Nathan Lambert publishes open-source AI reading list

The reading list arrives as a structured overview rather than a single news event, pulling together pieces on why labs release open models, how businesses use them, and where adoption differs between open and closed systems. It includes Mark Zuckerberg's July 2024 explanation for Meta releasing Llama 3, alongside arguments that open models will lag closed models in performance but still serve as complements for enterprise agentic workflows.

The list also traces how Western companies have shifted to Chinese open models. Perplexity adopted DeepSeek R1, and Thomson Reuters built on Qwen to move off Claude, while DoorDash, Airbnb, Anysphere / Cursor, and Apple have drawn lawmaker scrutiny over their use of Chinese models. Those examples sit alongside the list's claim that the open-closed performance gap is now roughly 4-6 months.

The most contested ground remains distillation. The list includes both an argument that distillation helps Chinese labs without negating their innovation and a rebuttal to what it calls a political panic that distillation is the only reason those labs are near the frontier. How regulators weigh evidence of distillation techniques against the list's account of open-model safety and cyber risk may shape whether Western enterprises feel comfortable adopting these models.