Shashwat Singh

Shashwat Singh, Tal Linzen, Shauli Ravfogel

Can large language models detect and report their own internal states? A number of studies have argued that the answer to this question is yes. We argue, based on lessons from human metacognition research, that this conclusion may be premature: to be convinced of this conclusion we need to distinguish genuine introspection from pattern matching based on surface-level cues. Furthermore, we argue that behavioral evidence alone is inherently insufficient to establish strong introspective claims. We re-examine two recently introduced evaluation paradigms in light of this consideration. In the first paradigm, models are expected to detect whether their internal states have been tampered with. We find that models cannot reliably distinguish such interventions on their internal states from manipulations of the input, suggesting that their success in the original studies reflects their ability to detect anomalies more generally, as opposed to interventions on their internal states in particular. In the second paradigm we examine, models are tasked with predicting labels derived from their own hidden states. Here, we find that classifiers that only have access to the input achieve equivalent performance to the model's own in-context predictions, indicating that the original results do not conclusively demonstrate that the model has privileged access to its internal representations. We further introduce a relabeled control setting, where models cannot rely on the semantics of the task to solve it, and instead must rely on the internal representation; models perform closer to chance on this better-controlled version of the task. Taken together, these results indicate that current evidence is insufficient to establish that LLMs display metacognitive monitoring.

Ashish Bhoopesh Gulaguli, Shashwat Singh, Rakesh Kumar Bansal

In this work we extend the results developed in 2022 for a sequential change detection algorithm making use of Page's CUSUM statistic, the empirical distribution as an estimate of the pre-change distribution, and a universal code as a tool for estimating the post-change distribution, from the i.i.d. case to the Markov setup.

Shashwat Singh, Sheri Zhang, Spencer Matonis et al.

Mudskippers are unique amphibious fish capable of locomotion in diverse environments, including terrestrial surfaces, aquatic habitats, and highly viscous substrates such as mud. This versatile locomotion is largely enabled by their powerful tail, which stores and rapidly releases energy to produce impulsive jumps. Inspired by this biological mechanism, we present the design and development of a multi-terrain centimeter-scale skipping and crawling robot. The robot is predominantly 3D printed and features onboard sensing, computation, and power. It is equipped with two side fins for crawling, each integrated with a hall effect sensor for gait control, while a rotary springtail driven by a 10mm planetary gear motor enables continuous impulsive skipping across a range of substrates to achieve multi-terrain locomotion. We modeled and experimentally characterized the tail, identifying an optimal length of 25mm that maximizes the mean propulsive force (4N, peaks up to 6N) for forward motion. In addition, we evaluated skipping on substrates where fin based crawling alone fails, and varied the moisture content of uniform sand and bentonite clay powder to compare skipping with crawling. Skipping consistently produced higher mean velocities than crawling, particularly on viscous and granular media. Finally, outdoor tests on grass, loose sand, and hard ground confirmed that combining skipping on entangling and granular terrain with crawling on firm ground extends the operational range of the robot in real-world environments.

Tianhui Han, Shashwat Singh, Sarvesh Patil et al.

Origami-inspired mechanisms can transform flat sheets into functional three-dimensional dynamic structures that are lightweight, compact, and capable of complex motion. These properties make origami increasingly valuable in robotic and deployable systems. However, accurately simulating their folding behavior and interactions with the environment remains challenging. To address this, we present a design framework for origami mechanism simulation that utilizes MuJoCo's deformable-body capabilities. In our approach, origami sheets are represented as graphs of interconnected deformable elements with user-specified constraints such as creases and actuation, defined through an intuitive graphical user interface (GUI). This framework allows users to generate physically consistent simulations that capture both the geometric structure of origami mechanisms and their interactions with external objects and surfaces. We demonstrate our method's utility through a case study on an origami catapult, where design parameters are optimized in simulation using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) and validated experimentally on physical prototypes. The optimized structure achieves improved throwing performance, illustrating how our system enables rapid, simulation-driven origami design, optimization, and analysis.

Shashwat Singh, Shauli Ravfogel, Jonathan Herzig et al.

Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. In the case of neural language models, an encoding of the undesirable behavior is often present in the model's representations. Thus, one natural (and common) approach to prevent the model from exhibiting undesirable behavior is to steer the model's representations in a manner that reduces the probability of it generating undesirable text. This paper investigates the formal and empirical properties of steering functions, i.e., transformation of the neural language model's representations that alter its behavior. First, we derive two optimal, in the least-squares sense, affine steering functions under different constraints. Our theory provides justification for existing approaches and offers a novel, improved steering approach. Second, we offer a series of experiments that demonstrate the empirical effectiveness of the methods in mitigating bias and reducing toxic generation.