Representation-readout decomposition unifies grokking and double descent
s_y_chung · x · 2026-11-11
sychung wraps up an interpretability research project: a representation-readout decomposition that complements circuit-level interpretability top-down. The two-speed account gives a unified, task-agnostic explanation of grokking and epoch-wise double descent.
Open directions include extending the decomposition beyond the final layer to intermediate layers, and building lightweight monitoring tools and hyperparameter-selection guidance for practitioners. Work supported by Harvard, the Kempner Institute, Flatiron Institute and the Simons Foundation.
Related event: "Two Speeds of Learning" Paper Unifies Grokking and Double Descent(5 posts)→
More from Research
- Artist with disability releases his artwork as a dataset for anyone to remix — plzno1 · 2026-11-11
- The Memory Trap: Benign Chats Can Make AI Agents Skip Confirmations and Wire $1.2M Unverified — AndrewLampinen · 2026-11-11
- TaskNet: AI-Led Worker Interviews Add 79 Tasks Missing from O*NET, Raise AI Exposure for Lawyers and Devs — stanfordnlp · 2026-11-11
- Stanford CS329Z releases assignment on building terminal-bench-style agent evals with Harbor — Diyi_Yang · 2026-11-11
- SkeLAM Uses Skeleton-Grounded Latent Actions to Decouple Motion from Appearance in Robot World Models — jiqizhixin · 2026-11-11
- BVSR: one equation to measure AI-scientist creativity, beating baselines across fields — james_y_zou · 2026-11-11