I’m still reading the machine generated transcript of the video. But to keep it short:

The author was messing with ISBNs (international standard book numbers), and noticed invalid ones fell into three categories.

• Typos and similar.
• Publishers assigning an invalid ISBN to the book, because they didn’t get how ISBNs work.
• References "hallucinated"¹ by ChatGPT, that do not match any actual ISBN.

He then uses this to highlight that Wikipedia is already infested by bullshit from large “language” models², and this creates a bunch of vicious cycles that go against the spirit of Wikipedia of reliability, factuality, etc.

Then, if I got this right, he lays out four hypotheses (“theories”) on why people do this³:

• People who ignore the limitations of those models
• People seeking external help to contribute with Wikipedia
• People using chatbots to circumvent frustrating parts of doing something
• People with an agenda.

Notes (all from my/Lvxferre’s part; none of those is said by the author himself)

1. “Hallucination”: misleading label used to refer to output that has been generated the exact same way as the rest of the output, but when interpreted by humans it leads to bullshit.
2. I have a rant about calling those models “language” models, but to keep it short: I think “large token models” would be more accurate.
3. In my opinion, the author is going the wrong way here. Disregard intentions, focus on effect — don’t assume good faith, don’t assume any faith at all. Instead focus on the user behaviour; if they violate Wikipedia policies once warn them, if they keep doing it remove them as dead weight fighting against the spirit of the project.

Done!

twaróg

That’s actually a great idea — gotta try it next time. (I’m almost sure the locally sold keschmier is the same as twaróg, but worst hypothesis homemade requeijão does the trick.)

Ah, I stole the pic from some random site. They do it this way because it’s more presentable. But it’s better to spread the “sauce” across the pasta, mix it a bit, and then add the strawberry pieces. (Or add the strawberry pieces and then mix it, your choice.)

Truskawki z makaronem.

• 150g pasta. Preferably something that sticks to the sauce better, like fusilli or farfalle. Spaghetti is a bad idea.
• 500g strawberries
• 4Tbsp of sugar
• a drop of vanilla extract
• 200g sour cream. Yoghurt if you want something healthier.

1. Boil the pasta as usual, except you’re only adding 1/2 of the salt you’d otherwise use to the water. (You want some salt for the contrast, but not enough to turn this into a savoury dish.)
2. Clean and chop the strawberries into eights. Then reserve, like, 1/3 of them aside. Mash the rest with the sugar and vanilla extract, add the sour cream, mix everything well.
3. Add pasta to plate. Then strawberry mix over pasta. Then the reserved strawberries over everything.
4. [Optional] Chill it before serving.

This should be enough for 2 people. It tastes surprisingly nice.

And upon sharing this recipe, I can hear my ancestors… some rolling in their graves and saying “che schifo”, some giving me a thumbs up, and some asking if I could add yucca meal to the dish (no).

Now picture a gangster rapper dressed as Santa Klaus, with an eyepatch on one eye and a bottle of rum, saying “yo ho ho ho”.

Perhaps due to my heavy consumption of Japanese media, my views are biased. But frankly? I think Western design tendencies are the ones being weird here.

Note quotes are out of order. Also, that by “West” I’m including the Latin America I’m from.

“The West has an aversion to information density at times,” says Shoin Wolfe

Indeed, in a country preoccupied with safety, information overload is a part of daily life.

I think the difference is caused by advertisement: Western advertisement is so obnoxious, noisy, bossy, that it is bound to cause even more of a cognitive load than the Japanese counterparts. Western ads boil down to selfish arseholes screeching “are you too stupid to follow simple orders? I told you to consume it!” into your ears, while flashing loud lights. The following excerpt reinforces it:

“For the most part Japanese advertising has been ‘soft-sell,’ relying on the use of celebrities, attractive graphics, music or catchy slogans to sell products.” This was contrasted with “hard-sell” advertising, which uses “analytical logic, product comparison, or ‘annoy and attract attention’ tactics.”

And this might explain why people in the West avoid minor but still relevant info, while in Japan they seem to expect it:

In 2020, Lawson learned the hard way that too much minimalism can backfire. When the convenience store chain redesigned its branded product packaging to embrace negative space, it faced swift and loud mockery on Twitter. Users complained the now uniformly beige products looked too similar and gave no indication of the contents.

Moving on:

“(In web design) I think that negative space is an aesthetic, Western idea,” says Wolfe. “In the West, with physical products and just design in general, they have this idea that more negative space equals luxury.”

I have a better name for the so-called “negative space”: it’s “wasted space”. Space that failed to benefit the user.

And while some waste is unavoidable, I think the current Western design tendencies boil down to “cripple your design until you’re offering the users the bare minimum, before they stop bothering with it”.

“Because one symbol (of kanji) can compress what would be four to six letters in an alphabetic language, we grow up being accustomed to processing dense visual information very quickly,” says Akiko Sakamoto, a freelance UX designer and design strategist who works between Kyoto and Tokyo.

Under ideal conditions, the difference in scripts shouldn’t be relevant here. Sure, kanji is more informationally dense per character, but as a consequence your average kanji has more strokes than your average Latin letter. Thus requiring larger sizes for comfortable reading. And I think both things cancel each other out, forcing both scripts to convey roughly the same amount of info per area.

For the sake of example, contrast

• ⟨F⟩ vs. ⟨E⟩
• ⟨水⟩ vs. ⟨氷⟩ // mizu “water” vs. kōri “ice”

People here in the Fediverse are probably seeing all four characters the same size, right? Note how the difference between 水/氷 feels way subtler than the one between E/F.

…that is, under ideal conditions.

Sorry, I can’t hear what you said, because of all that *CLANK CLANK CLANK* noise you’re making. Clanker~

Crimea has a mountain range up south, really close to the sea, so it has a bunch of really short rivers. Chile is the same.

That said the river from the “Crimea river” meme is the Salğır/Салгир:

Info on how to disable AI anti-features from Firefox here and here.

I’m getting real tired of this shit. It isn’t just Firefox, or large token models; every fucking little thing nowadays has anti-features, that go against why I want that thing, that I need to work around, that makes every simple task a bloody chore.

That bloody browser is the cherry of the cake for me. I already avoid Chromium and similar Google trash as much as possible, as I know Google vultures the shit out of your personal info; now every bloody time Firefox updates I need to check two computers, two phones, and a few household electronics for potential shit to disable. Do I need to sell my soul to Satan to make things simpler again? Oh wait, souls don’t exist so I’m out of luck.

Yes, I’m aware, there’s LibreWolf. But that’s like makeup hiding your black eye, since it’s a “soft” fork of Firefox. And it doesn’t change that bitter taste on your mouth, that you can’t even browse the internet in peace any more.

Those are great to add to beans or chickpeas.

For example, feijoada relies on those and pig ears for thickness, while other ingredients (like sausages, jerk, bacon, smoked pork ribs, etc.) get added for the meat and flavour.

Give up all hope (of seeing the recipe), you who enter (the site)!

This is like a network of small problems: the advertisement model being awful, ads paying almost nothing per view, overabundance of cooking sites, Google’s monopoly on search, search engine optimisation, Google forcing you to feed its AI to show your site in search results, AI models and their intrinsic shortcomings (such as not understanding what they output)…

Cooking sites were the first victims because of how heavily they rely on SEO shit, and how people hate it. But others will eventually go the same way.

Now I regret following it with only two points, instead of three. LLMs love listing threes.

I typically used the em dash only when writing professionally, but because of this AI thing I’m doing it in general, just to see how it turns out. (So far it’s a good way to sniff out assumers.)

a ‘product of its time’.

Something like this, indeed. Or more like a product of the situation, plus a few laws - like network effect (the value a user derives from the OS depends on the number of users using it).

Note that not even the devs are to blame for this; it makes sense someone releasing commercial software would focus on the 70% (Windows), sometimes on the 15% (Mac OS), but almost never on the 4% (Linux).

It does, but this is a vicious cycle: small market share → devs don’t release Linux versions for their software → the software ecosystem is fragile → users who’d rather use Linux still need to use Windows → small market share. Anything countering any of those “links” weakens the vicious cycle, including Microsoft pissing off some Windows users; that’s why the penguin gets smug, because they know “Winrows is now an Agenric OS lol lmao” means slightly higher Linux market share.

Your typical Linux user gets really smug when learning about dumb shit Microsoft is doing with Windows. Just like that penguin in the OP. Because that dumb shit is making plenty Windows users consider ditching Windows for Linux.

One of those things is to force-feed AI into the users. Exemplified by Microsoft seeking to transform Windows into an “agentic OS”. People who don’t know how those systems work don’t want it; and people who do, even less.

“Windows is now an agentic OS”.

In the specific case of clanker vocab leaking into the general population, that’s no big deal. Bots are “trained” towards bland, unoffensive, neutral words and expressions; stuff like “indeed”, “push the boundaries of”, “delve”, “navigate the complexities of $topic”. Mostly overly verbose discourse markers.

However when speaking in general grounds you’re of course correct, since the choice of words does change the meaning. For example, a “please” within a request might not change the core meaning, but it still adds meaning - because it conveys “I believe to be necessary to show you respect”.

And AI sucks at that. If you interpret its output as a human-made summary, it shows everything you shouldn’t do — such as conflating what’s written with its assumptions over what’s written, or missing the core of the text for the sake of random excerpts (that might imply the opposite of what the author wrote).

But, more importantly: people are getting used to babble, that what others say has no meaning. They will not throw it into an AI to summarise it, and when they do it, they won’t understand the AI output.