Scaling Interpretable Transformers with Parity Bottleneck Layers

Language models are thought to exhibit the phenomenon of superposition, representing many more features than dimensions in their residual streams. Sparse autoencoders (SAEs) are designed to recover such features post-hoc, but training models that are interpretable by construction has remained impractical, as a per-layer over-complete bottleneck is prohibitively expensive in both memory and compute. To overcome this issue, we introduce the ParityTransformer, a GPT-2-scale architecture whose intermediate representations are efficient and wide / sparse by design. At each layer, a Deep Parity Bottleneck (DPB) replaces a learned over-complete basis with a parameter-free algebraic dictionary, providing a deterministic incoherence guarantee and eliminating the memory requirements that have prevented per-layer interpretable bottlenecks at scale. A DPB is a hierarchically structured sparse bottleneck which efficiently enforces sparsity using a multi-level mixture-of-experts approach: a hardware-aware implementation that closes the cost gap between activation sparse and dense training to a manageable interpretability tax. Empirically, ParityTransformers perform at least as well as post-hoc SAEs on sparse probing tasks, while out-performing on measures of feature absorption, steering effectiveness, and fine-grained causal interventions. Because subsequent computation acts only on features that survive the sparse bottleneck, the ParityTransformer's features are native to the model's forwards pass by construction, addressing the question of whether SAEs probe features the model actually uses during computation. We see this as a step toward training models whose internal representations are interpretable by design rather than recovered post hoc.

Mechanistically Eliciting Latent Behaviors in Language Models

We aim to discover diverse, generalizable perturbations of LLM internals that can surface hidden behavioral modes. Such perturbations could help reshape model behavior and systematically evaluate potential risks. We introduce Causal Perturbative Elicitation (CPE), an unsupervised method for discovering interpretable low-rank adapters (LoRAs) that can elicit these latent behaviors. CPE decomposes the computations of a deep transformer slice using a heuristic tensor-decomposition-based algorithm. CPE exhibits remarkable data efficiency, learning a large number of interpretable LoRAs from a single example. Even though CPE is unsupervised, we find that in some cases it can be competitive with supervised elicitation methods via brute-force enumerative search over weight space. For instance, CPE performs similarly to matched-wall-clock-time GRPO on the Countdown task for Qwen3-8B (85% vs 87%), demonstrating that CPE can efficiently elicit complex multi-token behaviors. Since CPE is unsupervised, it can also surface hidden failure modes, such as sandbagging, restoring 85% of locked BigCodeBench performance on a password-locked version of Llama3-70B introduced by Taylor et al. (2025). Additionally, since CPE explores behaviors in weight-space rather than token-space it can potentially ameliorate exploration hacking, a misalignment failure which may arise in sufficiently self-aware AI models (Ngo, 2022). In fact, we find that CPE virtually eliminates alignment-faking (Greenblatt et al., 2024) behavior in a Llama3-70B-based model organism developed by Hughes et al. (2025). Finally, we find that CPE can be used to initialize GPT-OSS-20B in an aligned basin when running GRPO on an environment prone to reward-hacking. By providing a data-efficient method to systematically explore the space of latent model behaviors, CPE yields a powerful tool for aligning AI systems and evaluating their safety.

Towards Worst-Case Guarantees with Scale-Aware Interpretability

Neural networks organize information according to the hierarchical, multi-scale structure of natural data. Methods to interpret model internals should be similarly scale-aware, explicitly tracking how features compose across resolutions and guaranteeing bounds on the influence of fine-grained structure that is discarded as irrelevant noise. We posit that the renormalisation framework from physics can meet this need by offering technical tools that can overcome limitations of current methods. Moreover, relevant work from adjacent fields has now matured to a point where scattered research threads can be synthesized into practical, theory-informed tools. To combine these threads in an AI safety context, we propose a unifying research agenda -- \emph{scale-aware interpretability} -- to develop formal machinery and interpretability tools that have robustness and faithfulness properties supported by statistical physics.