Google DeepMind has released Gemma Scope 2, an open set of interpretability tools that now works across the full Gemma 3 family of language models. The announcement positions the release as a resource for the AI safety community, offering a way to probe how these models process and represent information internally.
According to the source, the new tools are designed to deepen understanding of complex model behavior. This suggests a focus on making model internals more transparent, which could help researchers identify failure modes or biases. The release follows earlier interpretability efforts from DeepMind, though the source does not detail specific technical changes or performance metrics.
Because there is only one source, there are no conflicting viewpoints to compare. The key claim is straightforward: Gemma Scope 2 is now available and covers the entire Gemma 3 line, expanding access to interpretability methods that were previously more limited in scope.