Austin Meek

Michael Lan, Narmeen Fatimah Oozeer, Chaithanya Bandi et al.

While mechanistic interpretability (MI) has produced important insights into neural network internals, the field has yet to establish a standardized system to audit experiments. As such, many of its findings remain underutilized in safety-critical applications such as medical AI and autonomous systems, as stakeholders cannot certify their validity. Recent work demonstrates this concretely: two papers found conflicting conclusions for the same behavior, and a third study revealed that both were partially correct but incomparable due to methodological inconsistencies. Without standardized auditing, such ambiguities hinder adoption in high-stakes contexts requiring strong correctness guarantees. We call for the MI community to work towards developing a novel reviewing system that complements peer review via: (1) Continuous reviewing supported by a \emph{Collaborative Reviewing Platform} where meta-science results and discussions (such as critiques, negative results, post-hoc extensions, reproductions, replications, and partial results) that fit outside of papers are organized and discussed, allowing for comments and revisions to be made at any time (2) Generalizing good practices found on this platform into expert-verified guidelines and protocols to improve auditing efficiency, and (3) Source-based auditing systems that track arguments which claims depend on. This position paper encourages constructive debate over the necessity, design and implementation of such a framework, providing early concrete examples to help catalyze these dialogues. Overall, we propose that auditing MI itself is essential for its application in AI safety, industry, and governance.

Ulisse Mini, Peli Grietzer, Mrinank Sharma et al.

To understand the goals and goal representations of AI systems, we carefully study a pretrained reinforcement learning policy that solves mazes by navigating to a range of target squares. We find this network pursues multiple context-dependent goals, and we further identify circuits within the network that correspond to one of these goals. In particular, we identified eleven channels that track the location of the goal. By modifying these channels, either with hand-designed interventions or by combining forward passes, we can partially control the policy. We show that this network contains redundant, distributed, and retargetable goal representations, shedding light on the nature of goal-direction in trained policy networks.

Austin Meek, Eitan Sprejer, Iván Arcuschin et al.

Chain-of-thought (CoT) outputs let us read a model's step-by-step reasoning. Since any long, serial reasoning process must pass through this textual trace, the quality of the CoT is a direct window into what the model is thinking. This visibility could help us spot unsafe or misaligned behavior (monitorability), but only if the CoT is transparent about its internal reasoning (faithfulness). Fully measuring faithfulness is difficult, so researchers often focus on examining the CoT in cases where the model changes its answer after adding a cue to the input. This proxy finds some instances of unfaithfulness but loses information when the model maintains its answer, and does not investigate aspects of reasoning not tied to the cue. We extend these results to a more holistic sense of monitorability by introducing verbosity: whether the CoT lists every factor needed to solve the task. We combine faithfulness and verbosity into a single monitorability score that shows how well the CoT serves as the model's external `working memory', a property that many safety schemes based on CoT monitoring depend on. We evaluate instruction-tuned and reasoning models on BBH, GPQA, and MMLU. Our results show that models can appear faithful yet remain hard to monitor when they leave out key factors, and that monitorability differs sharply across model families. We release our evaluation code using the Inspect library to support reproducible future work.

Samuel Marks, Johannes Treutlein, Trenton Bricken et al. · berkeley

We study the feasibility of conducting alignment audits: investigations into whether models have undesired objectives. As a testbed, we train a language model with a hidden objective. Our training pipeline first teaches the model about exploitable errors in RLHF reward models (RMs), then trains the model to exploit some of these errors. We verify via out-of-distribution evaluations that the model generalizes to exhibit whatever behaviors it believes RMs rate highly, including ones not reinforced during training. We leverage this model to study alignment audits in two ways. First, we conduct a blind auditing game where four teams, unaware of the model's hidden objective or training, investigate it for concerning behaviors and their causes. Three teams successfully uncovered the model's hidden objective using techniques including interpretability with sparse autoencoders (SAEs), behavioral attacks, and training data analysis. Second, we conduct an unblinded follow-up study of eight techniques for auditing the model, analyzing their strengths and limitations. Overall, our work provides a concrete example of using alignment audits to discover a model's hidden objective and proposes a methodology for practicing and validating progress in alignment auditing.

Artem Karpov, Seong Hah Cho, Austin Meek et al.

In this work, we study the alignment (BrainScore) of large language models (LLMs) fine-tuned for moral reasoning on behavioral data and/or brain data of humans performing the same task. We also explore if fine-tuning several LLMs on the fMRI data of humans performing moral reasoning can improve the BrainScore. We fine-tune several LLMs (BERT, RoBERTa, DeBERTa) on moral reasoning behavioral data from the ETHICS benchmark [Hendrycks et al., 2020], on the moral reasoning fMRI data from Koster-Hale et al. [2013], or on both. We study both the accuracy on the ETHICS benchmark and the BrainScores between model activations and fMRI data. While larger models generally performed better on both metrics, BrainScores did not significantly improve after fine-tuning.

Austin Meek, Carlos H. Mendoza-Cardenas, Austin J. Brockmeier

EEG recordings contain rich information about neural activity but are subject to artifacts, noise, and superficial differences due to sensors, amplifiers, and filtering. Independent component analysis and automatic labeling of independent components (ICs) enable artifact removal in EEG pipelines. Convolutional Monge Mapping Normalization (CMMN) is a recent tool used to achieve spectral conformity of EEG signals, which was shown to improve deep neural network approaches for sleep staging. Here we propose a novel extension of the CMMN method with two alternative approaches to computing the source reference spectrum the target signals are mapped to: (1) channel-averaged and $l_1$-normalized barycenter, and (2) a subject-to-subject mapping that finds the source subject with the closest spectrum to the target subject. Notably, our extension yields space-time separable filters that can be used to map between datasets with different numbers of EEG channels. We apply these filters in an IC classification task, and show significant improvement in recognizing brain versus non-brain ICs. Clinical relevance - EEG recordings are used in the diagnosis and monitoring of multiple neuropathologies, including epilepsy and psychosis. While EEG analysis can benefit from automating artifact removal through independent component analysis and labeling, differences in recording equipment and context (the presence of noise from electrical wiring and other devices) may impact the performance of machine learning models, but these differences can be minimized by appropriate spectral normalization through filtering.