Open-Source AI & Open Models Reading List

Newsletter · Watch/Read · (gathered 2026-09-13)

• Nathan Lambert’s reading list offers a comprehensive overview of open models, covering their foundational concepts, strategic implications, safety debates, and the intense US-China competition in the AI landscape.
• Key concepts include the “Gradient of Generative AI Release” (Irene Solaiman), which frames open models on a spectrum rather than a binary, and the claim that “open models in perpetual catch-up” (Nathan Lambert) will consistently lag closed models in performance.
• “Distillation”—training models on output tokens from stronger models—is highlighted as the “single most eventful debate around open models in 2026,” significantly contributing to Chinese labs’ ability to close the performance gap (now roughly 4-6 months) with frontier closed models.
• A warning is issued regarding the risk of “vibe regulation” potentially leading to a ban on frontier open models in the US, even as Western companies increasingly adopt advanced Chinese open models like DeepSeek R1 and Qwen for cost savings and performance.