At this point I simply don't care about Anthropic or OAI in the slightest, the exciting stuff is coming from China with great, cheap SOTA models and local AI that people can run themselves.

I use bubblewrap directly, never liked Flatpak.

Same except for Steam. Hope this money revitalizes bubblewrap's development, it'd be nice to get proper signal handling cleanup up then merged (even util-linux's unshare has it, these days).

I have bad news for you: The flatpak developers plan to stop using bubblewrap. They say it is no longer necessary because most distros now support unprivileged user namespaces. https://youtu.be/NsVhkz2Xl0E?si=-ypxvTlRukCZLql9

Bubblewrap is a CLI interface for these, you know? It's been some time since bwrap recommended avoiding their SUID wrapper.

From my knowledge its a yes and no. One of the "benefits" of bwrap was that it was a SUID binary and thus maximised compatibility with systems that disabled unprivileged user namespaces. With that barrier gone you can implement a more tailored sandbox to the usecase instead of working around limitations in bwrap. Like, all things considered, bubblewrap as a whole isn't that much code (quick check seems like ~6k LOC) and does somewhat limit the API you can work with.

I tried to find out what unprivileged user namespaces was, and all I can find is people talking loudly about how insecure they are, e.g. https://secureblue.dev/articles/userns

But does bubblewrap also use them?

> 90% of features for systems half a decade old

This can be a good tradeoff. An M1 Ultra still runs circles around 90% of consumer hardware sold today, and with Linux it has easily another 10+ years. Now support for M3 is coming.

> Asahi is doing incredible work but it's still 5 people in a garage vs the huge corporation.

Not sure why the "vs" scenario, this is a project that makes no promises. And I think they do have some support from Fedora / Red Hat fwiw.

> This can be a good tradeoff.

That's a good tradeoff for durable hardware with available spare parts. That's not a good tradeoff for hardware that was designed so it cannot be repaired, such as newer Macs.

> Not sure why the "vs" scenario, this is a project that makes no promises. And I think they do have some support from Fedora / Red Hat fwiw.

They are building something in an environment that's not supportive of what they're doing. Contrast with Intel or AMD.

Yeah for a lot of us macOS really gets in the way of using the otherwise excellent hardware. My daily driver is an M1 Pro with Asahi and it's easily my favorite computer ever. It's a real shame that the newer M models can't run Linux.

I get ~10 tok/sec on a M1 Ultra, but the real issue is abysmal prompt processing which makes it unusable. On the M1 and M2 series MTP is actively harmful for performance.

What signals? Looking more like a tiered release.

Any good LLM should be able to come up with the equivalent git and curl commands fairly easily.

I'm doing much the same, avoid the long thinking loops and instead have more iterations on the plan with reviews from different angles.

When I set up my local AI server I tried different harnesses and one prompt was to build a site for a consultancy in my domain. Not long after an actual agency I know had a website redesign and it was the exact same thing I got from my local model and the Hermes agent with the web design skill.

6 or 7 years here, no news or socials. I keep telling people the same thing, you find out about the important stuff real quick regardless. The rest is noise and rage bait.

It's sad how effective rage bait is on many people, and how oblivious they are to it.