A robot in a human-centric environment needs to account for the human's intent and future motion in its task and motion planning to ensure safe and effective operation. This requires symbolic reasoning about probable future actions and the ability to tie these actions to specific locations in the physical environment. While one can train behavioral models capable of predicting human motion from past activities, this approach requires large amounts of data to achieve acceptable long-horizon predictions. More importantly, the resulting models are constrained to specific data formats and modalities. Moreover, connecting predictions from such models to the environment at hand to ensure the applicability of these predictions is an unsolved problem. We present a system that utilizes a Large Language Model (LLM) to infer a human's next actions from a range of modalities without fine-tuning. A novel aspect of our system that is critical to robotics applications is that it links the predicted actions to specific locations in a semantic map of the environment. Our method leverages the fact that LLMs, trained on a vast corpus of text describing typical human behaviors, encode substantial world knowledge, including probable sequences of human actions and activities. We demonstrate how these localized activity predictions can be incorporated in a human-aware task planner for an assistive robot to reduce the occurrences of undesirable human-robot interactions by 29.2% on average.

Humans possess a strong capability for reasoning beyond common sense. For example, given an unconventional image of a goldfish laying on the table next to an empty fishbowl, a human would effortlessly determine that the fish is not inside the fishbowl. The case, however, may be different for a vision-language model, whose reasoning could gravitate towards the common scenario that the fish is inside the bowl, despite the visual input. In this paper, we introduce a novel probing dataset named ROME (reasoning beyond commonsense knowledge) to evaluate whether the state-of-the-art pre-trained vision-language models have the reasoning capability to correctly interpret counter-intuitive content. ROME contains images that defy commonsense knowledge with regards to color, shape, material, size and positional relation. Experiments on the state-of-the-art pre-trained vision-language models reveal that most of these models are still largely incapable of interpreting counter-intuitive scenarios. We hope that ROME will spur further investigations on reasoning beyond commonsense knowledge in vision-language research.

Given a query and a document corpus, the information retrieval (IR) task is to output a ranked list of relevant documents. Combining large language models (LLMs) with embedding-based retrieval models, recent work shows promising results on the zero-shot retrieval problem, i.e., no access to labeled data from the target domain. Two such popular paradigms are generation-augmented retrieval or GAR (generate additional context for the query and then retrieve), and retrieval-augmented generation or RAG (retrieve relevant documents as context and then generate answers). The success of these paradigms hinges on (i) high-recall retrieval models, which are difficult to obtain in the zero-shot setting, and (ii) high-precision (re-)ranking models which typically need a good initialization. In this work, we propose a novel GAR-meets-RAG recurrence formulation that overcomes the challenges of existing paradigms. Our method iteratively improves retrieval (via GAR) and rewrite (via RAG) stages in the zero-shot setting. A key design principle is that the rewrite-retrieval stages improve the recall of the system and a final re-ranking stage improves the precision. We conduct extensive experiments on zero-shot passage retrieval benchmarks, BEIR and TREC-DL. Our method establishes a new state-of-the-art in the BEIR benchmark, outperforming previous best results in Recall@100 and nDCG@10 metrics on 6 out of 8 datasets, with up to 17% relative gains over the previous best.

Large language models (LLMs) have demonstrated remarkable capabilities in various tasks. However, their suitability for domain-specific tasks, is limited due to their immense scale at deployment, susceptibility to misinformation, and more importantly, high data annotation costs. We propose a novel Interactive Multi-Fidelity Learning (IMFL) framework for the cost-effective development of small domain-specific LMs under limited annotation budgets. Our approach formulates the domain-specific fine-tuning process as a multi-fidelity learning problem, focusing on identifying the optimal acquisition strategy that balances between low-fidelity automatic LLM annotations and high-fidelity human annotations to maximize model performance. We further propose an exploration-exploitation query strategy that enhances annotation diversity and informativeness, incorporating two innovative designs: 1) prompt retrieval that selects in-context examples from human-annotated samples to improve LLM annotation, and 2) variable batch size that controls the order for choosing each fidelity to facilitate knowledge distillation, ultimately enhancing annotation quality. Extensive experiments on financial and medical tasks demonstrate that IMFL achieves superior performance compared with single fidelity annotations. Given a limited budget of human annotation, IMFL significantly outperforms the human annotation baselines in all four tasks and achieves very close performance as human annotations on two of the tasks. These promising results suggest that the high human annotation costs in domain-specific tasks can be significantly reduced by employing IMFL, which utilizes fewer human annotations, supplemented with cheaper and faster LLM (e.g., GPT-3.5) annotations to achieve comparable performance.

Multi-agent systems driven by large language models (LLMs) have shown promising abilities for solving complex tasks in a collaborative manner. This work considers a fundamental problem in multi-agent collaboration: consensus seeking. When multiple agents work together, we are interested in how they can reach a consensus through inter-agent negotiation. To that end, this work studies a consensus-seeking task where the state of each agent is a numerical value and they negotiate with each other to reach a consensus value. It is revealed that when not explicitly directed on which strategy should be adopted, the LLM-driven agents primarily use the average strategy for consensus seeking although they may occasionally use some other strategies. Moreover, this work analyzes the impact of the agent number, agent personality, and network topology on the negotiation process. The findings reported in this work can potentially lay the foundations for understanding the behaviors of LLM-driven multi-agent systems for solving more complex tasks. Furthermore, LLM-driven consensus seeking is applied to a multi-robot aggregation task. This application demonstrates the potential of LLM-driven agents to achieve zero-shot autonomous planning for multi-robot collaboration tasks. Project website: westlakeintelligentrobotics.github.io/ConsensusLLM/.

Large language models (LLMs) have achieved significant progress from pre-training on and memorizing a wide range of textual data, however, this process might suffer from privacy issues and violations of data protection regulations. As a result, the ability to easily remove data related to individual users from such models while not deteriorating their predictive quality after the removal becomes increasingly important. To address these issues, in this work, we propose an efficient unlearning framework that could efficiently update LLMs without having to retrain the whole model after data removals, by introducing lightweight unlearning layers learned with a selective teacher-student objective into the transformers. In addition, we introduce a fusion mechanism to effectively combine different unlearning layers that learns to forget different sets of data to handle a sequence of forgetting operations. Experiments on classification and generation tasks demonstrate the effectiveness of our proposed methods compared to the state-of-the-art baselines.

Prompt tuning prepends a soft prompt to the input embeddings or hidden states and only optimizes the prompt to adapt pretrained models (PTMs) to downstream tasks. The previous work manually selects prompt layers which are far from optimal and failed to exploit the potential of prompt tuning. In this work, we propose a novel framework, \underline{S}elective \underline{P}rompt \underline{T}uning (SPT), that learns to select the proper prompt layers by inserting a prompt controlled by a learnable probabilistic gate at each intermediate layer. We further propose a novel bi-level optimization framework, SPT-DARTS, that can better optimize the learnable gates and improve the final prompt tuning performances of the learned prompt layer settings. We conduct extensive experiments with ten benchmark datasets under the full-data and few-shot scenarios. The results demonstrate that our SPT framework can perform better than the previous state-of-the-art PETuning baselines with comparable or fewer tunable parameters.

Although large language models (LLMs) have advanced the state-of-the-art in NLP significantly, deploying them for downstream applications is still challenging due to cost, responsiveness, control, or concerns around privacy and security. As such, trainable models are still the preferred option in some cases. However, these models still require human-labeled data for optimal performance, which is expensive and time-consuming to obtain. In order to address this issue, several techniques to reduce human effort involve labeling or generating data using LLMs. Although these methods are effective for certain applications, in practice they encounter difficulties in real-world scenarios. Labeling data requires careful data selection, while generating data necessitates task-specific prompt engineering. In this paper, we propose a unified data creation pipeline that requires only a single formatting example, and which is applicable to a broad range of tasks, including traditionally problematic ones with semantically devoid label spaces. In our experiments we demonstrate that instruction-following LLMs are highly cost-effective data creators, and that models trained with these data exhibit performance better than those trained with human-labeled data (by up to 17.5%) on out-of-distribution evaluation, while maintaining comparable performance on in-distribution tasks. These results have important implications for the robustness of NLP systems deployed in the real-world.

The accurate classification of student help requests with respect to the type of help being sought can enable the tailoring of effective responses. Automatically classifying such requests is non-trivial, but large language models (LLMs) appear to offer an accessible, cost-effective solution. This study evaluates the performance of the GPT-3.5 and GPT-4 models for classifying help requests from students in an introductory programming class. In zero-shot trials, GPT-3.5 and GPT-4 exhibited comparable performance on most categories, while GPT-4 outperformed GPT-3.5 in classifying sub-categories for requests related to debugging. Fine-tuning the GPT-3.5 model improved its performance to such an extent that it approximated the accuracy and consistency across categories observed between two human raters. Overall, this study demonstrates the feasibility of using LLMs to enhance educational systems through the automated classification of student needs.

In programming education, plagiarism and misuse of artificial intelligence (AI) assistance are emerging issues. However, not many relevant studies are focused on web programming. We plan to develop automated tools to help instructors identify both misconducts. To fully understand the issues, we conducted a controlled experiment to observe the unfair benefits and the characteristics. We compared student performance in completing web programming tasks independently, with a submission to plagiarize, and with the help of AI assistance (ChatGPT). Our study shows that students who are involved in such misconducts get comparable test marks with less completion time. Plagiarized submissions are similar to the independent ones except in trivial aspects such as color and identifier names. AI-assisted submissions are more complex, making them less readable. Students believe AI assistance could be useful given proper acknowledgment of the use, although they are not convinced with readability and correctness of the solutions.