Ablation: even/odd tokens, not 'cat', control the zigzag — and it fits only one probe
Sauers_ · x · 2026-10-20
Ablations localize the mechanism: knocking out 'cat' from the prompt doesn't break the zigzag, but 'even'/'odd' tokens are required — the model learned an execution circuit keyed to the parity instruction, with cat as the target direction.
The behavior still doesn't generalize to other cat probes or logit-lens P(cat), remaining tightly fit to the specific probe used in training.
Related event: Researcher Trains Qwen3's Internal Representations Into a Zigzag Pattern(9 posts)→
More from Research
- New paper: let models reward hack in RL, then disable the parameters at deployment — nptacek · 2026-10-20
- Victor Taelin: smart enough compilers will understand and swap your algorithms — bronzeagepapi · 2026-10-20
- Galbot proposes observational scaling law for physical recursive self-improvement in VLM agents — Galbot · 2026-10-20
- ActionRoPE encodes action as world coordinates for spatially consistent game world models — Zhiyang Deng · 2026-10-20
- UniAVC unifies audio-video generation via progressive training and asymmetric noise scheduling — Kien T. Pham · 2026-10-20
- AHE paper auto-evolves coding-agent harnesses with decoupled files, lifting Terminal-Bench 2 to 77% — morgymcg · 2026-10-20