Related papers: Circuit Breaking: Removing Model Behaviors with Ta…

Large language models (LLMs) can detect software vulnerabilities, but how do they actually identify vulnerable code? We address this question using mechanistic interpretability; analyzing the internal computations of a neural network to…

Where should we intervene in a language model (LM) to localize and control behaviors that are diffused across many tokens of a long-form response? We introduce Generative Causal Mediation (GCM), a procedure for selecting model components…

As large language models achieve impressive scores on traditional benchmarks, an increasing number of researchers are becoming concerned about benchmark data leakage during pre-training, commonly known as the data contamination problem. To…

Employing language models to generate explanations for an incoming implicit hate post is an active area of research. The explanation is intended to make explicit the underlying stereotype and aid content moderators. The training often…

Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in architecture, this method still requires…

Recent work has shown that computation in language models may be human-understandable, with successful efforts to localize and intervene on both single-unit features and input-output circuits. Here, we introduce an approach which extends…

Refusal on harmful prompts is a key safety behaviour in instruction-tuned large language models (LLMs), yet the internal causes of this behaviour remain poorly understood. We study two public instruction-tuned models, Gemma-2-2B-IT and…

Current alignment evaluation mostly measures whether models encode dangerous concepts and whether they refuse harmful requests. Both miss the layer where alignment often operates: routing from concept detection to behavioral policy. We…

Context: Large Language Models (LLMs) are increasingly used in modern software development, aiding in code generation, code completion, and refactoring through AI-powered assistants. While they accelerate development workflows, they often…

The unit selection problem aims to find objects, called units, that optimize a causal objective function which describes the objects' behavior in a causal context (e.g., selecting customers who are about to churn but would most likely…

A causal-decoder block is hierarchical: lower layers build the residual basis that upper layers attend over. We identify a failure mode in GPT pretraining: upper layers commit to sharp attention patterns before lower-layer features…

Large language models (LLMs) have shown various ability on natural language processing, including problems about causality. It is not intuitive for LLMs to command causality, since pretrained models usually work on statistical associations,…

Topic modeling has emerged as a dominant method for exploring large document collections. Recent approaches to topic modeling use large contextualized language models and variational autoencoders. In this paper, we propose a negative…

Pre-trained language models (LMs) are able to perform complex reasoning without explicit fine-tuning. To understand how pre-training with a next-token prediction objective contributes to the emergence of such reasoning capability, we…

Language models are now prevalent in software engineering with many developers using them to automate tasks and accelerate their development. While language models have been tremendous at accomplishing complex software engineering tasks,…

Sequence generation applications require satisfying semantic constraints, such as ensuring that programs are correct, using certain keywords, or avoiding undesirable content. Language models, whether fine-tuned or prompted with few-shot…

Large pre-trained language models are often trained on large volumes of internet data, some of which may contain toxic or abusive language. Consequently, language models encode toxic information, which makes the real-world usage of these…

Lack of moderation in online communities enables participants to incur in personal aggression, harassment or cyberbullying, issues that have been accentuated by extremist radicalisation in the contemporary post-truth politics scenario. This…

Conversational large language models are trained to refuse to answer harmful questions. However, emergent jailbreaking techniques can still elicit unsafe outputs, presenting an ongoing challenge for model alignment. To better understand how…

Large pre-trained models have transformed machine learning, yet adapting these models effectively to exhibit precise, concept-specific behaviors remains a significant challenge. Task vectors, defined as the difference between fine-tuned and…