sharktheone

Oh what? That was not on my bingo card for 2026. Mold already was incredible when it still was in C++ and I assume it is even better now that it is in rust

Am I the only one who thought of the BEAM VM after the first word of thee title?

I think there are a few more. Especially cranelift, but I would not use it for production stuff and it is missing a lot of llvm intrinsics that will just trap if you are trying to use them. But for most it seems to be just fine for regular dev builds with a HUGE speedup.

Still waiting for remote functions to become ready :/

I am hoping for portable SIMD so much. But I still think that often a manually rolled SIMD will be faster.

Also the state of SIMD in Cranelift is also very WIP. They pretty much just support a subset of 128bit vectors with some rare exceptions.

The linked website uses one <style></style> for the blink animation, and also inline style (style="...") which also was only introduced in HTML4.

"no css" look inside css

First of all, this is VERY impressive. I wonder how this might effect the output quality. I wrote my own AI execution engines in the past and from that I can tell that it is really hard to get right and event small mistakes can make the model way dumber than expected.

While qwen3-4b is smarter than I remembered it, newer models seem to be way better, K2 Horizon 3.7B for example. Is there a reason why that qwen model was choosen?

My last question would be if the same optimizations could be made to a Jev-like model. I mean Jev is already fast and probably has a high throughput per H100. So could we maybe get to >2B tokens/minute/GPU with such a model?

I hope portable simd will be stabilized some time in rust :/

so nice. FDroid always looked a bit dated to me

Debuggers can lie pretty often. Especially if you are just doing things with ASM. I had some issues with V8 debugging and the debugger just saying that a function got called which for sure wasn't. Even with the option that V8 should emit gdb info :/

interesting. I mean day-to-day usage is a requirement if you want to buy some of these machines for real. But good to know that graphics works now. Last time I checked it wasn't and that was significantly after launch

Yeah, this just sucks imo. I mean I think for the copilot button there was something upstreamed, but the useful stuff wasn't

Could they please focus on their website rendering it better? Like I'm sorry, but no one actually cares about the Github Copilot App

poor Kimi K3 :/

Still waiting for the announcement "Claude kills 50% of human population after it cured cancer"

I always thought that just kernel mode makes it better.

Didn't know that WG can't use hardware crypto accelerators. I hope it will someday

Interesting to see that they are publishing a new tts model. I remember that they refused to release one of them a few years ago because they were too afraid of abusing it. Now they just release it without any much thought lol

Why does "gemini 3.8" make me not be too interested in their tts model?

I mean it technically can run linux. But I think non one actually got any desktop or graphics working. I hope it just happens but you are sadly correct here.

yeah. K2.6 can run on insane speeds. So sad that they don't have K3 yet.

But it can apparently also run 5.6 Sol

this feels like "we got the same benches as gpt-oss-120b but are also potentially slower while saying it is great"

still worried about how much data this thing will collect about you

I think I am still going to stick with the Steam Frame lmao

yeah everyone was hyped and it was said to be getting linux support. Some things were upstreamed into the kernel, but it never really got there

Sure, not too far, but single core perf is still. And also still very much behind the new M6 in single core and multi core https://browser.geekbench.com/v7/cpu/389219

I think this is huge. I thought about getting one of these chips a long time ago, but linux support was just a breaker. I ended up sticking to my current laptop and getting a macbook some years after

The "original" X Elite was also supposed to get good linux support but it unfortunately never happened.

So hopefully the X2 will actually happen!

hmm, it somehow continued the trend of being basically the same score on https://artificialanalysis.ai/ as the 5.6 variants.

I kind of hated Astra for it's poor instruction following and stopping all the time plus bad code quality. It somehow feels a bit like some of the popular open models but with a lot more knowledge or peek capability. But it doesn't reach peek that often

That is a lot. I thought Anthropic models would just do the opposite because they are greedy for money.

Interesting to see it now. I've used it a bunch before it came out and i pretty much didn't notice it. It might have been slightly better code quality, but still not great in that. I guess it just was slightly less frustrating to work with, but still AI...