If we sort out continuous learning such that large models don’t just get a batch of snippet stew out of a blender, and read sources linearly, it may help with credibility assessment.

→ View original post on X — @id_aa_carmack,