olejorgenb

Not sure if they are the "inventors", but https://spritz.com/ was where I first saw this ages ago

I'm in the middle of reading this long article, and it's really good IMO! (I usually don't post comments like this, but since there's no comments and the submission is soon completely burried)

Chat control 1.0

"A temporary derogation from the ePrivacy Directive that allowed (but did not require) providers to scan private messages of unsuspected users for potential child sexual abuse material."

Does that imply it's currently not allowed?

EDIT: apparently not enforced at least:

"Chat Control 1.0 expires

The legal ground for voluntary, indiscriminate scanning ends. Google, Meta, Microsoft, and Snap state they will continue scanning private messages regardless. "

I think the idea is that the images are decrypted by the client. See how Ente does it: https://ente.com/architecture

Of course - this sacrifice quite a bit of functionality since more or less all functions which require looking at the pixels need to be client-side. But to be fair - the client is part of the "app", so it's not "just" encrypted storage.

They said they only want to control one of the ends

Even opus refuse to discuss micro biology for more than a around 15 turns in my experience.

I also see more of these, but I'd say still around half require more click for "reject non-essential"

That would require a tax hike to fund though, which is not very palatable.

And either 80% of banners are not respecting the law, or the law managed to omit mandating making it as easy to reject as accept... Rejecting usually require you to enter into settings and sometimes click "reject" for every individual partner(!)

This wasp seems quite specialized towards ticks: https://en.wikipedia.org/wiki/Ixodiphagus_hookeri

Isn't this against the ToS though? I don't share my sessions with Anthropic, but I would be happy to share selected sessions to improve an open model.

Off topic, but the other articles are well made too. I enjoyed this one: https://moultano.wordpress.com/2025/02/24/you-should-make-cr...

"This is a good time to spare a thought for our red-green colorblind brethren. [...] it is to them that we owe the beautiful color of green traffic lights. The spectral requirements that make the green signals distinguishable from red in their eyes make them beautiful in ours."

https://archive.ph/rSUKz

throwing shitload of money to the big actors of a field

My reply was directed at this part. Based on my memory seeing ironcalc specifically getting funding. Unless they hide it well they are not a big actor. And the project looks interesting and worthy to me. (I see I should have omitted the nextgraph link as I'm not familiar at all with that project)

Few of the projects listed here seems to be big actors: https://nlnet.nl/project/index.html

Some projects funded by NLnet: Organic Maps, KDE Connect, KDE Plasma Wayland, Bottles (Builds on Wine IIRC), Briar, mitmproxy, Nextcloud, Wireguard

Note: NLnet is an independent organization, but it seems to get quite some support from EU. Maybe you would argue NLnet itself is a big actor?

I think funding already established, respected donor organizations is a decent strategy.

One counter example: https://nextgraph.org/elfa-consortium-encrypted-local-first-... (eg.. https://www.ironcalc.com/)

The source code is published [1] and there are MIT course videos available [2] (albeit 9 years old...).

The source code appear to be primarily in Matlab(!) though.

[1] https://github.com/Accla

[2] https://ocw.mit.edu/courses/res-ll-005-mathematics-of-big-da...

Yes. Isn't that "giant PITA" is referring to here?

your own repo reviewing and merging from upstream as needed. Would be a giant PITA though

Except most of the attacks so far has not landed actually source code changes to git IIRC. They have targeting the release files directly.

pnpm also support this

Their transcription (STT) models are good IMO

And the gotcha has been known about since 2014:

This is the class of attack documented by Adnan Khan in 2024. It's not a TanStack-specific bug; it's a known GitHub Actions design issue that requires conscious mitigation.

While it seems the maintainers kinda went-out-of-their way to enable this - GitHub could easily have at least turned of cache-sharing between fork jobs and the main jobs...

I've had good experiences with the Mistral Voxtral models (I've used the API, but some of the model-variants are open weight)

Idk, it almost seem a workaround for slow/broken go-back? If go-back is fast and state preserving, it's basically a fullscreen modal.

All(?) browser open links in a new tab when middle-clicked?

Can you please fix https://github.com/anthropics/claude-code/issues/43713 ?

Wish reword took a commit range though

I don't want "standby", I want suspend where the only power usage is keeping the ram alive...

Maybe they should fix bugs like this then https://github.com/anthropics/claude-code/issues/17979#issue... ...

I see. If that's just hashes and not encrypted content I can't see how they can resume old sessions properly. IIRC they have a 30 days retention policy and surely the thinking traces must be considered data. Wonder how this works with the zero-retention enterprise plans...

git bisect --first-parent