In October 2025, Andrej Karpathy said on Dwarkesh Patel’s podcast that current large models lack continuous learning, stating, “You can’t tell them something and expect them to remember,” and estimated it would take about another decade to address these cognitive limitations. Two months later, in his annual review, he noted that the major leap in model capabilities in 2025 came primarily from verifiable reward reinforcement learning (RLVR), but memory, multimodal perception, continuous learning, and the ability to operate computers remain clear weaknesses. He added, “We have prototypes, but no agents ready to work as colleagues.” Continual learning, also known as lifelong learning, refers to a model’s ability to continuously learn from new tasks, knowledge, and experiences after deployment without forgetting what it has already learned. A core obstacle is catastrophic forgetting, where fine-tuning a model with new data overwrites parameters responsible for previous capabilities. Over the past year, academia and industry have diverged into several technical pathways, including attaching external memory, continuously updating weights, full re-pretraining, and redefining what “learning” means.