H.264 is supported on Linux in DR Studio, it’s just not in the free version.

And to bring up a likely reason: my impression is that H.264 and AAC still have active patents that apply to the major implementations, and those are not licensed gratis for all uses, and that applies more heavily to the encoding side. Raspberry Pi used to (and might still, I don't know) offer add-on license keys to unlock certain video codec support for what I assume was a similar reason. A lot of individual Linux desktop users just do their small-scale FFmpeg runs under the radar, but commercial software delivery is different.

With proprietary operating systems, the commercial entity behind the OS can negotiate for licensing. Some commercially-associated Linux distributions have efforts in similar directions, like Fedora for instance has some special secondary-repo stuff around H.264 support, but that's at minimum much more fragmentary as a target for an application developer concerned with the legalities.

Yep, but AAC is not, so you're still stuck needing to encode multiple times. This adds delay and also massive disk churn.

I rarely care about the date of the commit on the branch to determine recency, I care about when I last switched to the branch.

This comment gets it correct https://news.ycombinator.com/item?id=49504860

I wrote my own version to do this nearly 10 years ago: https://github.com/amarshall/git-recent-branches

Qubes uses VMs, Jails use a shared kernel. So I wouldn’t say Jails are safer nor that they have a smaller exposed area with regard to host isolation than VMs.

Bigger issue was that native encryption doesn’t (didn’t?) have a maintainer, so even though most bugs got fixed, it’s not a great thing to build on. Some of the key storage is also a bit outdated.

IMO use LUKS unless you really need the features of ZFS native enc.

Has it? Most of the LLM-generated commit messages and PR descriptions I read are needlessly verbose and miss the point: elucidating the why of a change rather than the what. If I actually pose a question based on them the author often tells me “yea that was some AI generated nonsense, I rewrote it now”.

Use uBlock Origin and add custom filters to block scripts on http:// ; e.g.

    http://$inline-script,domain=~localhost
    http://$script,domain=~localhost

It’s possible to make a fake display to avoid that; both Linux and Windows have this ability. No dummy plug is required (though it is simpler than software solutions).

I don't use the dummy plug for display, it's only that consumer nvidia gpus won't have working 3D acceleration if there is nothing physically connected to the card. My dummy plug doesn't support the resolutions I run games at anyway.

EDIT: Ok, correction, I bothered to go unplug the dummy plug and it still works. So I was wrong, Sunshine does not need a dummy plug at all, not even on Nvidia. Now it remains a mystery why I have one in the first place if it's useless after all…

Sunshine was significantly easier for me to get working using a dummy plug. I'm currently exploring other options but that's definitely the happy path.

Pretty light on details, heavy on fluff. 9.8s to 3.1s was userfaultd + hugepages, 500ms was PS/2 mouse and… where is the rest of the time to get to 400ms?

Thinking doesn’t change output speed. Anthropic’s models are ~ 40–60 t/s median output speed.

Do you have access to Anthropic model weights to run them locally?

No, and having that is not required to know output speed nor the effect of thinking, so I don’t see the point in such a superfluous, indirect question.

As for the question you’re likely asking: benchmarks that include speed across many models and providers available at various places e.g. https://artificialanalysis.ai/leaderboards/models

Can take the geometric mean of total and active parameters of MoE to get approximate equivalent quality to dense model params. So sqrt(35*10)≈18.7.

The trade-off of MoE is that it is worse but faster for the same total size.