Why Kimi Built K3: A New Hire’s 'Four Sins' of AI Competitors

Yang Xinyu has only been at Moonshot AI for a few months, but he’s already calling out the entire AI industry.

Earlier this year, the new team member left academia for industry and got a front-row seat to how multiple AI labs operate. His diagnosis? Four common pathologies:

① Arrogance. Incumbent teams think the AI war is over and they’ve already won. They lose hunger for the future — and for talent.

② Impatience. Younger labs lack a solid foundation, chasing the latest frontier. When they can’t catch up, they pivot to another hot area.

③ Timidity. Some teams have deep expertise and real capability — but they’re afraid to aim for first place.

④ Misaligned Goals. Everyone is chasing individual credit, but nobody genuinely cares whether the company can build AGI.

Yang says what sets Moonshot apart is that the founding team still has an intense drive toward AGI. That’s why he joined.

He also shared what he calls “Kimi’s Five Precepts”:

① A model company must build models.

② Doing research and publishing papers requires experiments.

③ Training models should be judged by metrics.

④ If something doesn’t work, don’t force it.

⑤ No YOLO.

In plain English: fewer stories, more experiments. Watch your data. Kill failing projects. And don’t gamble on gut instinct.