kinda "work".

> could have caused

That's why they didn't.

enhance!

But those people who bought expensive properties are the ones with the power to change the situation.

myspace managed to be killed

This is the argument for books against TV.

Shadows help ground images (e.g. characters in games seem to float without contact shadows). Seems simple, but requires a 3D model of the room to work properly. LiDAR

Lag between reality and virtual would be noticable side-by-side, no matter how much the "R1" chip reduces lag (12ms)... Ah! they might do the Guitar Hero trick, and delay the reality feed by the same amount, so they are exactly syncronized... see-through AR (like google glasses) can't delay reality.

The outward display of reverse faux-transparency is a nice touch, displaying a fuzzy image of the wearer's eye on the front. (Though not sure what to make of the back-side of the displayed image - could a user ever see it?)

I think this is the first major Apple developmemt without Jobs' involvement?

perhaps related: https://wikipedia.org/wiki/Learned_helplessness

> a singularity would almost surely be preceded by a world in which machines are 0.01% intelligent (say)

I'm not sure that fractional intelligence makes sense. Otherwise, Boole's Laws of Thought, mechanisms like clockwork, even books, even counting, and many other ideas would be seen as capturing some "fraction" of intelligence.

I think it's a threshold, like Turing equivalence: either you can compute anything or you can't. Once over that threshold, percentages make a difference, such as 10,000 seconds to generate 1 second of human-level intelligence.

There's a democratizing aspect to this position, that all humans possess human-level intelligence, not just geniuses (or researchers).

The bitter lesson, that data beats understanding, reminds me of the more-energy beats efficiency in "Review: Energy and Civilization" (https://news.ycombinator.com/item?id=36013309) which notes that the pursuit of efficiency suffers from diminishing returns, whereas abundent energy does not.

In a similar way, parsimonious concepts and knowledge, while personally satisfying to researchers, are inherently limited; whereas massive data (including self-generated data, as for alpha-zero) are not.

Knuth sides with actual understanding:

> [I shall continue] to devote my time to developing concepts that are authentic and trustworthy (https://news.ycombinator.com/item?id=36012360)

Inventions often seem to be simultaneous, as if the time had come for this problem to be recognized, and for the solution to be possible. However, I do think in some cases, a "great man" can make an invention earlier than otherwise - we should measure genius in units of time.

We wouldn't be stuck in 1950 without him... but maybe we'd now be in 2018 without him.

OTOH the "great man" fallacy encourages the intense and unpredictable work required for many inventions - why undermine that?

"Explicit" would seem to be "obvious" - is that necessarily true? Too much code can obscure what is happening.

Familiarity helps implicitness be understood. Some would argue lack of familiarity is the only problem with the languages you mention. Certainly, it helps: as you become more expert in anything, you start using short-hand and jargon, and skipping details.

Adoption is easier when it's similar to the familiar. Hence, "c-like". And python, psuedo-code-like.

But are some things easier to grasp, by nature? I think, "intuitive" often means familiar. And there are things that all human beings are familiar with, like space, trade and language. Perhaps this is where intrinsically obvious and familiarity coincide?

implicitness examples: precedence, type coersion, overflow, polymorphism.

But I suppose (e.g.) precedence doesn't arise that often; it "should" be familiar from high-school algebra; it can be made explicit (with parentheses or successive assignments).

> implicitness in language design

In a sense, all languages have implicit aspects. Can you characterize the problematic implicitness? (I guess python's TOOWTDI helps, when reading code).

I've been studying bidirectional transformations, and a survey paper notes that "relational" methods have died off (somewhat "declarative", where you specify how things should be connected, with the details implicitly worked out for you... something like set-builder notation), while various coding methods remain (where you code for forward transformation, and the backward transformation is automatically derived). They hypothesise it's because relational tools (libraries and projects) are difficult to maintain - but I suspect it's because they are difficult to use.

Which is a shame: something that automatically "does what you want" seems like a great idea! But being difficult to predict and diagnose - not being in control - is not great. So, to rephrase my question, what characterizes "good" automation?

Only for uniform probability density.

All that's needed is a tendency to sustain itself, instead of diminishing it. This isn't that unusual; consider waves. Structures also spontaneously arise, such as bubbles (foam), crystals, streams, rocks.

But if pre-life RNA existed, it would be hard to identify: very small, similar to life detritus, and quickly processed as food.

Stackexchange is most directly under threat (from the current "chat" AI UI).

Christensen's disruptive vs sustaining innovations is more descriptive than predictive. But if it's the same customers, solving the same problem, in the same way (from their point of view), then it's probably "sustaining" and incumbents win.

Different customers, problems, ways - and all bets are off. Worse, incumbents are dependent on their customers, having optimized the company around them. Even if they know the opportunity and could grasp it, if it means losing customers they simply can't do it.

Larry is thinking people will still search... right?

Very true, though for compilation, perceptively-instant is nice rather than a few seconds.

OK, so it's a normal unix release, and uses proot for a virtual directory structure. It says it only uses the terminal emulator from termux.

termux doesn't use proot - its distribution has all paths rewritten (usually in source) to cope with the android filesystem (and other changes). I'm always surprised that this hackery works so well!

Therefore, I'd expect identical programs to be slower in nix-on-droid.