Clément Dumas

Clément Dumas, Chris Wendler, Veniamin Veselovsky et al.

A central question in multilingual language modeling is whether large language models (LLMs) develop a universal concept representation, disentangled from specific languages. In this paper, we address this question by analyzing latent representations (latents) during a word-translation task in transformer-based LLMs. We strategically extract latents from a source translation prompt and insert them into the forward pass on a target translation prompt. By doing so, we find that the output language is encoded in the latent at an earlier layer than the concept to be translated. Building on this insight, we conduct two key experiments. First, we demonstrate that we can change the concept without changing the language and vice versa through activation patching alone. Second, we show that patching with the mean representation of a concept across different languages does not affect the models' ability to translate it, but instead improves it. Finally, we generalize to multi-token generation and demonstrate that the model can generate natural language description of those mean representations. Our results provide evidence for the existence of language-agnostic concept representations within the investigated models.

Julian Minder, Clément Dumas, Caden Juang et al.

Model diffing is the study of how fine-tuning changes a model's representations and internal algorithms. Many behaviors of interest are introduced during fine-tuning, and model diffing offers a promising lens to interpret such behaviors. Crosscoders are a recent model diffing method that learns a shared dictionary of interpretable concepts represented as latent directions in both the base and fine-tuned models, allowing us to track how concepts shift or emerge during fine-tuning. Notably, prior work has observed concepts with no direction in the base model, and it was hypothesized that these model-specific latents were concepts introduced during fine-tuning. However, we identify two issues which stem from the crosscoders L1 training loss that can misattribute concepts as unique to the fine-tuned model, when they really exist in both models. We develop Latent Scaling to flag these issues by more accurately measuring each latent's presence across models. In experiments comparing Gemma 2 2B base and chat models, we observe that the standard crosscoder suffers heavily from these issues. Building on these insights, we train a crosscoder with BatchTopK loss and show that it substantially mitigates these issues, finding more genuinely chat-specific and highly interpretable concepts. We recommend practitioners adopt similar techniques. Using the BatchTopK crosscoder, we successfully identify a set of chat-specific latents that are both interpretable and causally effective, representing concepts such as $\textit{false information}$ and $\textit{personal question}$, along with multiple refusal-related latents that show nuanced preferences for different refusal triggers. Overall, our work advances best practices for the crosscoder-based methodology for model diffing and demonstrates that it can provide concrete insights into how chat-tuning modifies model behavior.

Adam Karvonen, James Chua, Clément Dumas et al.

Large language model (LLM) activations are notoriously difficult to understand, with most existing techniques using complex, specialized methods for interpreting them. Recent work has proposed a simpler approach known as LatentQA: training LLMs to directly accept LLM activations as inputs and answer arbitrary questions about them in natural language. However, prior work has focused on narrow task settings for both training and evaluation. In this paper, we instead take a generalist perspective. We evaluate LatentQA-trained models, which we call Activation Oracles (AOs), in far out-of-distribution settings and examine how performance scales with training data diversity. We find that AOs can recover information fine-tuned into a model (e.g., biographical knowledge or malign propensities) that does not appear in the input text, despite never being trained with activations from a fine-tuned model. Our main evaluations are four downstream tasks where we can compare to prior white- and black-box techniques. We find that even narrowly-trained LatentQA models can generalize well, and that adding additional training datasets (such as classification tasks and a self-supervised context prediction task) yields consistent further improvements. Our best AOs match or exceed white-box baselines on all four tasks and the best overall baseline on 3 of 4. These results suggest that diversified training to answer natural-language queries imparts a general capability to verbalize information about LLM activations.

Clément Dumas

Mechanistic interpretability research requires reliable tools for analyzing transformer internals across diverse architectures. Current approaches face a fundamental tradeoff: custom implementations like TransformerLens ensure consistent interfaces but require coding a manual adaptation for each architecture, introducing numerical mismatch with the original models, while direct HuggingFace access through NNsight preserves exact behavior but lacks standardization across models. To bridge this gap, we develop nnterp, a lightweight wrapper around NNsight that provides a unified interface for transformer analysis while preserving original HuggingFace implementations. Through automatic module renaming and comprehensive validation testing, nnterp enables researchers to write intervention code once and deploy it across 50+ model variants spanning 16 architecture families. The library includes built-in implementations of common interpretability methods (logit lens, patchscope, activation steering) and provides direct access to attention probabilities for models that support it. By packaging validation tests with the library, researchers can verify compatibility with custom models locally. nnterp bridges the gap between correctness and usability in mechanistic interpretability tooling.

Stewart Slocum, Julian Minder, Clément Dumas et al.

Knowledge editing techniques promise to implant new factual knowledge into large language models (LLMs). But do LLMs really believe these facts? We develop a framework to measure belief depth and use it to evaluate the success of knowledge editing techniques. We operationalize belief depth as the extent to which implanted knowledge 1) generalizes to related contexts (e.g. Fermi estimates several logical steps removed), 2) is robust to self-scrutiny and direct challenge, and 3) is represented similarly to genuine knowledge (as measured by linear probes). Our evaluations show that simple prompting and mechanistic editing techniques fail to implant knowledge deeply. In contrast, Synthetic Document Finetuning (SDF) - where models are trained on LLM-generated documents consistent with a fact - often succeeds at implanting beliefs that behave similarly to genuine knowledge. However, SDF's success is not universal, as implanted beliefs that contradict basic world knowledge are brittle and representationally distinct from genuine knowledge. Overall, our work introduces measurable criteria for belief depth and enables the rigorous evaluation necessary for deploying knowledge editing in real-world applications.

Julian Minder, Clément Dumas, Stewart Slocum et al.

Finetuning on narrow domains has become an essential tool to adapt Large Language Models (LLMs) to specific tasks and to create models with known unusual properties that are useful for research. We show that narrow finetuning creates strong biases in LLM activations that can be interpreted to understand the finetuning domain. These biases can be discovered using simple tools from model diffing - the study of differences between models before and after finetuning. In particular, analyzing activation differences on the first few tokens of random text and steering by adding this difference to the model activations produces text similar to the format and general content of the finetuning data. We demonstrate that these analyses contain crucial information by creating an LLM-based interpretability agent to understand the finetuning domain. With access to the bias, the agent performs significantly better compared to baseline agents using simple prompting. Our analysis spans synthetic document finetuning for false facts, emergent misalignment, subliminal learning, and taboo word guessing game models across different architectures (Gemma, LLaMA, Qwen) and scales (1B to 32B parameters). We suspect these biases reflect overfitting and find that mixing pretraining data into the finetuning corpus largely removes them, though residual risks may remain. Our work (1) demonstrates that narrowly finetuned models have salient traces of their training objective in their activations and suggests ways to improve how they are trained, (2) warns AI safety and interpretability researchers that the common practice of using such models as a proxy for studying broader finetuning (e.g., chat-tuning) might not be realistic, and (3) highlights the need for deeper investigation into the effects of narrow finetuning and development of truly realistic case studies for model-diffing, safety and interpretability research.