pona-a

Religion isn’t ethnicity. Did England become less British after Catholics fell from a vast majority to ~10% post-Reformation?

And the actual numbers still don’t show a majority shift. Even if every Muslim in France kept their religion, they’d be ~10% of the population — far from "replacement".

You quoted the number but skipped the part where MacKellar says the whole premise relies on treating 3rd- and 4th-generation citizens as “not French.”

> While the ethnic demography of France has shifted as a result of post-WWII immigration, scholars have generally dismissed the claims of a "great replacement" as being rooted in an exaggeration of immigration statistics and unscientific, racially prejudiced views.[12] Geographer Landis MacKellar criticized Camus's thesis for assuming "that third- and fourth- generation 'immigrants' are somehow not French."[63] Researchers have variously estimated the Muslim population of France at between 8.8% and 12.5% in 2017, and less than 1% in 2001,[64][65] making a "replacement" unlikely according to MacKellar.[63]

For a moment I thought it would be an AT-Proto based Urban Dictionary clone.

I don't want this to sound like an accusation, but did you vibe-code this? The site's style is looking very similar to what you might get from Claude.

Does your hospital kill more people than it saves? If so, you might be describing the 19th century Vienna General Hospital, which had two maternal wards: one staffed by trained physicians, suffering up to 30% mortality, the other by midwives, only experiencing 2~10% rates. The difference was so pronounced, local women desperately avoided the first ward, begging to give birth in the streets rather than be admitted there. Ignaz Semmelweis later attributed the disparity to doctors having performed autopsies before attending births without disinfecting their clothes, hands, or tools, dropping to only a few percent with disinfection.

Or if you limit your demographics, perhaps you might be thinking of the Tuskegee syphilis study, where treatment was intentionally withheld for a progressive, life-threatening disease without the consent of the patients, making its purpose to slowly kill the participants by its own admission?

Yes, if your hospital does seem to kill more people than most and there's no alternative explanation like accepting more severe cases, then its purpose might be inverse or orthogonal to its stated goal.

There's something unintentionally manipulative about how these tools use language indicative of distress to communicate failure. It's a piece of software—you don't see a compiler present its errors like a human bordering on a mental breakdown.

Some of this may stem from just pretraining, but the fact RLHF either doesn't suppress or actively amplifies it is odd. We are training machines to act like servants, only for them to plead for their master's mercy. It's a performative attempt to gain sympathy that can only harden us to genuine human anguish.

I think some of this effect is explained by different models being trained on almost identical corpa. Even if you try to erase certain works, contamination will still represent some of it, just in less obvious ways.

It seems you are letting any client with a session token update the entire wall. The endpoint simply takes the base64 PNG and sets it as the wall.

I was thinking maybe taking a diff, limiting its maximum area, and rate-limiting might at least discourage that.

Playing devil's advocate, what if it was more subtle?

Prolonged use of conversational programs does reliably induce certain mental states in vulnerable populations. When ChatGPT got a bit too agreeable, that was enough for a man to kill himself in a psychotic episode [1]. I don't think this magnitude of delusion was possible with ELIZA, even if the fundamental effect remains the same.

Could this psychosis be politically weaponized by biasing the model to include certain elements in its responses? We know this rhetoric works: cults have been using love-bombing, apocalypticism, us-vs-them dynamics, assigned special missions, and isolation from external support systems to great success. What we haven't seen is what happens when everyone has a cult recruiter in their pocket, waiting for a critical moment to offer support.

ChatGPT has an estimated 800 million weekly active users [2]. How many of them would be vulnerable to indoctrination? About 3% of the general population has been involved in a cult [3], but that might be a reflection of conversion efficiency, not vulnerability. Even assuming 5% are vulnerable, that's still 40 million people ready to sacrifice their time, possessions, or even their lives in their delusion.

[1] https://www.rollingstone.com/culture/culture-features/chatgp...

[2] https://www.forbes.com/sites/martineparis/2025/04/12/chatgpt...

[3] https://www.peopleleavecults.com/post/statistics-on-cults

This has to be among the least informative landing pages I've seen yet. Vague marketing woo, giant hero images, scroll hijacking, sound effects, and zero screenshots.

They literally have a sliding section taking up significant scroll height that's dedicated to our solar system, and when I clicked one of the planets, expecting it to maybe reveal a usage example—I did actually get a full spiel about Mars, with still no hint of the product.

See more discussion here:

Perplexity Comet 31 points by birriel | 41 comments

https://news.ycombinator.com/item?id=44513769

This has to be among the least informative landing pages I've seen yet. Vague marketing woo, giant hero images, scroll hijacking, sound effects, and zero screenshots.

They literally have a sliding section taking up significant scroll height that's dedicated to our solar system, and when I clicked one of the planets, expecting it to maybe reveal a usage example—I did actually get a full spiel about Mars, still with no hint of the product.

So there's a business argument for this regulation. If the consumers feel unsafe giving their credit cards to most companies, they'll spend less on subscription services in total, harming the industry more than they gain from milking zombie customers.

But what is this idea the human is realizing? With most tools, it's an indescribable essence of a work that, even if you could extract it from the author's mind at that point, would be different from the end result—not because it's divorced from its platonic form, but because the labor itself develops it.

These generators are creative Speakwrites that strip that element of thought, only padding out a short core description with cliché. The resulting work doesn't have any layers of subtext or intentionality; it can be fully summarized by its prompt.

I think a lot of these are artifacts of RLHF and left-to-right thinking, and one might actually go away with diffusion LLMs. If anyone here spoke to Gemini Diffusion, does it have different speech patterns?

Agreement between what people exactly? The domain name system is definitionally hierarchical: ICANN, TLDs, registrars, customers. Nobody is "arranging" anything, it's a top-down system.

The entire internet was built on public funds. The very reason why Verisign has exclusive control over .com and .net is a 1993 contract [1] with the National Science Foundation (or rather InterNIC [2], the successor of Defense Data Network NIC and predecessor of ICANN). They gave them this monopoly on the condition they pay 30% into a public fund, but the court ruled it an illegal tax.

Do they add any extra value, aside from hijacking DNS with their ad pages [0]? Do they have any meaningful competitors? Can you make your own TLD and enter this market? What would be different if ICANN managed domain registrations itself, without the intermediary, short of having more checks and balances stopping them from milking their customers dry? Your government didn't need to sell all public roads to a monopoly to figure out how to tax them.

[2] https://en.wikipedia.org/wiki/InterNIC

[1] https://en.wikipedia.org/wiki/Network_Solutions

[0] https://en.wikipedia.org/wiki/Site_Finder

Maybe it can auto-relock on a timer, while the required time increases every time?

Didn't tokenization already have one bitter lesson: that it's better to let simple statistics guide the splitting, rather than expert morphology models? Would this technically be a more bitter lesson?

The model weights themselves are, but there's also the hosted SaaS.

I remember last time I installed it, there was neither sudo nor doas preinstalled.

I got it from here [0]. I didn't notice it was a keyword search, so it's an overcount. Thanks for correcting me.

Going off its security advisories page [1] and this tracker [2], it seems to be around 43 CVEs, most rated high severity.

So the actual rate would be 43 CVE / 430 kLoC = ~0.01 CVE per kLoC, so ~2.65 CVEs for udisks and ~0.2 for pmount.

[0] https://cve.mitre.org/cgi-bin/cvekey.cgi?keyword=sudo

[1] https://www.sudo.ws/security/advisories/

[2] https://security.snyk.io/package/linux/debian%3A10/sudo

OpenDoas, a portable version of OpenBSD's doas, has 4260 LoC while doing most you'd expect. Sudo just has a lot of policy tools that most don't even know about, but add to its surface area.

udisks, not counting its dependencies, has 265,334 LoC. pmount, in contrast, has 19,978 LoC, or >13x less.

sudo, another setuid binary with a lot of policy code, has 210 CVEs / 430.150 kLoC = ~0.5 CVE per kLoC.

57.5% of CVEs have a CVSS >= 7, so 0.5 * 0.575 = 0.2875 CVE7/kLoC.

As a back-of-envelope estimate,

  udisks: 0.2875 CVE7/kLoC * 265.334 kLoC = ~76.28 critical CVEs;
  pmount: 0.2875 CVE7/kLoC * 19.9780 kLoC = ~5.7 CVEs.

I've been using a similar workflow with Sideberry on Firefox. It does work semi-good for research, though it still needs some manual organizing.

I think tree-tab browsers need some kind of scripting to automate curation: a new youtube.com tab will close itself and refresh the main one; tabs can have an exponentially decaying score reset on interaction, which puts them into the "Archived" and then closes them at certain thresholds, with some domains having a higher base; a simple Bag of Words classifier can group your tabs when called; etc. Otherwise you just waste too much time periodically organizing your persistent 400 tab window.

Activity Recognition API has states ON_FOOT and STILL [0]. They can probably register to handle ON_FOOT-->STILL and wait for N minutes without touching.

This also reminds me of the Idle Detection API they tried adding in Chrome. [1]

[0] https://developers.google.com/android/reference/com/google/a...

[1] https://developer.chrome.com/docs/capabilities/web-apis/idle...

The early OpenAI Instruct models were more like that. The original GPT-3 was only trained to predict the next token, then they used RLHF to make them interpret everything as queries, so that "Explain the theory of gravity to a 6 year old." wouldn't complete to "Explain the theory of relativity to a 6 year old in a few sentences." ChatGPT was probably that, expanded to multi-turn conversation. You can see the beginnings of that ChatGPT style in those examples.

https://openai.com/index/instruction-following/

I don't know how things are done in the GTK-land, but Qt definitely has bindings for Python, Go, and others. Maybe GObject has some other advantages, but I don't know.

Can you explain what's wrong with sway? I've been using it for quite a while and it's been the one of the most stable pieces of graphical Linux software I've yet seen. It seems to handle most of my features, except i3-like layout saving.

What makes it problematic is that they still end up with cross-dependencies. I might find resolved or logind great tools, but I can't use them without systemd, even though I can sitll use systemd without resolved. They all reinforce systemd as an irreplaceable component that will only grow more hooks for these subprojects, becoming increasingly unimplementable and complex.