Cross-seed dictionary stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured cost, while recovering over half of them, and the external check shows the surfaced motifs are significantly enriched for their claimed annotations.
Abstract
Sparse autoencoders (SAEs), and other interpretability methods could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. However, a latent is only useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemble them into a single pipeline and report the practical innovations each stage required. First, cross-seed dictionary stability prioritises which latents are worth spending resources to investigate. Second, an intruder-detection task asks whether a latents activating examples share a recognizable pattern. Third, a separate pass proposes a candidate biological description which we convert into falsifiable predictions which can be tested in silico. Deployed on the Boltz-1 Pairformer trunk, stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured cost, while recovering over half of them, and the external check shows the surfaced motifs are significantly enriched for their claimed annotations. The results also suggest a possible tension: the cross- seed stability might be selecting for some types of features, like structure-related ones, much more than others, such as function-related features.
The Koopman spectrum is an identifiable, model-intrinsic fingerprint with a stated error bar, not a legible decomposition, and the spectrum is recoverable from calibration samples at rate $M^{-1/2}$ up to permutation.
Across AI-driven science, generative capacity has outpaced investigative judgment. It is now possible to produce millions of plausible candidates in hours, but not yet to decide which merit expensive validation. This asymmetry defines the central bottleneck of scientific discovery for the next decade. This blue sky paper casts scientific investigation as an information market. A strategically intelligent agent must exploit cost-information asymmetries across fidelity levels, purchasing cheap inferences to decide whether expensive ones are worth acquiring. The paper formalizes this principle as Inferential Arbitrage: a multi-fidelity Markov Decision Process in which agents maximize inferential return on investment. The formulation is pluralistic by design, admitting reinforcement learning, evolutionary algorithms, and multi-armed bandits. The framework is developed in protein design, where the vendor hierarchy from sequence embeddings to wet-lab assays spans several orders of magnitude, but the same structure recurs in materials science and autonomous chemistry. Three grand challenges (cross-modal calibration, asynchronous decision logic, interpretable strategy) anchor a 2030 benchmark: an agent that discovers a functional protein at 1% of the compute and 10% of the lab cost of human-designed pipelines. The ultimate measure of an autonomous AI scientist is not what it can compute, but what it chooses not to.
Amarda Shehu· Proceedings of the 32nd ACM...· 0 citations
P-Bench is built, a benchmark comprising 425 open-ended, realistic hypothesis-testing tasks spanning economics, biology, and medicine and introduces Fisher-R1, an open-weight LLM agent trained for rigorous hypothesis testing using synthetic tasks and reinforcement learning.
Jia-Cheng Miao, Jin Mu, Guanhua Chen et al.· 0 citations
Recent work has shown that large language models (LLMs) exhibit strong numerical sequence modeling capabilities and show promise in time-series prediction. While LLMs display in-context learning capabilities, the mechanisms with which they accomplish time-series prediction remain unclear. Specifically, whether they truly understand the underlying structure, which at a minimum requires reasoning over first differences in the sequence of numbers. To study this, we investigate Llama 3.1-8B from a mechanistic interpretability point of view. Mechanistic interpretability is an emerging field concerned with the reverse engineering of the algorithms learned by neural networks such as LLMs. To assess Llamas'numerical sequence modeling capabilities and to facilitate our mechanistic interpretability analysis, we create a sequence modeling task that cannot be solved without picking up structural cues. Specifically, we sample n random numbers and repeat them with an offset. We find that Llama displays strong performance on our tasks suggesting that it can pick up on the underlying structure. To understand the mechanisms that allow it to do so, we perform probing experiments and activation patching based counterfactual analysis. Probing reveals that the model computes and stores first differences in its internal representations without explicit supervision, indicating that it tracks structural information about the sequence. Activation patching reveals that Llama retrieves the relevant first-difference with a mechanism similar to an induction circuit and subsequently adds it to the current value. Notably, our work represents one of the first studies to identify this form of concept induction in LLMs.
Rahul Chowdhury, Timothy Rupprecht, Senhao Cao et al.· 0 citations
A central challenge in interpreting learned decision-making systems is to determine whether their internal representations contain concepts that help explain their behavior. We report interpretability experiments for a tokenized autoregressive Transformer agent in the Game of Hidden Rules (GOHR). We focus on a compact two-rule task in which both hidden rules map object shapes to target buckets, but with different permutations. The policy is trained on episodes sampled from these two hidden rules and then evaluated with fixed weights. It is never given a rule label and does not use an explicit rule classifier; any rule information must be inferred implicitly from interaction history. In this setting, the correct rule is not identifiable before the agent tries an informative move and observes accept/reject feedback. Sparse autoencoders (SAEs) trained on the agent's decision-token embeddings recover this structure. When held-out decisions are labeled by simple concepts such as the chosen shape or bucket, SAE dimensions that are highly selective for a concept cover most decisions where that concept is present. Individual SAE dimensions also correspond to interpretable strategies such as probing one rule hypothesis and switching after negative feedback.
CON decomposition is introduced, which quantifies how much of a layer's variance each concept explains given all other concepts and the outcome, and how much none of them explains, yielding layer-comparable, calibrated scores that suppress false positives.
R. Rane, Marco Simnacher, Manuel Pfeuffer et al.· 0 citations
