Simulating learner actions helps stress-test open-ended interactive learning environments and prototype new adaptations before deployment. While recent studies show the promise of using large language models (LLMs) for simulating human behavior, such approaches have not gone beyond rudimentary proof-of-concept stages due to key limitations. First, LLMs are highly sensitive to minor prompt variations, raising doubts about their ability to generalize to new scenarios without extensive prompt engineering. Moreover, apparently successful outcomes can often be unreliable, either because domain experts unintentionally guide LLMs to produce expected results, leading to self-fulfilling prophecies; or because the LLM has encountered highly similar scenarios in its training data, meaning that models may not be simulating behavior so much as regurgitating memorized content. To address these challenges, we propose Hyp-Mix, a simulation authoring framework that allows experts to develop and evaluate simulations by combining testable hypotheses about learner behavior. Testing this framework in a physics learning environment, we found that GPT-4 Turbo maintains calibrated behavior even as the underlying learner model changes, providing the first evidence that LLMs can be used to simulate realistic behaviors in open-ended interactive learning environments, a necessary prerequisite for useful LLM behavioral simulation.

Despite the success of Instruction Tuning (IT) in training large language models (LLMs), such models often leverage spurious or biased features learnt from their training data and can become misaligned, leading to undesired behaviours. While existing techniques can steer model behaviour at inference-time, they are often post-hoc and do not embed steering as an intrinsic model feature. In this work, we introduce Focus Instruction Tuning (FIT), which trains LLMs to condition their responses by focusing on specific features whilst ignoring others, leading to different behaviours based on what features are specified. Across diverse benchmarks, we demonstrate that FIT: (i) successfully steers behaviour at inference time; (ii) increases robustness by amplifying core task signals and down-weighting spurious cues; (iii) mitigates social bias by suppressing demographic attributes; and (iv) generalises under distribution shifts and to previously unseen focus features. FIT therefore offers a lightweight, intrinsic mechanism for building more robust, fair, and easily controllable LLMs.

Large Language Models (LLMs) have demonstrated impressive abilities in symbol processing through in-context learning (ICL). This success flies in the face of decades of critiques asserting that artificial neural networks cannot master abstract symbol manipulation. We seek to understand the mechanisms that can enable robust symbol processing in transformer networks, illuminating both the unanticipated success, and the significant limitations, of transformers in symbol processing. Borrowing insights from symbolic AI and cognitive science on the power of Production System architectures, we develop a high-level Production System Language, PSL, that allows us to write symbolic programs to do complex, abstract symbol processing, and create compilers that precisely implement PSL programs in transformer networks which are, by construction, 100% mechanistically interpretable. The work is driven by study of a purely abstract (semantics-free) symbolic task that we develop, Templatic Generation (TGT). Although developed through study of TGT, PSL is, we demonstrate, highly general: it is Turing Universal. The new type of transformer architecture that we compile from PSL programs suggests a number of paths for enhancing transformers’ capabilities at symbol processing. We note, however, that the work we report addresses computability, and not learnability, by transformer networks.

Neural image classifiers are known to undergo severe performance degradation when exposed to inputs that are sampled from environmental conditions that differ from their training data. Given the recent progress in Text-to-Image (T2I) generation, a natural question is how modern T2I generators can be used to simulate arbitrary interventions over such environmental factors in order to augment training data and improve the robustness of downstream classifiers. We experiment across a diverse collection of benchmarks in single domain generalization (SDG) and reducing reliance on spurious features (RRSF), ablating across key dimensions of T2I generation, including interventional prompting strategies, conditioning mechanisms, and post-hoc filtering, showing that modern T2I generators like Stable Diffusion can indeed be used to implement a powerful interventional data augmentation (IDA) mechanism, outperforming previously state-of-the-art data augmentation techniques regardless of how each dimension is configured.

Despite the importance of shape perception in human vision, early neural image classifiers relied less on shape information for object recognition than other (often spurious) features. While recent research suggests that current large Vision-Language Models (VLMs) exhibit more reliance on shape, we find them to still be seriously limited in this regard. To quantify such limitations, we introduce IllusionBench, a dataset that challenges current cutting-edge VLMs to decipher shape information when the shape is represented by an arrangement of visual elements in a scene. Our extensive evaluations reveal that, while these shapes are easily detectable by human annotators, current VLMs struggle to recognize them, indicating important avenues for future work in developing more robust visual perception systems. The full dataset and codebase are available at: https://arshiahemmat.github.io/illusionbench/

Walkability has many health, environmental, and economic benefits. That is why web and mobile services have been offering ways of computing walkability scores of individual street segments. Those scores are generally computed from survey data and manual counting (of even trees). However, that is costly, owing to the high time, effort, and financial costs. To partly automate the computation of those scores, we explore the possibility of using the social media data of Flickr and Foursquare to automatically identify safe and walkable streets. We find that unsafe streets tend to be photographed during the day, while walkable streets are tagged with walkability-related keywords. These results open up practical opportunities (for, e.g., room booking services, urban route recommenders, and real-estate sites) and have theoretical implications for researchers who might resort to the use social media data to tackle previously unanswered questions in the area of walkability.