kevinsxu's profile picture. Building/Investing @ Interconnected Capital; Writing @interconnect_ed; ex. GitHub, Obama White House/Commerce Dept; no one's quant

Kevin S. Xu

@kevinsxu

Building/Investing @ Interconnected Capital; Writing @interconnect_ed; ex. GitHub, Obama White House/Commerce Dept; no one's quant

置頂

2023: 170.13% 2024: 66.71% 2023-2024 average annual return: 112.21% 28 more years to go to reach my inner scorecard of 30+% return per year for 30 years More in my annual letter 👇

NEW POST: 2024 Annual Letter: Generational Company, Generational Wealth buff.ly/40Y0gW5 by @kevinsxu



Kevin S. Xu 已轉發

As far as I know, there isn't any chatbot or API that gives you access to an IMO 2025 gold-medalist model. Not only does this change today, but you get to download the weights with the Apache 2.0 open-source release of @deepseek_ai Math-V2 on @huggingface! Imagine owning the…

ClementDelangue's tweet image. As far as I know, there isn't any chatbot or API that gives you access to an IMO 2025 gold-medalist model. Not only does this change today, but you get to download the weights with the Apache 2.0 open-source release of @deepseek_ai Math-V2 on @huggingface!

Imagine owning the…

Kevin S. Xu 已轉發

Google TPU v6e vs AMD MI300X vs NVIDIA H100/B200: Artificial Analysis’ Hardware Benchmarking shows NVIDIA achieving a ~5x tokens-per-dollar advantage over TPU v6e (Trillium), and a ~2x advantage over MI300X, in our key inference cost metric In our metric for inference cost…

ArtificialAnlys's tweet image. Google TPU v6e vs AMD MI300X vs NVIDIA H100/B200: Artificial Analysis’ Hardware Benchmarking shows NVIDIA achieving a ~5x tokens-per-dollar advantage over TPU v6e (Trillium), and a ~2x advantage over MI300X, in our key inference cost metric

In our metric for inference cost…

The Middle East is surely becoming the epicenter of self-driving and robotaxi -- the future of urban transportation Mr Market seems to be sleeping on two things: 1. Uber's global robotaxi ambition; 2. WeRide, a tiny (~2-3B market cap) company, that is the tech backbone of this…

The future is arriving today! In partnership with @WeRide_ai, we’ve launched the Middle East’s first-ever fully driverless robotaxi commercial operations — starting today in Abu Dhabi.



Nice launch by another Nebius customer!

FLUX.2 is here - our most capable image generation & editing model to date. Multi-reference. 4MP. Production-ready. Open weights. Into the new.



I guess Taiwan now belongs in the "etc." bucket? 🤔

kevinsxu's tweet image. I guess Taiwan now belongs in the "etc." bucket? 🤔
kevinsxu's tweet image. I guess Taiwan now belongs in the "etc." bucket? 🤔

China speed! 😂

I asked (on ChinaTalk) the head of product at Z ai, one of the leading Chinese companies building open models, how long it takes them to get their model out the door once its done training. Incredible stuff: "a few hours" and the model is on HuggingFace.

natolambert's tweet image. I asked (on ChinaTalk) the head of product at Z ai, one of the leading Chinese companies building open models, how long it takes them to get their model out the door once its done training. Incredible stuff:

"a few hours" and the model is on HuggingFace.


Chinese tech companies have been "stealing" western open source (Linux, Android, MySQL/Postgres) to reduce dependency and improve profitability American AI companies are now "stealing" Chinese open source (DeepSeek, Qwen, Kimi) to reduce dependency and improve profitability…


This is a must read. Thoughtful and sobering

Today, I write about US-China competition in AI and associated technologies as I see it. “Race” is a fairly poor analogy, in my view, for understanding the magnitude of what is underway.

deanwball's tweet image. Today, I write about US-China competition in AI and associated technologies as I see it. 

“Race” is a fairly poor analogy, in my view, for understanding the magnitude of what is underway.


Kevin S. Xu 已轉發

WeRide was at Abu Dhabi Autonomous Week! 🇦🇪 Founder and CEO Tony Han showcased our AVs to His Highness Sheikh Hamdan bin Mohamed bin Zayed Al Nahyan & spoke on a panel on how Abu Dhabi’s supportive ecosystem, led by regulators like @ad_mobility, makes it a global AV hub. $WRD

WeRide_ai's tweet image. WeRide was at Abu Dhabi Autonomous Week! 🇦🇪 Founder and CEO Tony Han showcased our AVs to His Highness Sheikh Hamdan bin Mohamed bin Zayed Al Nahyan & spoke on a panel on how Abu Dhabi’s supportive ecosystem, led by regulators like @ad_mobility, makes it a global AV hub.
$WRD
WeRide_ai's tweet image. WeRide was at Abu Dhabi Autonomous Week! 🇦🇪 Founder and CEO Tony Han showcased our AVs to His Highness Sheikh Hamdan bin Mohamed bin Zayed Al Nahyan & spoke on a panel on how Abu Dhabi’s supportive ecosystem, led by regulators like @ad_mobility, makes it a global AV hub.
$WRD
WeRide_ai's tweet image. WeRide was at Abu Dhabi Autonomous Week! 🇦🇪 Founder and CEO Tony Han showcased our AVs to His Highness Sheikh Hamdan bin Mohamed bin Zayed Al Nahyan & spoke on a panel on how Abu Dhabi’s supportive ecosystem, led by regulators like @ad_mobility, makes it a global AV hub.
$WRD
WeRide_ai's tweet image. WeRide was at Abu Dhabi Autonomous Week! 🇦🇪 Founder and CEO Tony Han showcased our AVs to His Highness Sheikh Hamdan bin Mohamed bin Zayed Al Nahyan & spoke on a panel on how Abu Dhabi’s supportive ecosystem, led by regulators like @ad_mobility, makes it a global AV hub.
$WRD

Incredible launch! No wonder all the Nebius GPUs are sold out and fully utilized (World Labs is a $NBIS customer )

Introducing Marble by World Labs: a foundation for a spatially intelligent future. Create your world at marble.worldlabs.ai



Following Kimi's AMA on Reddit (reddit.com/r/LocalLLaMA/c…), Team Zhipu is doing one tomorrow! (reddit.com/r/LocalLLaMA/c…) Zhipu has done one 3 months ago too on its GLM models (reddit.com/r/LocalLLaMA/c…) Chinese AI labs' increasing community engagement and sophistication in strategy…


Kevin S. Xu 已轉發

I’m starting a new series of interviews on @interconnectsai with all the leading open model labs around the world to show why people are doing this, how people train great models, and where the ecosystem is going. The first one is Ant Group’s Ling (@AntLingAGI) / InclusionAI…

natolambert's tweet image. I’m starting a new series of interviews on @interconnectsai with all the leading open model labs around the world to show why people are doing this, how people train great models, and where the ecosystem is going. 

The first one is Ant Group’s Ling (@AntLingAGI) / InclusionAI…

A few high level thoughts after processing both CoreWeave and Nebius's earnings: Main storyline: CoreWeave vs Nebius is now becoming CoreWeave wants to become Nebius Here's why: - Nebius's strength is having developer-friendly software and multi-tenant expertise in operating a…

kevinsxu's tweet image. A few high level thoughts after processing both CoreWeave and Nebius's earnings:

Main storyline: CoreWeave vs Nebius is now becoming CoreWeave wants to become Nebius

Here's why:

- Nebius's strength is having developer-friendly software and multi-tenant expertise in operating a…
kevinsxu's tweet image. A few high level thoughts after processing both CoreWeave and Nebius's earnings:

Main storyline: CoreWeave vs Nebius is now becoming CoreWeave wants to become Nebius

Here's why:

- Nebius's strength is having developer-friendly software and multi-tenant expertise in operating a…

Nor did he pull any punches about what's happening today, albeit with his usual diplomatic subtlety "Finally, Berkshire will always be managed in a manner that will make its existence an asset to the United States and eschew activities that would lead it to become a supplicant.…

Warren Buffett releases his “farewell letter” ahead of Thanksgiving. What a life, and what a great man he turned out to be. It is an ode to America, to Omaha, his shareholders, friends, his loved ones, his family and an exhortation to all to be grateful in life.…



Nebius more than doubles its 2026 power guidance from 1 to 2.5GW, when power is the mother of all constraints for AI Incredible 🔥🚀

We now expect to have over 2.5 GW of contracted power by the end of 2026, including 800 MW to 1 GW of connected power. ⚡️

nebiusai's tweet image. We now expect to have over 2.5 GW of contracted power by the end of 2026, including 800 MW to 1 GW of connected power. ⚡️


Liang Wenfeng of DeepSeek said this in a 2024 interview, when DeepSeek V2 was released: "When the technology is disruptive, a moat built on being closed-source is only temporary. Even if OpenAI stays closed-source, it can't stop others from catching up. That's why we invest in…

Commodification by way of open source to erode the moat of closed source technology has been a well-trodden business strategy. Except this time, it's applied on a country-level, where economic pragmatism (what SV startups are doing with Chinese models) is pushing back on the…



Commodification by way of open source to erode the moat of closed source technology has been a well-trodden business strategy. Except this time, it's applied on a country-level, where economic pragmatism (what SV startups are doing with Chinese models) is pushing back on the…

The number of major US companies building on Chinese open source AI models (Qwen, Kimi, etc.) is really underappreciated



"Kimi chose INT4 over 'fancier' MXFP4/NVFP4 to better support non-Blackwell GPUs, with strong existing kernel support" The venn diagram of GPUs China cares to buy and ones the US is willing to sell is getting smaller and smaller by the day

🚀 "Quantization is not a compromise — it's the next paradigm." After K2-Thinking's release, many developers have been curious about its native INT4 quantization format. 刘少伟, infra engineer at @Kimi_Moonshot and Zhihu contributor, shares an insider's view on why this choice…

ZhihuFrontier's tweet image. 🚀 "Quantization is not a compromise — it's the next paradigm."
After K2-Thinking's release, many developers have been curious about its native INT4 quantization format.
刘少伟, infra engineer at @Kimi_Moonshot and Zhihu contributor, shares an insider's view on why this choice…
ZhihuFrontier's tweet image. 🚀 "Quantization is not a compromise — it's the next paradigm."
After K2-Thinking's release, many developers have been curious about its native INT4 quantization format.
刘少伟, infra engineer at @Kimi_Moonshot and Zhihu contributor, shares an insider's view on why this choice…
ZhihuFrontier's tweet image. 🚀 "Quantization is not a compromise — it's the next paradigm."
After K2-Thinking's release, many developers have been curious about its native INT4 quantization format.
刘少伟, infra engineer at @Kimi_Moonshot and Zhihu contributor, shares an insider's view on why this choice…
ZhihuFrontier's tweet image. 🚀 "Quantization is not a compromise — it's the next paradigm."
After K2-Thinking's release, many developers have been curious about its native INT4 quantization format.
刘少伟, infra engineer at @Kimi_Moonshot and Zhihu contributor, shares an insider's view on why this choice…


Loading...

Something went wrong.


Something went wrong.