Loom Lens Experiment: Parallel Multi-Line MTP Sampling Shows Potential for Interp Agents
voooooogel · x · 2026-11-12
The full chain lens post co-authored with Raymond Douglas is out, with a big results table and WorkspaceBench battle stories. Since the MTP head can be sampled like a regular LM, the authors also tried a 'loom lens' inspired by the logit loom project — sampling multiple parallel continuation lines. Not yet significantly better (too complicated for the judge), but they think it has potential in the hands of an interp agent.
Related event: vgel team explores Loom Lens: sampling MTP heads like a language model(2 posts)→
More from Research
- Axiom Math formalization finds broken step in finite simple group classification proof — jiqizhixin · 2026-11-12
- SimEX: agents learn manipulation skills in imperfect simulation and transfer them to real robots — chris_j_paxton · 2026-11-12
- Lie bracket visualized: how second-order residue lets wheeled robots move without sliding — burny_tech · 2026-11-12
- On PostTrainBench, AI post-training now beats the human-tuned SmolLM3-3B from the same base — maksym_andr · 2026-11-12
- FigBrains: models are outpacing our ability to build training environments by hand — vsikka · 2026-11-12
- Open CHI Reviews site archives real CHI peer reviews, crowdsourced from authors — IanArawjo · 2026-11-12