> This site is an independent informational resource and is not officially affiliated with Zhipu AI.

What about using LLMs to automatically update docs?

I think separating thinking tokens from "representing" tokens might be a better approach, like what those thinking models does

Japanese kanji is not the same as Chinese characters.

Yes, and the vast majority of Chinese would now write it as 汉字 instead of 漢字

Do C++ modules solve this problem?

Kind of, the current problem is they aren't fully mature, you great a good experience with VC++ and clang (vLatest), alongside MSBuild/CMake/ninja.

VS IDE tooling is stil kind of broken because they rely on EDG instead of VC++, and fixing module intelisense has been low priority for EDG, and MS as well as I would assume a $4 trillion valued company would care about their suppliers.

Clion has better support on that regard.

GCC is still not fully there, and Apple clang, well Apple is happy with module header maps for their Swift/Objective-C integration, so lets see.

C++ modules solve exactly one of the problems - the "one pch limit" one - at the cost of introducing several more. Certainly they are not more compiler-independent!

> ollama gpt-oss 120b mxfp4 1 94.67 11.66

This is insanely slow given its 200+GB/s memory bandwidth. As a comparison, I've tested GPT OSS 120B on Strix Halo and it obtains 420tps prefill and >40tps decode.

Probably the quants have higher perplexity, but the Sparks performance seems to be lack lustre. The reviewer videos I've seen so far tries their best not to offend Nvidia or, rather, not break their contracts.

"Train a 100,000,000,000B multimodal large language model" got 75/100

I think AI doesnt give a chance LLMs. :)

> https://preview.redd.it/wca7kzfq5w2f1.png?width=1190&format=...

"4x gated residual streams" look quite weird. Is there any paper or technique report for this?

Why is this post flagged?

Right? The feature is just being abused at this point, it's gross. I wish dang would look into it

And you can't even vouch for flagged articles like you can for dead comments

There are far, far more israelis and jews on HN than muslims.

kind of ridiculous. they should learn how to keep my grub entries first.