AI alignment aims to make AI systems behave in line with human intentions and values. As
AI systems grow more capable, so do risks from misalignment. To provide a comprehensive …

Understanding AI systems' inner workings is critical for ensuring value alignment and safety.
This review explores mechanistic interpretability: reverse engineering the computational …

Inspired by the Kolmogorov-Arnold representation theorem, we propose Kolmogorov-Arnold
Networks (KANs) as promising alternatives to Multi-Layer Perceptrons (MLPs). While MLPs …

Through considerable effort and intuition, several recent works have reverse-engineered
nontrivial behaviors oftransformer models. This paper systematizes the mechanistic …

The capabilities of large language models (LLMs) have sparked debate over whether such
systems just learn an enormous collection of superficial statistics or a set of more coherent …

Sparse autoencoders provide a promising unsupervised approach for extracting
interpretable features from a language model by reconstructing activations from a sparse …

We analyze transformers from the perspective of iterative inference, seeking to understand
how model predictions are refined layer by layer. To do so, we train an affine probe for each …

The last decade of machine learning has seen drastic increases in scale and capabilities.
Deep neural networks (DNNs) are increasingly being deployed in the real world. However …

Large language models based on transformers have achieved great empirical successes.
However, as they are deployed more widely, there is a growing need to better understand …

Do neural networks, trained on well-understood algorithmic tasks, reliably rediscover known
algorithms? Several recent studies, on tasks ranging from group operations to in-context …