Laundry Lamps, Billion Salaries, China, & More

TLDR;

Welcome back! This week, we take a look at yet another fantastic release from Chinese models, yet another record-setting compensation package for an AI researcher coming from Meta, Anthropic’s business troubles, a laundry-folding lamp, and more.

Enjoy!

FRONTIER MODEL

China Releases SOTA Reasoning Model

The time has come. China now has a competitive benchmark, with US frontier models such as o3, Gemini 2.5 Pro, or Grok 4.

Zhipu AI has released the model in two versions:

GLM-4.5, the flagship model with just 355 billion parameters with 32 billion active (only 9% of parameters activate per prediction, making it fast despite the large size)

GLM-4.5 Air, a 100 billion parameter model with 12 active, meaning it’s as fast as small models despite being considerably large.

As you can see in the above results, the models are in the same league as the best of the best in the West, and with totally open-weight and permissible licenses, automatically making them the default choice for any open-source user alongside other great models like DeepSeek R1 or Kimi K2.

And what do all these have in common? Of course, you know what.

TheWhiteBox’s takeaway:

China is dominating the Pareto frontier; it’s not even a debate at this point. Something is very wrong with the US/EU strategy when the top 5 best options to run local models are all coming from China.

And as we’ll discuss in the next news, open-source is much more important than it seems.

AI TALENT

Meta Announces New Chief AI Scientist

Meta has announced Shengjia Zhao as the new Chief AI Scientist for its Superintelligence Unit, the highest AI rank in the entire company, ahead of Yann LeCun (who will remain as Chief AI Scientist at Fundamental AI Research (FAIR), the non-Generative AI unit).

Shengjia Zhao was a key member of OpenAI, considered one of the prime creators of ChatGPT, which means that this is, without a doubt, one of the hardest hits in Meta’s relentless OpenAI poaching spree.

Almost simultaneously, it has also hired three top researchers at Google DeepMind who played a key role in Gemini’s recent historic International Mathematics Olympiad gold medal.

TheWhiteBox’s takeaway:

Although unconfirmed, it’s believed that Shengjia Zhao was offered a total compensation, over the following years, that exceeds $1 billion, which could be considered the highest salary the world has ever seen.

Seeing Meta building a stacked AI team is excellent, but only if the company stays loyal to its open-source efforts, something that has been put into question lately.

As we have seen with Zhipu’s GLM models, China is not only releasing competitive models, but these are open-source too, which means all the best open-source models (or dare I say, open-weights, as Chinese Labs also don’t share datasets like US Labs) are Chinese, and the gap with the US isn’t only not falling, it’s increasing:

Source

I can’t overstate how concerning this trend is to the interests of the US. If academia and enterprises, both heavily inclined toward open-source, start using…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.