Interpreting residual streams to build better tag clusters and a dendrogram search engine
amplifiedamp · x · 2026-10-20
- The author applied a residual-stream interpretation method (developed by friends) to a base model, producing more useful tag clusters than plain k-means over the last layer's residual stream.
- He built a dendrogram viewer and a search engine on top of the clusters.
- The demo dataset (EAG NYC attendees) is boring and unshareable, so the search engine can't be released; he has more interesting internal datasets but won't share those results yet.
Related event: Residual Stream Interpretation Method Beats Last-Layer k-means Clustering(2 posts)→
More from Research
- Anima Anandkumar on neural operators: AI weather models beat supercomputers — AnimaAnandkumar · 2026-10-23
- Scaling GLM-5.3-Flash training to 1M token context, 64x over Megatron baseline — AndrewLBeam · 2026-10-23
- Experimental boolean-logic LM: microsecond forward pass on one CPU thread, but painfully slow training — cephaloform · 2026-10-23
- AdaDex turns seconds of human hand motion into generalizable dexterous robot skills — mangahomanga · 2026-10-23
- ReAPO paper: conditioning retries on failures beats GRPO in RLVR reasoning training — jasondeanlee · 2026-10-23
- Quantum-JEPA: first JEPA world model deployable on real quantum circuits (QPU) — DrMorganLevine · 2026-10-23