Motivated by the surge of large language models, there has been a push to formally characterize the symbolic abilities intrinsic to the transformer architecture. A programming language, called RASP, has been proposed, which can be directly compiled into transformer weights to implement these algorithms. However, the tasks that can be implemented in RASP are often uncommon to learn from natural unsupervised data, showing a mismatch between theoretical capabilities of the transformer architecture, and the practical learnability of these capabilities from unsupervised data. We propose tracr-injection, a method that allows us to distill algorithms written in RASP directly into a pre-trained language model. We showcase our method by injecting 3 different algorithms into a language model. We show how our method creates an interpretable subspace within the model's residual stream, which can be decoded into the variables present in the code of the RASP algorithm. Additionally, we found that the proposed method can improve out-of-distribution performance compared to our baseline, indicating that indeed a more symbolic mechanism is taking place in the inner workings of the model. We release the code used to run our experiments.

Open-domain semantic parsing remains a challenging task, as models often rely on heuristics and struggle to handle unseen concepts. In this paper, we investigate the potential of large language models (LLMs) for this task and introduce Retrieval-Augmented Semantic Parsing (RASP), a simple yet effective approach that integrates external lexical knowledge into the parsing process. Our experiments not only show that LLMs outperform previous encoder-decoder baselines for semantic parsing, but that RASP further enhances their ability to predict unseen concepts, nearly doubling the performance of previous models on out-of-distribution concepts. These findings highlight the promise of leveraging large language models and retrieval mechanisms for robust and open-domain semantic parsing.

Achieving a mechanistic understanding of transformer-based language models is an open challenge, especially due to their large number of parameters. Moreover, the lack of ground truth mappings between model weights and their functional roles hinders the effective evaluation of interpretability methods, impeding overall progress. Tracr, a method for generating compiled transformers with inherent ground truth mappings in RASP, has been proposed to address this issue. However, manually creating a large number of models needed for verifying interpretability methods is labour-intensive and time-consuming. In this work, we present a novel approach for generating interpretability test beds using large language models (LLMs) and introduce TracrBench, a novel dataset consisting of 121 manually written and LLM-generated, human-validated RASP programs and their corresponding transformer weights. During this process, we evaluate the ability of frontier LLMs to autonomously generate RASP programs and find that this task poses significant challenges. GPT-4-turbo, with a 20-shot prompt and best-of-5 sampling, correctly implements only 57 out of 101 test programs, necessitating the manual implementation of the remaining programs. With its 121 samples, TracrBench aims to serve as a valuable testbed for evaluating and comparing interpretability methods.

Foundation models and large language models have shown immense human-like understanding and capabilities for generating text and digital media. However, foundation models that can freely sense, interact, and actuate the physical world like in the digital domain is far from being realized. This is due to a number of challenges including: 1) being constrained to the types of static devices and sensors deployed, 2) events often being localized to one part of a large space, and 3) requiring dense and deployments of devices Figure 1: RASP autonomous payload reconfiguration to achieve full coverage. As a critical step towards enabling to execute user specified task foundation models to successfully and freely interact with the physical environment, we propose RASP, a modular and Scaling up, there are few works that explore the use reconfigurable sensing and actuation platform that allows of LLMs to actuate our environments, particularly smart drones to autonomously swap onboard sensors and actuators homes [16, 17, 32], where events and actions may occur anywhere in only 25 seconds, allowing a single drone to quickly adapt in the space. These works generally focus on adapting to a diverse range of tasks. We demonstrate through real LLMs as a human-like interface to actuate common internetconnected smart home deployments that RASP enables FMs and LLMs smart appliances (e.g., speakers, television, air to complete diverse tasks up to 85% more successfully by conditioning, etc.). Much like how FMs enable general human allowing them to target specific areas with specific sensors language and sensory understanding and responses, and actuators on-the-fly.

The availability of large annotated data can be a critical bottleneck in training machine learning algorithms successfully, especially when applied to diverse domains. Weak supervision offers a promising alternative by accelerating the creation of labeled training data using domain-specific rules. However, it requires users to write a diverse set of high-quality rules to assign labels to the unlabeled data. Automatic Rule Induction (ARI) approaches circumvent this problem by automatically creating rules from features on a small labeled set and filtering a final set of rules from them. In the ARI approach, the crucial step is to filter out a set of a high-quality useful subset of rules from the large set of automatically created rules. In this paper, we propose an algorithm (Filtering of Automatically Induced Rules) to filter rules from a large number of automatically induced rules using submodular objective functions that account for the collective precision, coverage, and conflicts of the rule set. We experiment with three ARI approaches and five text classification datasets to validate the superior performance of our algorithm with respect to several semi-supervised label aggregation approaches. Further, we show that achieves statistically significant results in comparison to existing rule-filtering approaches.

We propose a novel domain generalization technique, referred to as Randomized Adversarial Style Perturbation (RASP), which is motivated by the observation that the characteristics of each domain are captured by the feature statistics corresponding to style. The proposed algorithm perturbs the style of a feature in an adversarial direction towards a randomly selected class, and makes the model learn against being misled by the unexpected styles observed in unseen target domains. While RASP is effective to handle domain shifts, its naive integration into the training procedure might degrade the capability of learning knowledge from source domains because it has no restriction on the perturbations of representations. This challenge is alleviated by Normalized Feature Mixup (NFM), which facilitates the learning of the original features while achieving robustness to perturbed representations via their mixup during training. We evaluate the proposed algorithm via extensive experiments on various benchmarks and show that our approach improves domain generalization performance, especially in large-scale benchmarks.

Language models (LMs) proficiency in handling deterministic symbolic reasoning and rule-based tasks remains limited due to their dependency implicit learning on textual data. To enable fully rule comprehension ability, we explore how to incorporate compiled neural networks (CoNNs) which weight is specially designed into the architecture of LMs, to achieve high accuracy and robust performance. CoNNs are transformer-based neural networks that execute rules through artificially generated attention weights. Our method, which call "Neural Comprehension", by incorporating CoNN modules into the LM, the framework effectively tackles rule-intensive challenges. Our experiments on symbolic reasoning tasks and real-world arithmetic reasoning tasks demonstrate the superior performance of our method compared to existing techniques. Furthermore, our LM achieves flawless execution on symbolic operations tasks, highlighting the potential of our method in enabling LMs to possess true symbolic comprehension capabilities. Our code is publicly available at: https://github.com/WENGSYX/Neural-Comprehension.

Preference-based reasoning is a form of commonsense reasoning that makes many problems easier to express and sometimes more likely to have a solution. We present an approach to introduce preferences in the weight constraint construct, which is a very useful programming construct widely adopted in Answer Set Programming (ASP). We show the usefulness of the proposed extension, and we outline how to accordingly extend the ASP semantics.