Most recurring multilingual SAE features fail causal validation — only one truly drives translation in Gemma
ravithejads · x · 2026-09-08
A BlackboxNLP 2026 Special Track paper causally validates multilingual sparse autoencoder (SAE) translation features in Gemma 2 and Gemma 3:
- Reproduces Wu et al.'s SAE feature discovery and extends it across varying prompt/source/target languages
- In both models, 20+ features fire frequently across all discovery settings, but causal validation (amplifying/ablating activations) shows nearly all have small or inconsistent effects — feature recurrence can overstate cross-lingual transfer
- The exception: Gemma 2's (L10, 5717) and Gemma 3's (L20, 2456) consistently improve COMET scores when amplified and degrade when ablated across 23 language settings — a language-agnostic translation-initiation direction
A methodological warning for interpretability researchers: activation frequency alone is unreliable; causal validation is a must.
More from Research
- Stanford Data-Centric AI Workshop reopens site, extends poster deadline — jure · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Ito: a browser that detects browsing intent and auto-organizes tabs into Flows — andrewkuznet · 2026-10-26
- Cornell study says brains process language as linear sequences, challenging Chomsky's hierarchy — anirbanbandyo · 2026-10-26
- Hi3D V1.0 open-sourced with 2M training assets and 90% slicing success rate — kimmonismus · 2026-10-26
- Ortet evaluates 34 open-weight models on 50 health tasks, finds general leaderboards insufficient — elmanmansimov · 2026-10-26