Full threadrfv6723·> using imagenet-1k for pretrainingLecun still can't show JEPA competitive at scale with autoregressive LLM.View on HN