David Steinmann

Lukas Helff, Quentin Delfosse, David Steinmann et al.

As reinforcement Learning with Verifiable Rewards (RLVR) has become the dominant paradigm for scaling reasoning capabilities in LLMs, a new failure mode emerges: LLMs gaming verifiers. We study this phenomenon on inductive reasoning tasks, where models must induce and output logical rules. We find that RLVR-trained models systematically abandon rule induction. Instead of learning generalizable patterns (e.g., ``trains carrying red cars go east''), they enumerate instance-level labels, producing outputs that pass verifiers without capturing the relational patterns required by the task. We show that this behavior is not a failure of understanding but a form of reward hacking: imperfect verifiers that check only extensional correctness admit false positives. To detect such shortcuts, we introduce Isomorphic Perturbation Testing (IPT), which evaluates a single model output under both extensional and isomorphic verification, where the latter enforces invariance under logically isomorphic tasks. While genuine rule induction remains invariant, shortcut strategies fail. We find that shortcut behavior is specific to RLVR-trained reasoning models (e.g., GPT-5, Olmo3) and absent in non-RLVR models (e.g., GPT-4o, GPT-4.5, Ministral). Moreover, shortcut prevalence increases with task complexity and inference-time compute. In controlled training experiments, extensional verification directly induces shortcut strategies, while isomorphic verification eliminates them. These results show that RLVR can incentivize reward hacking not only through overt manipulation but also by exploiting what the verifier fails to enforce.

David Steinmann, Antonia Wüst, Kristian Kersting et al.

While interpretable models such as concept bottleneck models (CBMs) and program synthesis methods enable verification of model decisions, their evaluation is typically limited to simple tasks, leaving complex reasoning on real-world images largely unexplored. We introduce COCOLogic-V2, an object-centric dataset for visual inductive reasoning on real-world images covering a broad subset of first-order logic. By categorizing samples into positive variants, near-boundary (NB), and far-from-boundary (FB) negatives, COCOLogic-V2 enables fine-grained diagnosis of model accountability. Our evaluations show that models tend to separate positive and FB samples well but fail on NB samples, while perceptual noise and large rule-induced search spaces pose additional challenges in few-shot settings. Together, these results highlight that visual inductive reasoning remains an open challenge and COCOLogic-V2 provides a concrete foundation for advancing methods in this direction.

Nils Grandien, David Steinmann, Felix Friedrich et al.

Sparse autoencoders (SAEs) have become an important tool for unsupervised concept discovery in large models. To make the resulting feature spaces more interpretable and manageable, recent approaches have begun imposing hierarchical structure, either explicitly or as an implicit effect of training constraints, yet rigorous comparison remains difficult. There are no agreed-upon requirements for what a meaningful feature hierarchy should satisfy, and evaluation has largely relied on qualitative illustrations with fragmented quantitative protocols. To address this, we derive a set of key requirements for generalization/specialization hierarchies in unsupervised concept discovery, drawing on semantic net and taxonomy research alongside recent SAE work, and use them to derive a concrete evaluation protocol. Applying this protocol to current SAE approaches trained on visual data, we find that while feature spaces generally provide a basis for sensible hierarchies, establishing good hierarchical structure remains challenging. In particular, feature absorption, both in its well-known hard form and in a continuous, soft form, systematically compromises hierarchy quality, pointing to a fundamental tension that future approaches will need to navigate.