Concept detail
continual learning
3 tagged videos · 6 connection videos · degree 0
Publication timeline
- 2026-071 video
- 2026-085 videos
- 2026-091 video
Source videos
Observed neighbors
Bayesian inference
Posterior updating is a theoretical route to avoiding catastrophic forgetting.
1 source videos
The mathematics of AI uncertaintyexpertise
required mechanism to convert on-the-job experience into specialized expertise
1 source videos
Intelligence + Continual Learning = Expertise — Yu Su, NeoCognitionmetalearning limits
Both bound how far a model can improve its own learning procedure without catastrophic regression on prior capability.
1 source videos
AI researchers debate how close we are to recursive self-improvementproduction traces
deployed interaction traces become the task-generation input for online RL, closing the deployment-to-training loop
1 source videos
Reinforcement Learning without Verifiable Rewards — Will Brown, Prime IntellectThe Bitter Lesson
Generalizes the compute-over-heuristics argument: continual adaptation, not curated prior knowledge, is the compute-scaling path.
1 source videos
Rich Sutton and Khurram Javed: Why AI Models Stop Learning, and How to Start It Againtrace mining
traces are the raw data substrate; without durable tracing there is no learning loop
1 source videos
Improving Agents is a Data Mining Problem — Vivek Trivedy, LangChain