yep!

A serious answer to this: this has also been a problem with programming and mathematics, but one which is starting to be solved, which is why engineers are freaking out en-masse. solving it for “fuzzier” tasks is an even bigger challenge but I am less and less confident that it is insurmountable.

oh right

Here it is: archive.ph/suv0 For some reason the NYT got him to review the last book; it isn’t very good

I hesitate to recommend it without revisiting, but I recall there was a Christopher Hitchens essay on this very subject

And it’s a bit of a mystery as to why not. There’s no obvious reason why, but things seem to divide neatly between low-dimensional polynomials (practical to compute) and exponentials (impractical)

We divide problems into those that can be solved in polynomial time (e.g. x^2, x^3) and those that require exponential time (2^x). The first are tractable, i.e. can be solved quickly. A problem solvable in x^1000 wouldn’t be tractable, but for some reason those don’t seem to come up in real problems

OK! I am not a mathematician but found plain old ChatGPT to be shockingly good at correctly explaining undergraduate-level maths and physics problems, and there are now plenty of examples of LLMs solving problems that are beyond me and definitely outside the training data.

Also, FWIW and contra OP, current LLMs are unreasonably good at solving difficult maths problems.

I think there is a very large difference between asking ChatGPT to do something and building a specialised tool which uses LLMs as an important component. The tools which are transforming software engineering are the latter and I frankly don't buy that indexing isn't a similarly tractable problem.

App Store pricing started out way too low because of the initial association with the cost of a song on iTunes — so there was never a premium tier and the low-cost stuff went FTP, which is poison. Also Apple encouraged it; some obscenely large slice of their services revenue is Candy Crush IAPs.

chatbots need to be situated within the longer history of "repl psychosis"

the more important question is whether or not LLMs can be usefully enrolled in the scientific process. i'd definitely answer yes, but "can" doesn't mean "will."

yeah, I think my main ask of something like claude code is to have more control and information about what's going on, especially with the subagents. I want to be able to selectively prune the past, toggle abilities and permissions etc. But no idea how to make it not-overwhelming to a human.

I'm really looking forward to the day when we collectively realise that in the field of context management, "append to chat log" is only one of many possible approaches. No idea what else is out there but it feels like a large possibility space to me

That’s true, but it’s not where their power comes from. Rather the opposite

I tend to think of SWE as a version of “labour aristocracy” for this reason. One of the few ways we are actually similar to engineers of yore is our social position within the firm. We’re not its owners exactly, but something like ownership accrues to the only people who know how to make it function

Sadly I think this is not possible because 1800 is the only 19thc year whose digits add up to 9

Honestly if you asked me “how many people would guess the correct URL and bother to try to access it in this half an hour window” I would answer “a small number” with not much more confidence than that. I wouldn’t be surprised if you told me it was 3 or 3000

If their analytics are really not set up to record requests which hit cache and the default cache time is 10 minutes, 32 hits in half an hour is too many. Also I’d be really surprised if whatever analytics package WP-engine uses did not take this into account

Do you know about Sid Meier’s CPU Bach?

L + Ratio + Send + Sync + Sized + Unpin

I think this might be the author half-remembering something about the Heisenberg-Schroedinger rivalry, where the elegance of waves vs the (at the time) unfamiliar-to-most-physicists clunkiness of linalg was a point of contention

Farewell to the Horse The relationship between horses and humans is an ancient, profound and complex one. For millennia horses provided the strength and speed that humans lacked. How we travelled, farmed and fought was d...

Recommend this if you haven't read it: www.penguin.co.uk/books/294612...

To be honest I just assume these things are fiction or a joke.

YouTube video by Developer Voices Elm & The Future of Open Source (with Evan Czaplicki)

Evan gave an interview last year where he was pretty vague but suggested he had been working on extending Elm to the backend: youtu.be/0SUM4869ODc

Without his notes!

Including for Americans; it provided the preconditions for US dominance of the continent

Speed/cost of generating the reply. Google are achieving an absolutely astonishing feat here in getting it to work at all, with a terrible end result

Rod Humble’s Stavka-OKH is essentially that, if you’ve not come across it before: www.rodvik.com/rodgames/STA...