Large Language Models (LLMs) possess remarkable reasoning abilities, yet their decision-making processes are often opaque, making it challenging to understand why they behave in unexpected ways. To address this, Gemma Scope 2 has been released as a comprehensive suite of interpretability tools for the Gemma 3 model family, ranging from 270 million to 27 billion parameters. This release is the largest open-source interpretability toolkit by an AI lab, designed to help researchers trace potential risks and better understand the internal workings of AI models. With the capability to store 110 petabytes of data and manage over a trillion parameters, Gemma Scope 2 aims to assist the AI research community in auditing and debugging AI agents, ultimately enhancing safety interventions against issues like jailbreaks and hallucinations. Interpretability research is essential for creating AI that is both safe and reliable as AI systems become more advanced and complex. Gemma Scope 2 acts like a microscope for the Gemma language models, using sparse autoencoders (SAEs) and transcoders to allow researchers to explore model internals and understand how their "thoughts" are formed and connected to behavior. This deeper insight into AI behavior is crucial for studying phenomena such as jailbreaks, where a model's internal reasoning does not align with its communicated reasoning. The new version builds on its predecessor by offering more refined tools and significant upgrades, including full coverage for the entire Gemma 3 family and advanced training techniques like the Matryoshka technique, which enhances the detection of useful concepts within models. Gemma Scope 2 also introduces tools specifically designed for analyzing chatbot behaviors, such as jailbreaks and chain-of-thought faithfulness. These tools are vital for deciphering complex, multi-step behaviors and ensuring models act as intended in conversational applications. By providing a full suite of interpretability tools, Gemma Scope 2 supports ambitious research into emergent behaviors that only appear at larger scales, such as those observed in models like the 27 billion parameter C2S Scale model. As AI technology continues to progress, tools like Gemma Scope 2 are crucial for ensuring that AI systems are not only powerful but also transparent and safe, ultimately benefiting the development of more robust AI safety measures. This matters because understanding and improving AI interpretability is crucial for developing safe and reliable AI systems, which are increasingly integrated into various aspects of society.
Large language models (LLMs) rely on the transformer architecture, a sophisticated neural network that processes sequences of token embeddings to generate text. The process begins with tokenization, where raw text is divided into discrete tokens, which are then mapped to identifiers. These identifiers are used to create embedding vectors that carry semantic and lexical information. Positional encoding is added to these vectors to provide information about the position of each token within the sequence, preparing the input for the deeper layers of the transformer. Inside the transformer, each token embedding undergoes multiple transformations. The first major component is multi-headed attention, which enriches each token's representation by capturing various linguistic relationships within the text. This component is crucial for understanding the role of each token in the sequence. Following this, feed-forward neural network layers further refine the token features, applying transformations independently to each token. This process is repeated across multiple layers, progressively enhancing the token embeddings with more abstract and long-range linguistic information. At the final stage, the enriched token representation is processed through a linear output layer and a softmax function to produce next-token probabilities. The linear layer generates unnormalized scores, or logits, which the softmax function converts into normalized probabilities for each possible token in the vocabulary. The model then selects the next token to generate, typically the one with the highest probability. Understanding this journey from input tokens to output probabilities is crucial for comprehending how LLMs generate coherent and context-aware text. This matters because it provides insight into the inner workings of AI models that are increasingly integral to various applications in technology and communication.