Recent work integrating Large Language Models (LLMs) has led to significant improvements in the Knowledge Base Question Answering (KBQA) task. However, we posit that existing KBQA datasets that either have simple questions, use synthetically generated logical forms, or are based on small knowledge base (KB) schemas, do not capture the true complexity of KBQA tasks. To address this, we introduce the SPINACH dataset, an expert-annotated KBQA dataset collected from forum discussions on Wikidata's "Request a Query" forum with 320 decontextualized question-SPARQL pairs. Much more complex than existing datasets, SPINACH calls for strong KBQA systems that do not rely on training data to learn the KB schema, but can dynamically explore large and often incomplete schemas and reason about them. Along with the dataset, we introduce the SPINACH agent, a new KBQA approach that mimics how a human expert would write SPARQLs for such challenging questions. Experiments on existing datasets show SPINACH's capability in KBQA, achieving a new state of the art on the QALD-7, QALD-9 Plus and QALD-10 datasets by 30.1%, 27.0%, and 10.0% in F1, respectively, and coming within 1.6% of the fine-tuned LLaMA SOTA model on WikiWebQuestions. On our new SPINACH dataset, SPINACH agent outperforms all baselines, including the best GPT-4-based KBQA agent, by 38.1% in F1.

A crucial requirement for deploying LLM-based agents in real-life applications is robustness against risky or irreversible mistakes. However, existing research lacks a focus on the preemptive evaluation of reasoning trajectories performed by LLM agents, leading to a gap in ensuring safe and reliable operations. To explore better solutions, this paper introduces InferAct, a novel approach that leverages the Theory-of-Mind capability of LLMs to proactively detect potential errors before critical actions are executed (e.g., "buy-now" in automatic online trading or web shopping). InferAct is also capable of integrating human feedback to prevent irreversible risks and enhance the actor agent's decision-making process. Experiments on three widely used tasks demonstrate the effectiveness of InferAct. The proposed solution presents a novel approach and concrete contributions toward developing LLM agents that can be safely deployed in different environments involving critical decision-making.

We study the problem of continual test-time adaption where the goal is to adapt a source pre-trained model to a sequence of unlabelled target domains at test time. Existing methods on test-time training suffer from several limitations: (1) Mismatch between the feature extractor and classifier; (2) Interference between the main and self-supervised tasks; (3) Lack of the ability to quickly adapt to the current distribution. In light of these challenges, we propose a cascading paradigm that simultaneously updates the feature extractor and classifier at test time, mitigating the mismatch between them and enabling long-term model adaptation. The pre-training of our model is structured within a meta-learning framework, thereby minimizing the interference between the main and self-supervised tasks and encouraging fast adaptation in the presence of limited unlabelled data. Additionally, we introduce innovative evaluation metrics, average accuracy and forward transfer, to effectively measure the model's adaptation capabilities in dynamic, real-world scenarios. Extensive experiments and ablation studies demonstrate the superiority of our approach in a range of tasks including image classification, text classification, and speech recognition.

Scaling up language models to billions of parameters has opened up possibilities for in-context learning, allowing instruction tuning and few-shot learning on tasks that the model was not specifically trained for. This has achieved breakthrough performance on language tasks such as translation, summarization, and question-answering. Furthermore, in addition to these associative "System 1" tasks, recent advances in Chain-of-thought prompt learning have demonstrated strong "System 2" reasoning abilities, answering a question in the field of artificial general intelligence whether LLMs can reason. The field started with the question whether LLMs can solve grade school math word problems. This paper reviews the rapidly expanding field of prompt-based reasoning with LLMs. Our taxonomy identifies different ways to generate, evaluate, and control multi-step reasoning. We provide an in-depth coverage of core approaches and open problems, and we propose a research agenda for the near future. Finally, we highlight the relation between reasoning and prompt-based learning, and we discuss the relation between reasoning, sequential decision processes, and reinforcement learning. We find that self-improvement, self-reflection, and some metacognitive abilities of the reasoning processes are possible through the judicious use of prompts. True self-improvement and self-reasoning, to go from reasoning with LLMs to reasoning by LLMs, remains future work.

We study the problem of online binary classification in settings where strategic agents can modify their observable features to receive a positive classification. We model the set of feasible manipulations by a directed graph over the feature space, and assume the learner only observes the manipulated features instead of the original ones. We introduce the Strategic Littlestone Dimension, a new combinatorial measure that captures the joint complexity of the hypothesis class and the manipulation graph. We demonstrate that it characterizes the instance-optimal mistake bounds for deterministic learning algorithms in the realizable setting. We also achieve improved regret in the agnostic setting by a refined agnostic-to-realizable reduction that accounts for the additional challenge of not observing agents' original features. Finally, we relax the assumption that the learner knows the manipulation graph, instead assuming their knowledge is captured by a family of graphs. We derive regret bounds in both the realizable setting where all agents manipulate according to the same graph within the graph family, and the agnostic setting where the manipulation graphs are chosen adversarially and not consistently modeled by a single graph in the family.

Large language models (LLMs) have demonstrated remarkable capabilities across numerous domains, as highlighted by OpenAI (2023) and Bubeck et al. (2023). However, whereas LLMs pre-trained on extensive language data excel in general language understanding, they may not be optimized for every specific task of interest prompted by instructions. Therefore, there is need for continual instruction learning to adapt LLMs to evolving tasks and domains. Indeed, continual instruction learning is essential for LLMs such as GPT (Radford et al., 2019) to maintain their effectiveness and relevance in handling a wide range of tasks and domains. Such models are trained on vast amounts of text data and fine-tuned for specific applications, often by learning tasks sequentially (Luo et al., 2023), i.e. learning on datasets pertaining to one task all at once, before moving on to the next task. The challenge lies in their ability to continually learn and adapt as they encounter new tasks and information.

A default assumption in the design of reinforcement-learning algorithms is that a decision-making agent always explores to learn optimal behavior. In sufficiently complex environments that approach the vastness and scale of the real world, however, attaining optimal performance may in fact be an entirely intractable endeavor and an agent may seldom find itself in a position to complete the requisite exploration for identifying an optimal policy. Recent work has leveraged tools from information theory to design agents that deliberately forgo optimal solutions in favor of sufficiently-satisfying or satisficing solutions, obtained through lossy compression. Notably, such agents may employ fundamentally different exploratory decisions to learn satisficing behaviors more efficiently than optimal ones that are more data intensive. While supported by a rigorous corroborating theory, the underlying algorithm relies on model-based planning, drastically limiting the compatibility of these ideas with function approximation and high-dimensional observations. In this work, we remedy this issue by extending an agent that directly represents uncertainty over the optimal value function allowing it to both bypass the need for model-based planning and to learn satisficing policies. We provide simple yet illustrative experiments that demonstrate how our algorithm enables deep reinforcement-learning agents to achieve satisficing behaviors. In keeping with previous work on this setting for multi-armed bandits, we additionally find that our algorithm is capable of synthesizing optimal behaviors, when feasible, more efficiently than its non-information-theoretic counterpart.

Long-form question answering (LFQA) aims to provide thorough and in-depth answers to complex questions, enhancing comprehension. However, such detailed responses are prone to hallucinations and factual inconsistencies, challenging their faithful evaluation. This work introduces HaluQuestQA, the first hallucination dataset with localized error annotations for human-written and model-generated LFQA answers. HaluQuestQA comprises 698 QA pairs with 4.7k span-level error annotations for five different error types by expert annotators, along with preference judgments. Using our collected data, we thoroughly analyze the shortcomings of long-form answers and find that they lack comprehensiveness and provide unhelpful references. We train an automatic feedback model on this dataset that predicts error spans with incomplete information and provides associated explanations. Finally, we propose a prompt-based approach, Error-informed refinement, that uses signals from the learned feedback model to refine generated answers, which we show reduces hallucination and improves answer quality. Furthermore, humans find answers generated by our approach comprehensive and highly prefer them (84%) over the baseline answers.

Online continual learning (OCL) involves deep neural networks retaining knowledge from old data while adapting to new data, which is accessible only once. A critical challenge in OCL is catastrophic forgetting, reflected in reduced model performance on old data. Existing replay-based methods mitigate forgetting by replaying buffered samples from old data and learning current samples of new data. In this work, we dissect existing methods and empirically discover that learning and replaying in the same feature space is not conducive to addressing the forgetting issue. Since the learned features associated with old data are readily changed by the features related to new data due to data imbalance, leading to the forgetting problem. Based on this observation, we intuitively explore learning and replaying in different feature spaces. Learning in a feature subspace is sufficient to capture novel knowledge from new data while replaying in a larger feature space provides more feature space to maintain historical knowledge from old data. To this end, we propose a novel OCL approach called experience replay with feature subspace learning (ER-FSL). Firstly, ER-FSL divides the entire feature space into multiple subspaces, with each subspace used to learn current samples. Moreover, it introduces a subspace reuse mechanism to address situations where no blank subspaces exist. Secondly, ER-FSL replays previous samples using an accumulated space comprising all learned subspaces. Extensive experiments on three datasets demonstrate the superiority of ER-FSL over various state-of-the-art methods.

Ancient manuscripts are frequently damaged, containing gaps in the text known as lacunae. In this paper, we present a bidirectional RNN model for character prediction of Coptic characters in manuscript lacunae. Our best model performs with 72% accuracy on single character reconstruction, but falls to 37% when reconstructing lacunae of various lengths. While not suitable for definitive manuscript reconstruction, we argue that our RNN model can help scholars rank the likelihood of textual reconstructions. As evidence, we use our RNN model to rank reconstructions in two early Coptic manuscripts. Our investigation shows that neural models can augment traditional methods of textual restoration, providing scholars with an additional tool to assess lacunae in Coptic manuscripts.