Our letter counting jobs are safe.

I stand corrected.

I wasn’t aware there was a ton of bluegrass in eastern Scandinavia.

I may also be bad at this. It was a curt addition!

I was just adding another direction of evidence to suzha’s post. They shared a lot of good examples of how waymos drive well in a way a machine would, and I added one example of how they fail in ways that a telepresence wouldn’t. Another would be when they make those traffic jams amongst themselves.

And why are they thwarted by traffic cones on the hood?

There are security architectures that can be made safe, and there are ones that cannot be. Unfortunately, the most popular one (access control lists) has ways it cannot be secured when used in normal ways. papers.agoric.com/papers/acls-...

Me, joking: I can’t believe we chose “destroying the economy” and “losing wars” over “being nice to trans people occasionally”

Conservatives, dead serious: it was a tough choice, but I’m glad we picked “destroying the economy” and “losing wars” over “being nice to trans people occasionally”

Well said. We need subjective feeds not because truth is subjective, but because <we don’t want to be beholden to bad mods>

Do we want to moderate? Hell yeah. But we’re gonna call it subjective so we don’t go creating a moderation cartel that can be abused. Subjective? Maybe “Bypassable”.

For purposes of suing anthropic, or as in fear of prosecution for forking/reading?

Evidence on Anthropic's Use of Copyrighted Training Data: Litigation, Settlements, and Legal Framework Through April 2026 Evidence on Anthropic's Use of Copyrighted Training Data: Litigation, Settlements, and Legal Framework Through April 2026

That’s a noble position reflecting a much more principled respect for IP than Anthropic showed the rest of the world. claude.ai/public/artif...

eliminate install scripts, use www.npmjs.com/package/@lav... to run the ones you actually need, set minimum package age, don't store unprotected ssh keys in .ssh

I'm also working on more tools:
github.com/LavaMoat/kip...

Look at all of lavamoat.github.io for runtime security on top of that.

Not as bad if you’re running LavaMoat. JS flexibility allows defense against supply chain attacks for those who care to. lavamoat.github.io

Seriously, though: who keeps going in for another round of this?

Betting on prediction markets of events this administration can influence because I like donating my money to the ultra wealthy.

I'm with MarkM's twinging at "fearless" cooperation. More like "more precisely constrained" collaboration, or something, although that just doesn't sound as trumpet-worthy, even though it's meant to express something as trumpet-worthy as computer security has to offer...

it sometimes works on entire paywalls!

ios reader mode and also the whole element hiding mode is pretty great. support.apple.com/en-us/120682

Glad he’s ok, that’s scary stuff. Sweet haunting tune.

Yeah, I don't know why I pitched it as certificate capabilities. I think that was an example of *me* trying to simplify the concept for the recipient. "Just a chain of signatures" feels like easier bait than "well, ok, a distributed vm runtime", yet the ocaps directly address the mental overhead.

Thanks for replying, it was very cool to learn you were a UCAN author, that gave me hope, lol.

I can squint and see this, and I am sympathetic to wanting to keep the protocol as simple as possible since you're already battling so many education fronts at once.

I just think it's too bad. I think it would be a huge win for scalable communities, and it'll just be won in another venue then.

So let's say I want to invite users who can invite other users to also post, but be able to revoke the tree freely.

Because I can build a capability system on a server that controls the owner and itself can update the ACL, we don't need more than an ACL in the protocol?

Handling revocations got complex? I'd love to hear how it got too complex, because from where I'm sitting, the ability to revoke invitations (and the ability for sub-administration) is a valuable feature that adds *useful complexity* that can save a space when it's needed.

You want to crawl the data to verify authenticity? A capability chain is *way more verifiable* than a 4-hour expiring auth token you have to request from the owner's server.

Boooo!

bsky.app/profile/danf...

Man, I'm so offended on behalf of all those human devs who we're protecting from learning.

"Human devs can't comprehend it, so we shouldn't bother adopting this superior paradigm" is the best argument for "LLMs should be writing code instead of humans" I've heard yet. They honestly get it just fine.

You know what else is unfamiliar to people? Social media that works like it oughtta for people.

Oh, it looks like they did consider a certificate-capability system (UCAN!), but wrote it off because people are unfamiliar with the model (womp-womp 🎺)

bsky.app/profile/dhol...

Anyways, that'd be one change I'd make.

I think you could make a very lightweight version of this for AT, you wouldn't even need a full OcapN type thing like Endo/Spritely are doing (although people who found this interesting might check those out).
endojs.org
spritely.institute