Better understanding of Large Language Models' (LLMs) legal analysis abilities can contribute to improving the efficiency of legal services, governing artificial intelligence, and leveraging LLMs to identify inconsistencies in law. This paper explores LLM capabilities in applying tax law. We choose this area of law because it has a structure that allows us to set up automated validation pipelines across thousands of examples, requires logical reasoning and maths skills, and enables us to test LLM capabilities in a manner relevant to real-world economic lives of citizens and companies. Our experiments demonstrate emerging legal understanding capabilities, with improved performance in each subsequent OpenAI model release. We experiment with retrieving and utilising the relevant legal authority to assess the impact of providing additional legal context to LLMs. Few-shot prompting, presenting examples of question-answer pairs, is also found to significantly enhance the performance of the most advanced model, GPT-4. The findings indicate that LLMs, particularly when combined with prompting enhancements and the correct legal texts, can perform at high levels of accuracy but not yet at expert tax lawyer levels. As LLMs continue to advance, their ability to reason about law autonomously could have significant implications for the legal profession and AI governance.

In this work, we empirically show that updating pretrained LMs (350M, 1.3B, 2.7B) with just a few steps of Gradient Ascent Post-training (GAP) on random, unlabeled text corpora enhances its zero-shot generalization capabilities across diverse NLP tasks. Specifically, we show that GAP can allow LMs to become comparable to 2-3x times larger LMs across 12 different NLP tasks. We also show that applying GAP on out-of-distribution corpora leads to the most reliable performance improvements. Our findings indicate that GAP can be a promising method for improving the generalization capability of LMs without any task-specific fine-tuning.

This paper describes the results of the first shared task on the generation of teacher responses in educational dialogues. The goal of the task was to benchmark the ability of generative language models to act as AI teachers, replying to a student in a teacher-student dialogue. Eight teams participated in the competition hosted on CodaLab. They experimented with a wide variety of state-of-the-art models, including Alpaca, Bloom, DialoGPT, DistilGPT-2, Flan-T5, GPT-2, GPT-3, GPT- 4, LLaMA, OPT-2.7B, and T5-base. Their submissions were automatically scored using BERTScore and DialogRPT metrics, and the top three among them were further manually evaluated in terms of pedagogical ability based on Tack and Piech (2022). The NAISTeacher system, which ranked first in both automated and human evaluation, generated responses with GPT-3.5 using an ensemble of prompts and a DialogRPT-based ranking of responses for given dialogue contexts. Despite the promising achievements of the participating teams, the results also highlight the need for evaluation metrics better suited to educational contexts.

In zero-shot learning (ZSL), generative methods synthesize class-related sample features based on predefined semantic prototypes. They advance the ZSL performance by synthesizing unseen class sample features for better training the classifier. We observe that each class's predefined semantic prototype (also referred to as semantic embedding or condition) does not accurately match its real semantic prototype. So the synthesized visual sample features do not faithfully represent the real sample features, limiting the classifier training and existing ZSL performance. In this paper, we formulate this mismatch phenomenon as the visual-semantic domain shift problem. We propose a dynamic semantic prototype evolving (DSP) method to align the empirically predefined semantic prototypes and the real prototypes for class-related feature synthesis. The alignment is learned by refining sample features and semantic prototypes in a unified framework and making the synthesized visual sample features approach real sample features. After alignment, synthesized sample features from unseen classes are closer to the real sample features and benefit DSP to improve existing generative ZSL methods by 8.5\%, 8.0\%, and 9.7\% on the standard CUB, SUN AWA2 datasets, the significant performance improvement indicates that evolving semantic prototype explores a virgin field in ZSL.

Generating intermediate steps, or Chain of Thought (CoT), is an effective way to significantly improve language models' (LM) multi-step reasoning capability. However, the CoT lengths can grow rapidly with the problem complexity, easily exceeding the maximum context size. Instead of increasing the context limit, which has already been heavily investigated, we explore an orthogonal direction: making LMs divide a problem into multiple contexts. We propose a new inference framework, called Recursion of Thought (RoT), which introduces several special tokens that the models can output to trigger context-related operations. Extensive experiments with multiple architectures including GPT-3 show that RoT dramatically improves LMs' inference capability to solve problems, whose solution consists of hundreds of thousands of tokens.

Stickers have become a ubiquitous part of modern-day communication, conveying complex emotions through visual imagery. To facilitate the development of more powerful algorithms for analyzing stickers, we propose a large-scale Chinese sticker dataset, namely Sticker820K, which consists of 820k image-text pairs. Each sticker has rich and high-quality textual annotations, including descriptions, optical characters, emotional labels, and style classifications. Although vision-language tasks in the domain of natural images have been well studied, directly applying the those models, such as CLIP, to sticker data is not an optimal solution due to the discrepant nature between natural and emotive image data. Therefore, we propose StickerCLIP as a benchmark model on the Sticker820K dataset. For the text-to-image retrieval task, our StickerCLIP demonstrates strong superiority over the CLIP, which achieves an absolute gain of 66.0\% in mean recall on the Sticker820K test set. Additionally, we endeavor to extend the recently popularized LLM by means of prompt tuning, integrating its ability for sticker retrieval and allowing users to retrieve stickers through instructions. We validate the feasibility of this method, demonstrating the immense potential of prompt tuning in expanding LLM abilities while not affecting the quality of upstream tasks.

Electronic health records (EHRs) serve as an essential data source for the envisioned artificial intelligence (AI)-driven transformation in healthcare. However, clinician biases reflected in EHR notes can lead to AI models inheriting and amplifying these biases, perpetuating health disparities. This study investigates the impact of stigmatizing language (SL) in EHR notes on mortality prediction using a Transformer-based deep learning model and explainable AI (XAI) techniques. Our findings demonstrate that SL written by clinicians adversely affects AI performance, particularly so for black patients, highlighting SL as a source of racial disparity in AI model development. To explore an operationally efficient way to mitigate SL's impact, we investigate patterns in the generation of SL through a clinicians' collaborative network, identifying central clinicians as having a stronger impact on racial disparity in the AI model. We find that removing SL written by central clinicians is a more efficient bias reduction strategy than eliminating all SL in the entire corpus of data. This study provides actionable insights for responsible AI development and contributes to understanding clinician behavior and EHR note writing in healthcare.

This perspective paper proposes a series of interactive scenarios that utilize Artificial Intelligence (AI) to enhance classroom teaching, such as dialogue auto-completion, knowledge and style transfer, and assessment of AI-generated content. By leveraging recent developments in Large Language Models (LLMs), we explore the potential of AI to augment and enrich teacher-student dialogues and improve the quality of teaching. Our goal is to produce innovative and meaningful conversations between teachers and students, create standards for evaluation, and improve the efficacy of AI-for-Education initiatives. In Section 3, we discuss the challenges of utilizing existing LLMs to effectively complete the educated tasks and present a unified framework for addressing diverse education dataset, processing lengthy conversations, and condensing information to better accomplish more downstream tasks. In Section 4, we summarize the pivoting tasks including Teacher-Student Dialogue Auto-Completion, Expert Teaching Knowledge and Style Transfer, and Assessment of AI-Generated Content (AIGC), providing a clear path for future research. In Section 5, we also explore the use of external and adjustable LLMs to improve the generated content through human-in-the-loop supervision and reinforcement learning. Ultimately, this paper seeks to highlight the potential for AI to aid the field of education and promote its further exploration.

Embodied AI focuses on the study and development of intelligent systems that possess a physical or virtual embodiment (i.e. robots) and are able to dynamically interact with their environment. Memory and control are the two essential parts of an embodied system and usually require separate frameworks to model each of them. In this paper, we propose a novel and generalizable framework called LLM-Brain: using Large-scale Language Model as a robotic brain to unify egocentric memory and control. The LLM-Brain framework integrates multiple multimodal language models for robotic tasks, utilizing a zero-shot learning approach. All components within LLM-Brain communicate using natural language in closed-loop multi-round dialogues that encompass perception, planning, control, and memory. The core of the system is an embodied LLM to maintain egocentric memory and control the robot. We demonstrate LLM-Brain by examining two downstream tasks: active exploration and embodied question answering. The active exploration tasks require the robot to extensively explore an unknown environment within a limited number of actions. Meanwhile, the embodied question answering tasks necessitate that the robot answers questions based on observations acquired during prior explorations.

Artificial agents have traditionally been trained to maximize reward, which may incentivize power-seeking and deception, analogous to how next-token prediction in language models (LMs) may incentivize toxicity. So do agents naturally learn to be Machiavellian? And how do we measure these behaviors in general-purpose models such as GPT-4? Towards answering these questions, we introduce MACHIAVELLI, a benchmark of 134 Choose-Your-Own-Adventure games containing over half a million rich, diverse scenarios that center on social decision-making. Scenario labeling is automated with LMs, which are more performant than human annotators. We mathematize dozens of harmful behaviors and use our annotations to evaluate agents' tendencies to be power-seeking, cause disutility, and commit ethical violations. We observe some tension between maximizing reward and behaving ethically. To improve this trade-off, we investigate LM-based methods to steer agents' towards less harmful behaviors. Our results show that agents can both act competently and morally, so concrete progress can currently be made in machine ethics--designing agents that are Pareto improvements in both safety and capabilities.