AI · 1h ago
Why Continuous Learning Can't Live in Model Weights
DeepSeek's Liang Wenfeng says continuous learning is the missing piece for AGI. Experts agree that frozen models can't learn per user due to cost, opacity, and lock-in. The solution lies in portable, inspectable file layers like Tolaria, which keep knowledge model-agnostic.
Meridian48 take
The article makes a compelling case that memory must be external to models, but the hard part—retrieval and runtime—remains unsolved.
continuous-learningmodel-memory