Kevin Meng

Kevin Meng, Arnab Sen Sharma, Alex Andonian et al. · mit

Recent work has shown exciting promise in updating large language models with new memories, so as to replace obsolete information or add specialized knowledge. However, this line of work is predominantly limited to updating single associations. We develop MEMIT, a method for directly updating a language model with many memories, demonstrating experimentally that it can scale up to thousands of associations for GPT-J (6B) and GPT-NeoX (20B), exceeding prior work by orders of magnitude. Our code and data are at https://memit.baulab.info.

Evan Hernandez, Arnab Sen Sharma, Tal Haklay et al. · microsoft-research, mit

Much of the knowledge encoded in transformer language models (LMs) may be expressed in terms of relations: relations between words and their synonyms, entities and their attributes, etc. We show that, for a subset of relations, this computation is well-approximated by a single linear transformation on the subject representation. Linear relation representations may be obtained by constructing a first-order approximation to the LM from a single prompt, and they exist for a variety of factual, commonsense, and linguistic relations. However, we also identify many cases in which LM predictions capture relational knowledge accurately, but this knowledge is not linearly encoded in their representations. Our results thus reveal a simple, interpretable, but heterogeneously deployed knowledge representation strategy in transformer LMs.

Jagdeep Singh Bhatia, Kevin Meng · mit

Perceptual hashes map images with identical semantic content to the same $n$-bit hash value, while mapping semantically-different images to different hashes. These algorithms carry important applications in cybersecurity such as copyright infringement detection, content fingerprinting, and surveillance. Apple's NeuralHash is one such system that aims to detect the presence of illegal content on users' devices without compromising consumer privacy. We make the surprising discovery that NeuralHash is approximately linear, which inspires the development of novel black-box attacks that can (i) evade detection of "illegal" images, (ii) generate near-collisions, and (iii) leak information about hashed images, all without access to model parameters. These vulnerabilities pose serious threats to NeuralHash's security goals; to address them, we propose a simple fix using classical cryptographic standards.

Kevin Meng, David Bau, Alex Andonian et al.

We analyze the storage and recall of factual associations in autoregressive transformer language models, finding evidence that these associations correspond to localized, directly-editable computations. We first develop a causal intervention for identifying neuron activations that are decisive in a model's factual predictions. This reveals a distinct set of steps in middle-layer feed-forward modules that mediate factual predictions while processing subject tokens. To test our hypothesis that these computations correspond to factual association recall, we modify feed-forward weights to update specific factual associations using Rank-One Model Editing (ROME). We find that ROME is effective on a standard zero-shot relation extraction (zsRE) model-editing task, comparable to existing methods. To perform a more sensitive evaluation, we also evaluate ROME on a new dataset of counterfactual assertions, on which it simultaneously maintains both specificity and generalization, whereas other methods sacrifice one or another. Our results confirm an important role for mid-layer feed-forward modules in storing factual associations and suggest that direct manipulation of computational mechanisms may be a feasible approach for model editing. The code, dataset, visualizations, and an interactive demo notebook are available at https://rome.baulab.info/