Language and vision-language models have shown impressive performance across a wide range of tasks, but their internal mechanisms remain only partly understood. In this work, we study how individual attention heads in text-generative models specialize in specific semantic or visual attributes. Building on an established interpretability method, we reinterpret the practice of probing intermediate activations with the final decoding layer through the lens of signal processing. This lets us analyze multiple samples in a principled way and rank attention heads based on their relevance to target concepts. Our results show consistent patterns of specialization at the head level across both unimodal and multimodal transformers. Remarkably, we find that editing as few as 1% of the heads, selected using our method, can reliably suppress or enhance targeted concepts in the model output.

Do neural networks build their representations through smooth, gradual refinement, or via more complex computational processes? We investigate this by extending the logit lens to analyze the policy network of Leela Chess Zero, a superhuman chess engine. Although playing strength and puzzle-solving ability improve consistently across layers, capability progression occurs in distinct computational phases with move preferences undergoing continuous reevaluation--move rankings remain poorly correlated with final outputs until late, and correct puzzle solutions found in middle layers are sometimes overridden. This late-layer reversal is accompanied by concept preference analyses showing final layers prioritize safety over aggression, suggesting a mechanism by which heuristic priors can override tactical solutions.

Named entities are fundamental building blocks of knowledge in text, grounding factual information and structuring relationships within language. Despite their importance, it remains unclear how Large Language Models (LLMs) internally represent entities. Prior research has primarily examined explicit relationships, but little is known about entity representations themselves. We introduce entity mention reconstruction as a novel framework for studying how LLMs encode and manipulate entities. We investigate whether entity mentions can be generated from internal representations, how multi-token entities are encoded beyond last-token embeddings, and whether these representations capture relational knowledge. Our proposed method, leveraging _task vectors_, allows to consistently generate multi-token mentions from various entity representations derived from the LLMs hidden states. We thus introduce the _Entity Lens_, extending the _logit-lens_ to predict multi-token mentions. Our results bring new evidence that LLMs develop entity-specific mechanisms to represent and manipulate any multi-token entities, including those unseen during training. Our code is avalable at https://github.com/VictorMorand/EntityRepresentations .

Pre-trained language models can be surprisingly adept at tasks they were not explicitly trained on, but how they implement these capabilities is poorly understood.

Although large language models (LLMs) have tremendous utility, trustworthiness is still a chief concern: models often generate incorrect information with high confidence. While contextual information can help guide generation, identifying when a query would benefit from retrieved context and assessing the effectiveness of that context remains challenging. In this work, we operationalize interpretability methods to ascertain whether we can predict the correctness of model outputs from the model's activations alone. We also explore whether model internals contain signals about the efficacy of external context. We consider correct, incorrect, and irrelevant context and introduce metrics to distinguish amongst them. Experiments on six different models reveal that a simple classifier trained on intermediate layer activations of the first output token can predict output correctness with about 75% accuracy, enabling early auditing. Our model-internals-based metric significantly outperforms prompting baselines at distinguishing between correct and incorrect context, guarding against inaccuracies introduced by polluted context. These findings offer a lens to better understand the underlying decision-making processes of LLMs. Our code is publicly available at https://github.com/jiarui-liu/LLM-Microscope

While large language models (LLMs) appear to be increasingly capable of solving compositional tasks, it is an open question whether they do so using compositional mechanisms. In this work, we investigate how feedforward LLMs solve two-hop factual recall tasks, which can be expressed compositionally as $g(f(x))$. We first confirm that modern LLMs continue to suffer from the "compositionality gap": i.e. their ability to compute both $z = f(x)$ and $y = g(z)$ does not entail their ability to compute the composition $y = g(f(x))$. Then, using logit lens on their residual stream activations, we identify two processing mechanisms, one which solves tasks $\textit{compositionally}$, computing $f(x)$ along the way to computing $g(f(x))$, and one which solves them $\textit{directly}$, without any detectable signature of the intermediate variable $f(x)$. Finally, we find that which mechanism is employed appears to be related to the embedding space geometry, with the idiomatic mechanism being dominant in cases where there exists a linear mapping from $x$ to $g(f(x))$ in the embedding spaces. We fully release our data and code at: https://github.com/apoorvkh/composing-functions .

Chain-of-thought (CoT) reasoning has enabled transformer-based language models to excel at complex mathematics and multi-step planning. However, in standard decoder-only architectures, these reasoning steps are externalized in natural language, improving interpretability at the cost of efficiency. To capture reasoning that is not easily represented in words, many works have explored recurrent architectures that aim to internalize reasoning in latent space, potentially supporting latent CoT. In this paper, we investigate whether such reasoning structures emerge in Huginn-3.5B, a depth-recurrent Transformer that reuses layers at inference time without increasing parameter count. We examine the model's internal behavior on arithmetic tasks using a suite of probing techniques including the Logit Lens and Coda Lens. Our findings reveal limited evidence of interpretable latent CoT by tracking rank trajectories of final and intermediate result tokens. Furthermore, we uncover significant probing inconsistencies across recurrent blocks, where the interpretability of hidden states depends heavily on both the layer index and the decoding method. Finally, we empirically show that increasing recurrence depth yields only marginal gains and falls well short of models that explicitly externalize reasoning steps. The code is available at https://github.com/wenquanlu/huginn-latent-cot.

Multi-digit addition is a clear probe of the computational power of large language models. To dissect the internal arithmetic processes in LLaMA-3-8B-Instruct, we combine linear probing with logit-lens inspection. Inspired by the step-by-step manner in which humans perform addition, we propose and analyze a coherent four-stage trajectory in the forward pass:Formula-structure representations become linearly decodable first, while the answer token is still far down the candidate list.Core computational features then emerge prominently.At deeper activation layers, numerical abstractions of the result become clearer, enabling near-perfect detection and decoding of the individual digits in the sum.Near the output, the model organizes and generates the final content, with the correct token reliably occupying the top rank.This trajectory suggests a hierarchical process that favors internal computation over rote memorization. We release our code and data to facilitate reproducibility.

This work explores entropy analysis as a tool for probing information distribution within Transformer-based architectures. By quantifying token-level uncertainty and examining entropy patterns across different stages of processing, we aim to investigate how information is managed and transformed within these models. As a case study, we apply the methodology to a GPT-based large language model, illustrating its potential to reveal insights into model behavior and internal representations. This approach may offer insights into model behavior and contribute to the development of interpretability and evaluation frameworks for transformer-based models