Liam Carroll

Jesse Hoogland, George Wang, Matthew Farrugia-Roberts et al.

Deep learning involves navigating a high-dimensional loss landscape over the neural network parameter space. Over the course of training, complex computational structures form and re-form inside the neural network, leading to shifts in input/output behavior. It is a priority for the science of deep learning to uncover principles governing the development of neural network structure and behavior. Drawing on the framework of singular learning theory, we propose that model development is deeply linked to degeneracy in the local geometry of the loss landscape. We investigate this link by monitoring loss landscape degeneracy throughout training, as quantified by the local learning coefficient, for a transformer language model and an in-context linear regression transformer. We show that training can be divided into distinct periods of change in loss landscape degeneracy, and that these changes in degeneracy coincide with significant changes in the internal computational structure and the input/output behavior of the transformers. This finding provides suggestive evidence that degeneracy and development are linked in transformers, underscoring the potential of a degeneracy-based perspective for understanding modern deep learning.

Simon Pepin Lehalleur, Jesse Hoogland, Matthew Farrugia-Roberts et al.

In this position paper, we argue that understanding the relation between structure in the data distribution and structure in trained models is central to AI alignment. First, we discuss how two neural networks can have equivalent performance on the training set but compute their outputs in essentially different ways and thus generalise differently. For this reason, standard testing and evaluation are insufficient for obtaining assurances of safety for widely deployed generally intelligent systems. We argue that to progress beyond evaluation to a robust mathematical science of AI alignment, we need to develop statistical foundations for an understanding of the relation between structure in the data distribution, internal structure in models, and how these structures underlie generalisation.

Liam Carroll, Jesse Hoogland, Matthew Farrugia-Roberts et al.

Modern deep neural networks display striking examples of rich internal computational structure. Uncovering principles governing the development of such structure is a priority for the science of deep learning. In this paper, we explore the transient ridge phenomenon: when transformers are trained on in-context linear regression tasks with intermediate task diversity, they initially behave like ridge regression before specializing to the tasks in their training distribution. This transition from a general solution to a specialized solution is revealed by joint trajectory principal component analysis. Further, we draw on the theory of Bayesian internal model selection to suggest a general explanation for the phenomena of transient structure in transformers, based on an evolving tradeoff between loss and complexity. We empirically validate this explanation by measuring the model complexity of our transformers as defined by the local learning coefficient.

Alistair Reid, Simon O'Callaghan, Liam Carroll et al.

Organisations are starting to adopt LLM-based AI agents, with their deployments naturally evolving from single agents towards interconnected, multi-agent networks. Yet a collection of safe agents does not guarantee a safe collection of agents, as interactions between agents over time create emergent behaviours and induce novel failure modes. This means multi-agent systems require a fundamentally different risk analysis approach than that used for a single agent. This report addresses the early stages of risk identification and analysis for multi-agent AI systems operating within governed environments where organisations control their agent configurations and deployment. In this setting, we examine six critical failure modes: cascading reliability failures, inter-agent communication failures, monoculture collapse, conformity bias, deficient theory of mind, and mixed motive dynamics. For each, we provide a toolkit for practitioners to extend or integrate into their existing frameworks to assess these failure modes within their organisational contexts. Given fundamental limitations in current LLM behavioural understanding, our approach centres on analysis validity, and advocates for progressively increasing validity through staged testing across stages of abstraction and deployment that gradually increases exposure to potential negative impacts, while collecting convergent evidence through simulation, observational analysis, benchmarking, and red teaming. This methodology establishes the groundwork for robust organisational risk management as these LLM-based multi-agent systems are deployed and operated.