AI Radar

Henry's source note frames this paper around In depth details of how Meta built Llama3 and the various configurations and hyperparameters.

OaK is Sutton's proposed model-based reinforcement-learning architecture for agents that learn continually from their own experience. Its components learn online, its parameter step sizes are meta-learned, and it creates state and time abstractions through the FC-STOMP progression: Feature Construction, SubTask, Option, Model, and Planning. This is an invited keynote and research program rather than a conventional peer-reviewed paper. Oak Lab's companion technical post demonstrates one piece of the motivation: adaptive credit assignment can isolate predictable signal in noisy online streams where the shown SGD baseline absorbs noise. The full OaK architecture remains a proposed system, not a validated end-to-end result.

Henry's source note frames this paper around conviction to scale up GPT2/3/4.

Henry's source note frames this paper around InstructGPT->ChatGPT.

Henry's source note frames this paper around foundation, self-attention.

Henry's source note frames this paper around Latest methods for solving ARC-AGI problems.

Seeded from Henry Shi's AI Crash Course reading list.

Henry's source note frames this paper around Building OSS o1-level reasoning model with pure RL, no SFT, no RM.

Henry's source note frames this paper around RL without prior knowledge of game or rules.

The classic RAG paper. It frames retrieval as a way to give generation a non-parametric memory instead of forcing every fact into model weights.

Reflexion gives agents a verbal self-feedback memory so failures can shape the next attempt without weight updates.