Large Language Models (LLMs) went from non-existent to ubiquitous in the machine
learning discourse within a few years. Due to the fast pace of the field, it is difficult to identify …

AI alignment aims to make AI systems behave in line with human intentions and values. As
AI systems grow more capable, the potential large-scale risks associated with misaligned AI …

This work identifies 18 foundational challenges in assuring the alignment and safety of large
language models (LLMs). These challenges are organized into three different categories …

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 …

Large language models (LLMs) are currently at the forefront of intertwining AI systems with
human communication and everyday life. Due to rapid technological advances and their …

Sparse autoencoders (SAEs) are an unsupervised method for learning a sparse
decomposition of a neural network's latent representations into seemingly interpretable …

We show how to" compile" human-readable programs into standard decoder-only
transformer models. Our compiler, Tracr, generates models with known structure. This …

Recent advances in machine learning have enabled the development of next-generation
predictive models for complex computational biology problems, thereby spurring the use of …

\emph {Circuit analysis} is a promising technique for understanding the internal mechanisms
of language models. However, existing analyses are done in small models far from the state …

One of the roadblocks to a better understanding of neural networks' internals is\textit
{polysemanticity}, where neurons appear to activate in multiple, semantically distinct …