Gemma Scope

anchor

Findings

• IC-054Gemma 2 2B performance degrades substantially when routed through Gemma Scope SAEs, and SAE-based feature suppression causes broad cross-domain degradation rather than targeted knowledge removal [builds-on]
• IC-054Gemma 2 2B performance degrades substantially when routed through Gemma Scope SAEs, and SAE-based feature suppression causes broad cross-domain degradation rather than targeted knowledge removal [primary]
• IC-205Gemma 2's SAE features exhibit depth-dependent organization, with polysemantic features in early layers and persistent, matchable features in later layers [builds-on]