Very happy to support this on behalf of Google. We have long benefited from open source, are big contributors to open source and in fact have consistently made open weights models with Gemma available from @GoogleDeepMind@demishassabis . Onwards!
Open-weight models are essential to a healthy AI ecosystem. Together with others across our industry, we are outlining a path for open-weight models to strengthen American competitiveness and expand economic opportunity, while protecting national security. https://t.co/Tr0sAzAxTD
When you're going through a tough time or feel like you're getting all the bad breaks, don't ask, "Why is this happening to me?"
Ask, "What can this do for me?"
What can I learn from this? What opportunity might this create? What door could this be opening that I just can't see yet?
And if it feels like every door is closing, then ask yourself one last question: What's the lesson?
There's always a lesson. If you learn from it, the setback wasn't a waste. It was a gift.
Stop being ungrateful when you get gifts in disguise. Say, thank you teacher.
a week ago @SemiAnalysis_ wrote that Chinese labs are "simply too compute poor to truly reach the frontier." today one of those "too compute poor" labs, a 300-person startup actually, shipped a model that compares to opus 4.8
the entire western consensus – export controls, the $650B hyperscaler capex race, the "compute moat" investment thesis – is built on one assumption: flops gate capability. if that were true, chip controls would keep chinese labs permanently behind the frontier.
but after reading through moonshot's stack i no longer think it is. training is efficiency-compressible: MoE routing, INT4-native quantization, better data curation, infra built around scarcity (their Mooncake stack exists because they don't have gpus!). a small lab with taste can compress the compute needed to make a frontier model, even if it can't afford to serve one
the frontier is no longer something money can buy
🚨 DeepSeek V4 is now running entirely on Huawei AI chips
China appears to be assembling a fully domestic frontier AI stack:
• DeepSeek V4 has been adapted to run entirely on Huawei Ascend clusters
• Huawei’s Atlas 950 SuperPoD debuts publicly tomorrow
• The system links thousands of Ascend processors into one computing cluster
• Designed for large scale AI training and inference
• Beijing will promote Chinese open-source AI as a low-cost alternative to Western closed models
This is one of China’s biggest steps yet towards removing Nvidia from its frontier AI ecosystem.
Will China finally compete with Nvidia?