But that response is grounded in the training data they've seen, so it's not entirely unreasonable to think their answer might provide actual insights, not just statistical parroting.

What do you mean? It is grounded on the text it is fed, the reason it said that was that humans have said that or something similar to it, not because it analyzed a lot of LLM information and thought up that answer itself.

LLM can "think" but that requires a lot of tokens to do, all quick answers are just human answers or answers it was fed with some basic pattern matching / interpolation.

There's nothing "basic" about the several months of training used to create a frontier model.

Can someone enlighten me what's been going on in the open-source office application space lately? Here we have LibreOffice and Collabora parting ways; meanwhile NextCloud used to integrate OnlyOffice until v18, then started integrating Collabora in v19 (and also in the recently announced stack for openDesk and "Office EU") but then the other day NextCloud announced they'd fork OnlyOffice to create EuroOffice, … which clearly neither Collabora nor OnlyOffice seem to like?![0]

Please just tell me what the canonical stack is that I'm supposed to use these days. I still have scar tissue from the OwnCloud vs. NextCloud situation…

[0]: https://www.theregister.com/2026/04/02/eurooffice_forks_only...

Agreed. "Prefer pure functions, push side effects to the boundary (functional core, imperative)" is at the top of my CLAUDE.md these days.

> The initial limitation to Google/Android is not great, we know that, and we have support for other OSs on our list (like, e.g., GrapheneOS).

GrapheneOS uses standard Android APIs for hardware attestation (as opposed to Google-specific ones), so why don't you just use those from the get-go?

They did. This is why Graphene works.

They, too, offer integrations for LLMs these days, presumably for better OCR and classification.

I've been using News Feed Eradicator[0], a Firefox/Chrome extension, for the same purpose and it's been working really well. For most sites you can configure what should vs. should not be "eradicated".

[0]: https://github.com/jordwest/news-feed-eradicator

> The $1 is measured in international dollars. This means it buys the same amount of goods and services in any country as a US dollar does in the United States. It is often used alongside purchasing power parity (PPP) data. The “time” refers to a day of life for anyone, at any age and in any circumstance — not just the hours worked by someone with a job.

So IIUC this "average poverty" (measured in time per international dollar) includes people living off social welfare? Otherwise, if it only included the working population, wouldn't we have

  average poverty ≝ (average yearly income* of the working population / 1yr)⁻¹

and so it should be inversely proportional to the average yearly income* metric mentioned in the article?

*) Adjusted for purchasing power, i.e. measured in international dollars.

From a linked article:

>For these purposes, income includes earnings from work, government benefits and other sources of money, and it is averaged among all family members.

Yes, it is supposed to include income from all sources.

https://theconversation.com/measuring-poverty-on-a-spectrum-...

Huh, I must have missed that. Thanks!

I don't think that's quite right, it would be the average of the inverse hourly wage not the inverse of the average hourly wage.

    >>> import statistics
    >>> 1/statistics.mean([10,30,100])
    0.02142857142857143
    >>> statistics.mean([1/10, 1/30, 1/100])
    0.04777777777777778

No, it's

average poverty ≝ average(1 / annual income)

Inversely proportional to the harmonic mean of average yearly income.

Ah, you're right. I added the (·)⁻¹ at the very end, when I noticed the units on both sides didn't agree, and forgot about the average.

> While standalone CSS is not yet Turing complete

Looks like it is, though? https://news.ycombinator.com/item?id=47558097

Oh yep, looks Turing complete just not performant enough for that use case. But that’s not an issue for APT-style attacks that take their sweet time. So am I off base here?

A wall hack is any technique allowing the player to see opponents through walls. Googling for screenshots or videos should give you a good idea of how it looks.

How is that setting supposed to carry over if I don't even have a Google account on my phone?

And even if I disregard that for a moment, what's up with the author being a mouthpiece for Google?

> Google's latest concession makes the sideloading controversy a big nothingburger

> Opting out is going to be even less of a problem than we thought

> This afternoon, Google’s Matthew Forsythe shares some answers to questions he’s gotten about the minutiae of how this process all works — and he’s got some very, *very* good news for us.

(emphasis theirs)

> Doing that once with every new phone already sounded perfectly manageable. But now Google clarifies that even that won’t be necessary, with the opt-out able to be transferred as we upgrade phones. That is maybe just the best news we could have gotten here, and hopefully it’s enough to calm everyone down about the sideloading-sky falling.