Aransentin

Sure, but the structure seems quite ChatGPTish as well, with e.g. the bullet points and section choice. You wouldn't get that by just faithfully translating a french source text from English.

This article has a strong ChatGPT smell. Things like "in the world of", "let's dive into", the bullet points, "conclusion" section, etc. Anyone else have the same feeling?

> We try to hide some of the differences between arches when reasonably feasible. e.g. Newer architectures no longer provide an open syscall, but do provide openat. We will still expose a sys_open helper by default that calls into openat instead.

Sometimes you actually want to make sure that the exact syscall is called; e.g. you're writing a little program protected by strict seccomp rules. If the layer can magically call some other syscall under the hood this won't work anymore.

I found Gwern's review of that quite good: https://gwern.net/review/crumb

No discussion in the article of how likely this is to be simply a coincidence?

There's a ton of ancient artifacts depicting humans in fanciful and non-realistic ways, and a ton of diseases that affect the body. That two of these will happen to superficially coincide is basically guaranteed.

The similarity here isn't even that great; if the statue had a huge goitre or something I'd be more inclined to buy it, but here it just looks like a little chubby squinting dude.

I live in Sweden and don't think I've touched physical money for about a decade. Cash is more or less on life support by now, with a small and dwindling list of places you can even use it anymore - basically grocery stores and e.g. hairdressers that cheat on their taxes. And illegal drugs, I suppose, though from recent news where people get busted by paying dealers using Swish (a Swedish Venmo-ish service) I'm not sure it's used even for that.

There are some hard-to-handle edge cases when doing display length truncation in Unicode, e.g. the character U+FDFD or "﷽" is four bytes but can be very long depending on the typeface*, so "completely" solving it is quite hard and has to depend on feedback from your rasterization engine.

(*Rendered version on Wikipedia: https://commons.wikimedia.org/wiki/File:Lateef_unicode_U%2BF... )

It should be noted that Xlib has been a compatibility wrapper around XCB – the "modern" X11 client library replacement – for about 15 years now, so it might make more sense to just target that.

(The only thing on the top of my head that absolutely requires Xlib is some niche Vulkan stuff; e.g. VK_EXT_acquire_xlib_display unfortunately has no XCB analogue.)

> very unlikely

Actually, what are the chances of this system – if it existed – is used to move an asteroid into the path of the Earth and killing everybody within the next 100 years? (This would include scenarios like "the US president or industrialist who builds it randomly goes insane"). 1%? 0.1%? If the risk of an asteroid hitting the earth is lower than that we should clearly not build such capability.

Probably not, it's unironically too incomprehensible to be AI.

In practice, changing tag type is rarely done.

In any case it wouldn't be particularly expensive; you'd have to make a new member of the destination type which is as expensive as an append. The "hole" that remains in the original vector can be filled in constant time by taking the last member of it and moving it there, then shrinking its size by 1.

We first finished a Zig application three years ago, but that was a fairly small demo project which we didn't have to support afterwards.

My company uses Zig to write software for small Linux devices, such as set-top boxes.

The biggest benefit isn't exactly performance, but the absurdly tiny binaries it produces (very convenient if you are working with small flash disks!), and how the Zig standard library doesn't depend on anything other than the kernel, which means you can often entirely ignore all library issues; more often than not our software 'just works' when built for some new customer platform even though I've never even targeted that environment (or even CPU architecture) before.

I wrote a quick script a while ago to bruteforce English 5x5 Sator squares using a wordlist, and the only one that makes some limited sense is

  t i m i d
  i r a n i
  m a d a m
  i n a r i
  d i m i t

I.e. "[the] timid Iranian madam that Inari* dimit(=dismissed)".

* https://en.wikipedia.org/wiki/Inari_%C5%8Ckami

(Of course, when googling for it somebody else had done the exact same thing and found the exact same square before: https://github.com/daniele-rapagnani/sator-squares)

Note too that the paper states "We generated five images for each test image and selected the generated images with highest PSMs [perceptual similarity metrics].", so it actually directly admits that the presented images are cherry-picked at least once in a way that IMO is blatantly cheating (why take the best of 5 and not 5 million? Then the pictures will look even closer to the ground truth!)

If the pictures they showcase at the top are among the best results they got, it means it's double-cherry picked. In that case it wouldn't surprise me if you could get similar results from random noise.

Surely in many cities this will simply lead to people being able to afford even larger loans to bid on the limited stock of housing, driving the prices even higher. If you have an inelastic good you need to actually increase supply, boosting demand will just make everything worse.

Bayes' theorem is ruthless. Even if your AI is 99.99% accurate, if you have one true positive and scan everybody the vast majority of the people you flag will be perfectly innocent. The people deploying the system and police are ignorant of the statistics and base rates, thinks that it's only 0.001% likely you are innocent, and chuck you in jail because of it.

If 90% of the prompts are human couldn't you reach 90% accuracy by just picking "human" every time?

Because pretty much everybody that reads the article will have taken away a grossly exaggerated idea of what the system is actually capable of. If Stable Diffusion was intentionally added "for flair" and really is unnecessary, then I would absolutely say that the researchers were being intentionally deceptive.

Even if we do a massive goalpost-move and grant that the system is only identifying the label "dog" with a brain scan of a person looking at a dog, we would need to see actual statistics of its labelling accuracy before judging it in that way. If the images in the paper are cherry-picked(1), it could easily be only able to extract a handful of bits to no bits at all, and the entire thing could very well turn out the be replicable from random noise.

(1) Note that the paper even states "We generated five images for each test image and selected the generated images with highest PSMs [perceptual similarity metrics].", so it even directly admits that the presented images are cherry-picked at least once.

I immediately found the results suspect, and think I have found what is actually going on. The dataset it was trained on was 2770 images, minus 982 of those used for validation. I posit that the system did not actually read any pictures from the brains, but simply overfitted all the training images into the network itself. For example, if one looks at a picture of a teddy bear, you'd get an overfitted picture of another teddy bear from the training dataset instead.

The best evidence for this is a picture(1) from page 6 of the paper. Look at the second row. The building generated by 'mind reading' subject 2 and 4 look strikingly similar, but not very similar to the ground truth! From manually combing through the training dataset, I found a picture of a building that does look like that, and by scaling it down and cropping it exactly in the middle, it overlays rather closely(2) on the output that was ostensibly generated for an unrelated image.

If so, at most they found that looking at similar subjects light up similar regions of the brain, putting Stable Diffusion on top of it serves no purpose. At worst it's entirely cherry-picked coincidences.

1. https://i.imgur.com/ILCD2Mu.png

2. https://i.imgur.com/ftMlGq8.png

As an obvious example, I uploaded a photograph I took of a painting I have by Julie Hagen-Schwarz* that's been in private possession since 1882 and never put online: https://www.stableattribution.com/?image=7730efac-bf52-4077-...

It still finds ostensible "source images" for the art. So, yes, it's clear this service is pretty much bogus.

* https://en.wikipedia.org/wiki/Julie_Wilhelmine_Hagen-Schwarz

> first day of class, write a N word essay

Initially I thought you meant having the student write an essay about slurs, as the AI will refuse to output anything like that. Then I realized you meant "N" as in "Number of words".

Still, that first idea might actually work; make the students write about hotwiring cars or something that's controversial enough for the AI to ban it but not controversial enough that anybody will actually care.

Two minor points:

GPT doesn't see letters, just words and parts of words as entire tokens. I would not expect it to understand that "no pun in ten did" and "no pun intended" are similar, since it can't see that at all. The only way it could plausibly understand it is if that exact comparison has been explained somewhere in the corpus before.

If you continue a long conversation where GPT has previously made mistakes and has been called out for it, it will start predicting that it's supposed to emulate a text where the joke-explainer keeps making mistakes. This causes it to start failing on purpose. Make sure to completely restart the prompt each time to not fall into that trap.

Now the council official just needs to run ChatGPT on their side to decide whether to accept or deny requests, and we'll have automated the entire process.

The "blind orchestra auditions" story gets repeated a lot, but the study it originates from did not find any statistically significant effect (and even so the ultimate outcome was that men benefit from blind auditions, not women).

http://jsmp.dk/posts/2019-05-12-blindauditions/

Considering the staggering speed that image generation is improving, that 10% gap will only continue to close.

Starting e.g. an art education right now seems likely to be extremely nerve-wracking as your talents may very well be woefully obsolete by the time you graduate; the exception perhaps being those top-0.1% talents that will feed the models of the future with new material.

> You answer a question with the verb that was asked. ‘Do you like coffee?’ ‘Like!’ or ‘不like‘.

Surely this explains that Chinese bootleg Star Wars translation meme where Darth Vader's "Nooo" is subtitled as "Do not want"?

https://knowyourmeme.com/memes/do-want-do-not-want

My immediate reaction was that it was some sort of 4chan project, tounge-in-cheek named after their softcore porn board (also named "/e/").

I'd wager I'm not the only one thinking that.

How about the Turpan water system?

https://en.wikipedia.org/wiki/Turpan_water_system

An artificial oasis city built in the Late Middle Ages* on the Taklamakan Desert for travellers along the Silk Road, it provided irrigation by having more than 5,000km of hand-dug underground tunnels to transport water from the mountains; an IMHO extremely impressive feat of geoengineering for the time.

* https://link.springer.com/chapter/10.1007/978-3-030-00728-7_...