New J++ Lens interpretability method delivers 55% more faithful readouts than J-Lens
wesg52 · x · 2026-10-13
New interpretability research introduces the J++ Lens, an improvement over J-Lens that reads out latent concepts more reliably and at earlier layers.
- Core idea: filter noisy gradients before averaging them
- Result: 55% more faithful readouts than J-Lens
- Recall@10: Logit Lens 31.1%, J-Lens 35.7%, R-Lens 37.7%, J++ Lens 55.2%
A meaningful upgrade to activation readout methods for probing model internals, especially in early layers.
More from Research
- DUET paper co-evolves agent and LLM grader, hits 96.5% on Finance Agent benchmark — dair_ai · 2026-10-13
- Base Labs investigates why models forget facts they iteratively learn, with 13B talkie — baseten · 2026-10-13
- Robot deployment is becoming the operating layer for post-training physical AI policies — paigeinsf · 2026-10-13
- Spatial biology slide costs stuck at $5-10k for five years, defying Moore's law — anshulkundaje · 2026-10-13
- Google Research trains user simulators with latent intrinsic feedback, cutting RL compute by over half — cocoweixu · 2026-10-13
- Helmholtz-SDE closes the approximation gap in simulation-free latent SDEs — scott_linderman · 2026-10-13