>>=

I have concerns from an accessibility point of view as we used to do everything hybrid and online, but it hasn't seemed to impact things too badly, and the students like the projects. The kids are also less bullish on GPT than they were a couple years ago. Ambivalent to negative.

In the classroom we've seen students less inclined to use AI and more honest about doing work themselves. That was achieved doing course design that makes plagiarism harder (we've gone back to blue book exams and prioritize group projects, while online quizzes are worth little, mostly study aides).

Our union recently won AI protection language for instructors. A first and it's thanks to a good friend of mine who bargained v hard for it. An issue is that administrators just don't understand the technology, so having experts on the bargaining team helps a lot. Hoping the language will spread.

I can't claim I'm a particularly good programmer. A look at my github will show you as much. But I value projects that are clearly very carefully designed, and the mark of careful design may only translate if the project is built under the control of the builder.

I see slopcoding as an extension of the kitbash ethos. When slopcoding, the programmer substantially gives up control over their program logic; reading one's "own" code comes to be like reading someone else's. But unlike libraries, which are usually deterministic, slop may have arbitrary errors.

The program logic, you want to write yourself and make clear, because you need to know what your program is doing, and be in control of it enough to debug it successfully.

But the important detail here is these are handling things like numeric computing, windowing, rendering, neural simulation. These are generic functions and operations that are specialized enough not to be trivial to just write myself, and generic enough not to be part of my program logic.

I want to pick a few that are generally useful, and use them across my projects, except where compactness is a concern. Then I might pick something more specialized, especially if it's well engineered and working with it feels pretty clean.

There are reasons why we want libraries; I wouldn't insist we all run in the opposite direction and subsist on a diet of glibc and our bare hands, but I prefer to use a small number of performant libraries I understand well that handle behaviour it'd be hard to guarantee the correctness of myself.

But what really bugs me is that it takes the programming out of programming. All the programming is abstracted away from the programmer, leaving them with an opaque API and some documentation.

Kitbashing contributes to bloat. It can also make applications buggy, if they aren't selective about what libraries they use and how those libraries are made to interact. Code quality, malware up the supply chain, and understanding how your software works are all concerns.

This was probably worst in the NodeJS ecosystem, where dependency trees became incomprehensible, but R is pretty bad for it too. Most languages with dependency managers will tend to avail themselves more to kitbashing, but some have worse patterns than others.

A programming phenomenon I found pretty frustrating in the last decade of my life is something I've tended to call "library kitbash". In a library kitbash, almost everything the program does is outsourced to imported software libraries. The program is just stitched together library calls.

What is the relationship between information loss and n(d) on a function computed between two populations? If we have n = kd then as k -> inf we should expect perfect representation and as k -> 1 we should expect a large amount of noise and bias.

Little Nengo things I don't know the answer to and should probably just send an email about or math out: when creating a population we decode as a vector on R^d, we choose the number of neurons n. Typically, we have that 10d ≤ n ≤ 200d, with n = 50 as the default...

I say "should probably". I will submit a PR tomorrow after I've done some grading, and after that I'll probably spend a week reimplementing it in spiking cells just for the trick shot. It's remarkable the amount of things that are incredibly easy to do that no one bothers with lol.

The VSA crowd should probably look into vector databases cause the way we do cleanup memories is what those in the biz call "the naïve approach" and I won't have my methods called naïve by people who generate slop slide decks about RAG all day. We can actually just implement a known logn solution.

"If you know so much about it, why shouldn't it be you?" because I have to jam another half-functional programming language into a vector space while struggling to understand how to read normalization semantics that's why

There were complaints on the forum that the standard is too complicated but you can literally just break it into core protocol and extensions and suddenly making something practicable doesn't sound so bad. That said I'm pretty sure the WP implementation covers the whole spec and it's 1500 SLoC.

Continuing to NOT post about cogsci but have we considered that sqrl never got the attention it deserved and if some work went into completing and polishing the RFC and making an open-source OAuth provider we could just never have to think about password security again.

Wow! Didn't know the European Union had run-ins with the Chiss. Those Bruxellois/es sure do get around

Oh, wouldn't you know, I forgot about Altan

As I recall, Soon isn't in that one

fwiw I like seeing your art a lot and I don't think you're wasting your time at all

Bro is your legal name Harry Fishman? That's incredible.

I'm only praying it's as self destructive as it looks. If this is to be the downfall of American power, good riddance!

They say arrogance is the graveyard of empires. America pivoted to a soft power orientation in the '80s because it realized coups and invasions weren't panning out. Obama changed military doctrine to rely more on proxies, because you need locals to hold power. Anyway I'm sure this will go fine.

Sorry I can't get enough this is too much of a self-own

"Blowback isn't real" says guy supporting government militarily retreating to its own hemisphere as costs of empire rise, which hasn't done a large-scale ground deployment since Iraq because it's too expensive, which cut the NED since American soft power is failing due to its hypocrisy, which...

That's fine for a civilian who's neither a politician nor a diplomat or IR person. If you're aiming to say how countries *should* behave internationally, it isn't very helpful to name who the good guys and bad guys are, because it's ultimately a matter of perspective, and doesn't help you win.