If this told me anything it's that most of the world falls within 5 point difference, guess we're doing good

Speaking of ASICs - how likely is it that as models get better we'll see someone baking a whole model directly into the silicon? It's like having l0 cache.

Only 8B currently but it's been done: https://taalas.com/products/

The only _public_ example we know.

This is definitely being done with private models by HFT/quant firms, data processing agencies/orgs (large intelligence agencies, _every_ data analytics org, etc).

You could do it but there would be no point, The only advantage over would be power consumption. And it would be quite expensive.

At the rate models are improving, it would be obsolete in six months.

Yes, right now it would be obsolete in six months, but I also must add that this never stopped crypto miners from making new ASICs. However, with how useful Kimi is right now - at some point if someone makes a dedicated hardware board with "good enough" model for daily tasks - that would be a very sought after commodity.

Power consumption and latency are very important on mobile

They're important everywhere of course, but especially on mobile. If AI researchers figure out how to offload knowledge and expertise from reasoning weights, then a core reasoning ASIC linked to the knowledge would totally rock.

Good thing they're not selling a printer then, they're selling a parts kit that you have to flash an open source controller firmware to and assemble yourself. Or you may just source them in any way you want. You can't really "outlaw" open source, we had this whole song and dance with https://en.wikipedia.org/wiki/Export_of_cryptography_from_th...

This would be the line for me. If at some point I'm unable to build an .apk and install it on my phone without Google letting me, I'm moving to Huawei.

If you're building the APK, you're probably installing via ADB, in which case none of the changes apply

Does Huawei not use android or Google play services?

It's Android but without Google's services, there's an alternative app store.

The irony of Chinese vendors providing a breath of fresh low-DRM air.

Partially true, HarmonyOS NEXT is its own thing, with a Typescript based language ArkTS.

https://developer.huawei.com/consumer/en/arkts/

And now they are adding yet another one, AOT compiled, Cangjie

https://cangjie-lang.cn/en

Using Android fork has been a transition step.

Neat, thanks for this correction! Interesting, an entire new programming language.

And a microkernel based OS with capabilities.

Another example that microkernels actually do have market share.

Low DRM? I looked at Huawei devices because I figured they'd have to sell them here super cheap because of this downside most Europeans people will even see as a showstopper ("how will I install my precious WhatsApp??"), but

- they're among the most expensive (I could afford that if needed though)

- they don't allow hardware unlock (ehh.. what's the point, then, if I get a locked-down device with Chinese surprises!)

OK yeah I didn't know they stopped allowing to root. Normal levels of DRM then, my mistake, you're right.

It seems like China is becoming the "freedom superpower" while USA is getting "corporate superpower" vibes. Huh

I'm curious why you think China is actually more open in this regard. The CCP has direct influence over the apps that are allowed to be installed on these phones. There is nothing more free about them.

No, Google is barred from providing any services to them by the US government.

not like that no, some US carriers don't allow them though like AT&T blocks you to google or apple phones. for them only pixel supports a way out with graphene.

No, they use AppGallery and HMS.

I swear, the more time goes on, the more I disrespect the "no to systemd crowd", not because they don't have valid arguments, but because in recent years it attracted a very *specific* kind of audience, and now it smells bad.

I've used Linux for over 20 years at this point, my first Linux computer was a PS2 console and my second one was a PS3 console. I remember the rc scripts and upstart, and I hated dealing with any of it, systemd, its timers and the user units were the most convinient change in decades. I hear so many anecdotal accounts of "sytemd destroying someone's system", yet personally it was nothing but a pleasant experience.

I think there's always needs to be an alternative, and I welcome anyone who's actually building alternative code paths and doing the work to create the alternative tools, but every time some tool gains a dependency on systemd to improve the functionality and the mob comes in complaining I can't help but get angry. It's open source. You can patch it. You can switch to something else, but instead you hurl insults at people volunteering their time to make software you run better.

The anti-systemd crowd just seems uneducated to be honest. They still say that systemd is bloated, because they don’t seem to understand systemd is a catalogue of software, not literally everything happening in PID 1.

Moreover none of them really seem to have operated servers at scale to understand why systemd is useful. Your init system is 1000 lines of C? Cool story bro that does literally none of the things sysadmins need to run a server fleet, enjoy gluing everything together with brittle bash scripts.

Pfft. Ever heard of runit or s6(6)?

( https://smarden.org/runit/ https://www.skarnet.org/software/s6/ )

While systemd may be convenient, and more than a simple init & supervision,

it's still slow compared to runit/s6(6).

Let's say I have a larger process-tree.

Let's say I give it's top, or any thread sighup, lazily out of htop.

Systemd? Takes seconds to tear it down, and restart.

Runit? In an instant. s6(6)? Same.

Just a flicker in the corner of my eye.

There is nothing brittle about https://github.com/skarnet/execline which is a dependency of s6.

Tested initially on almost obsolete systems.

First noticed with large firefoxen, with many windows and tabs.

But it isn't limited to that, just any larger process-tree suffices for comparison.

Does still apply to current versions and is reproducible on contemporary systems.

Runit and before it DJB's daemon-tools (and derivatives) have been around for a long time.

S6 came later. All rather good stuff, if one knows them, that they exist at all,

and is able to handle them, which isn't really complicated,

because they never had the history of glued together brittle bash-scripts.

Just saying...….

Exactly. They scraped the internet we all of us built with our own research, open source work, sharing, etc. I'm never going to agree that they own their models.

Can't say I agree. When I used to rent I lived in the same apartment for 3 years, and not a single thing changed about it.

And that's what you should do, since you're forcing the protocol update you should take the opportunity that might not come later

In hindsight though, and really also in foresight, it didn't work out that way.

Going to pour one out for all the Iranian refugees celebrating

They need it hoard datasets.