One of the most persistent biases in commentary on AI is many people's assumption that Silicon Valley VC-funded efforts are its main driver. But all the evidence points to something bigger than them - Open-Source.
Alibaba, the Chinese company that developed QwenLong-L1 isn't making their AI open-source out of the goodness of their heart. They get the benefit of lots of free workers improving it, and they also get to undermine their competitors. Many of the open-source AI are Chinese. It's possible China may be coordinating their efforts on a national scale, as a tit-for-tat retaliation measure responding to the US trying to weaken Chinese AI.
The global effect is the decentralized dispersion of AI technology with no one company or country dominating it - in the long run that will be its biggest outcome.
QwenLong-L1 solves long-context reasoning challenge that stumps current LLMs
“Run them locally”
It means the entire ai software is working on your own consumer computer without any connection to external servers you don't control.
“Open weights” means we have the tools to re-train and finetune the ai on whatever you want.
The default for llm like deepseek is of course they talk Chinese propaganda. But because its Open source we can just build a version that simply does not.