Continuous Learning Won't Come From the Weights

Short summary

DeepSeek's Liang Wenfeng identifies continuous learning as the key missing piece on the path to AGI, and this article argues it cannot live inside model weights due to economics, opacity, and vendor lock-in. Instead, durable agent memory should be stored in portable, inspectable formats like markdown and git, with a cognitive runtime handling retrieval, promotion, decay, and identity separation. The article warns that naive summarized memory can make agents overconfident about wrong facts, proving that how memory is structured matters as much as what it stores.

• •Continuous learning can't live in model weights — economics, opacity, and lock-in make it impractical
• •Portable file-based memory (markdown + git) with a cognitive runtime for retrieval, decay, and promotion is the emerging pattern
• •Synthetic summarized memory can be worse than no memory — provenance and uncertainty must be preserved to prevent false confidence

Generated with AI, which can make mistakes.