If an LLM Were a Character, Would It Know Its Own Story? Evaluating Lifelong Learning in LLMs

Large language models (LLMs) can carry out human-like dialogue, but unlike humans, they are stateless due to the superposition property. However, during multi-turn, multi-agent interactions, LLMs begin to exhibit consistent, character-like behaviors, hinting at a form of emergent lifelong learning. Despite this, existing benchmarks often fail to capture these dynamics, primarily focusing on static, open-ended evaluations. To address this gap, we introduce LIFESTATE-BENCH, a benchmark designed to assess lifelong learning in LLMs. It features two episodic datasets: Hamlet and a synthetic script collection, rich in narrative structure and character interactions. Our fact checking evaluation probes models' self-awareness, episodic memory retrieval, and relationship tracking, across both parametric and non-parametric approaches. Experiments on models like Llama3.1-8B, GPT-4-turbo, and DeepSeek R1, we demonstrate that nonparametric methods significantly outperform parametric ones in managing stateful learning. However, all models exhibit challenges with catastrophic forgetting as interactions extend, highlighting the need for further advancements in lifelong learning.

Are We Solving a Well-Defined Problem? A Task-Centric Perspective on Recommendation Tasks

Recommender systems (RecSys) leverage user interaction history to predict and suggest relevant items, shaping user experiences across various domains. While many studies adopt a general problem definition, i.e., to recommend preferred items to users based on past interactions, such abstraction often lacks the domain-specific nuances necessary for practical deployment. However, models are frequently evaluated using datasets from online recommender platforms, which inherently reflect these specificities. In this paper, we analyze RecSys task formulations, emphasizing key components such as input-output structures, temporal dynamics, and candidate item selection. All these factors directly impact offline evaluation. We further examine the complexities of user-item interactions, including decision-making costs, multi-step engagements, and unobservable interactions, which may influence model design and loss functions. Additionally, we explore the balance between task specificity and model generalizability, highlighting how well-defined task formulations serve as the foundation for robust evaluation and effective solution development. By clarifying task definitions and their implications, this work provides a structured perspective on RecSys research. The goal is to help researchers better navigate the field, particularly in understanding specificities of the RecSys tasks and ensuring fair and meaningful evaluations.

Knowledge Updating? No More Model Editing! Just Selective Contextual Reasoning

As real-world knowledge evolves, the information embedded within large language models (LLMs) can become outdated, inadequate, or erroneous. Model editing has emerged as a prominent approach for updating LLMs' knowledge with minimal computational costs and parameter changes. This approach typically identifies and adjusts specific model parameters associated with newly acquired knowledge. However, existing methods often underestimate the adverse effects that parameter modifications can have on broadly distributed knowledge. More critically, post-edit LLMs frequently struggle with multi-hop reasoning and continuous knowledge updates. Although various studies have discussed these shortcomings, there is a lack of comprehensive evaluation. In this paper, we provide an evaluation of ten model editing methods along four dimensions: reliability, generalization, locality, and portability. Results confirm that all ten popular model editing methods show significant shortcomings across multiple dimensions, suggesting model editing is less promising. We then propose a straightforward method called Selective Contextual Reasoning (SCR), for knowledge updating. SCR does not modify model parameters but harnesses LLM's inherent contextual reasoning capabilities utilizing the updated knowledge pieces. Under SCR, an LLM first assesses whether an incoming query falls within the scope of an external knowledge base. If it does, the relevant external knowledge texts are contextualized to enhance reasoning; otherwise, the query is answered directly. We evaluate SCR against the ten model editing methods on two counterfactual datasets with three backbone LLMs. Empirical results confirm the effectiveness and efficiency of contextual reasoning for knowledge updating.

A Flexible and Scalable Framework for Video Moment Search

Video moment search, the process of finding relevant moments in a video corpus to match a user's query, is crucial for various applications. Existing solutions, however, often assume a single perfect matching moment, struggle with inefficient inference, and have limitations with hour-long videos. This paper introduces a flexible and scalable framework for retrieving a ranked list of moments from collection of videos in any length to match a text query, a task termed Ranked Video Moment Retrieval (RVMR). Our framework, called Segment-Proposal-Ranking (SPR), simplifies the search process into three independent stages: segment retrieval, proposal generation, and moment refinement with re-ranking. Specifically, videos are divided into equal-length segments with precomputed embeddings indexed offline, allowing efficient retrieval regardless of video length. For scalable online retrieval, both segments and queries are projected into a shared feature space to enable approximate nearest neighbor (ANN) search. Retrieved segments are then merged into coarse-grained moment proposals. Then a refinement and re-ranking module is designed to reorder and adjust timestamps of the coarse-grained proposals. Evaluations on the TVR-Ranking dataset demonstrate that our framework achieves state-of-the-art performance with significant reductions in computational cost and processing time. The flexible design also allows for independent improvements to each stage, making SPR highly adaptable for large-scale applications.

Long Context vs. RAG for LLMs: An Evaluation and Revisits

Extending context windows (i.e., Long Context, LC) and using retrievers to selectively access relevant information (i.e., Retrieval-Augmented Generation, RAG) are the two main strategies to enable LLMs to incorporate extremely long external contexts. This paper revisits recent studies on this topic, highlighting their key insights and discrepancies. We then provide a more comprehensive evaluation by filtering out questions answerable without external context, identifying the most effective retrieval methods, and expanding the datasets. We show that LC generally outperforms RAG in question-answering benchmarks, especially for Wikipedia-based questions. Summarization-based retrieval performs comparably to LC, while chunk-based retrieval lags behind. However, RAG has advantages in dialogue-based and general question queries. These insights underscore the trade-offs between RAG and LC strategies, offering guidance for future optimization of LLMs with external knowledge sources. We also provide an in-depth discussion on this topic, highlighting the overlooked importance of context relevance in existing studies.

Hesitation and Tolerance in Recommender Systems

User interactions in recommender systems are inherently complex, often involving behaviors that go beyond simple acceptance or rejection. One particularly common behavior is hesitation, where users deliberate over recommended items, signaling uncertainty. Our large-scale surveys, with 6,644 and 3,864 responses respectively, confirm that hesitation is not only widespread but also has a profound impact on user experiences. When users spend additional time engaging with content they are ultimately uninterested in, this can lead to negative emotions, a phenomenon we term as tolerance. The surveys reveal that such tolerance behaviors often arise after hesitation and can erode trust, satisfaction, and long-term loyalty to the platform. For instance, a click might reflect a need for more information rather than genuine interest, and prolonged exposure to unsuitable content amplifies frustration. This misalignment between user intent and system interpretation introduces noise into recommendation training, resulting in suggestions that increase uncertainty and disengagement. To address these issues, we identified signals indicative of tolerance behavior and analyzed datasets from both e-commerce and short-video platforms. The analysis shows a strong correlation between increased tolerance behavior and decreased user activity. We integrated these insights into the training process of a recommender system for a major short-video platform. Results from four independent online A/B experiments demonstrated significant improvements in user retention, achieved with minimal additional computational costs. These findings underscore the importance of recognizing hesitation as a ubiquitous user behavior and addressing tolerance to enhance satisfaction, build trust, and sustain long-term engagement in recommender systems.

Knowledge-Enhanced Conversational Recommendation via Transformer-based Sequential Modelling

In conversational recommender systems (CRSs), conversations usually involve a set of items and item-related entities or attributes, e.g., director is a related entity of a movie. These items and item-related entities are often mentioned along the development of a dialog, leading to potential sequential dependencies among them. However, most of existing CRSs neglect these potential sequential dependencies. In this article, we first propose a Transformer-based sequential conversational recommendation method, named TSCR, to model the sequential dependencies in the conversations to improve CRS. In TSCR, we represent conversations by items and the item-related entities, and construct user sequences to discover user preferences by considering both the mentioned items and item-related entities. Based on the constructed sequences, we deploy a Cloze task to predict the recommended items along a sequence. Meanwhile, in certain domains, knowledge graphs formed by the items and their related entities are readily available, which provide various different kinds of associations among them. Given that TSCR does not benefit from such knowledge graphs, we then propose a knowledge graph enhanced version of TSCR, called TSCRKG. In specific, we leverage the knowledge graph to offline initialize our model TSCRKG, and augment the user sequence of conversations (i.e., sequence of the mentioned items and item-related entities in the conversation) with multi-hop paths in the knowledge graph. Experimental results demonstrate that our TSCR model significantly outperforms state-of-the-art baselines, and the enhanced version TSCRKG further improves recommendation performance on top of TSCR.

Sketch: A Toolkit for Streamlining LLM Operations

Large language models (LLMs) represented by GPT family have achieved remarkable success. The characteristics of LLMs lie in their ability to accommodate a wide range of tasks through a generative approach. However, the flexibility of their output format poses challenges in controlling and harnessing the model's outputs, thereby constraining the application of LLMs in various domains. In this work, we present Sketch, an innovative toolkit designed to streamline LLM operations across diverse fields. Sketch comprises the following components: (1) a suite of task description schemas and prompt templates encompassing various NLP tasks; (2) a user-friendly, interactive process for building structured output LLM services tailored to various NLP tasks; (3) an open-source dataset for output format control, along with tools for dataset construction; and (4) an open-source model based on LLaMA3-8B-Instruct that adeptly comprehends and adheres to output formatting instructions. We anticipate this initiative to bring considerable convenience to LLM users, achieving the goal of ''plug-and-play'' for various applications. The components of Sketch will be progressively open-sourced at https://github.com/cofe-ai/Sketch.

Not All Videos Become Outdated: Short-Video Recommendation by Learning to Deconfound Release Interval Bias

Short-video recommender systems often exhibit a biased preference to recently released videos. However, not all videos become outdated; certain classic videos can still attract user's attention. Such bias along temporal dimension can be further aggravated by the matching model between users and videos, because the model learns from preexisting interactions. From real data, we observe that different videos have varying sensitivities to recency in attracting users' attention. Our analysis, based on a causal graph modeling short-video recommendation, suggests that the release interval serves as a confounder, establishing a backdoor path between users and videos. To address this confounding effect, we propose a model-agnostic causal architecture called Learning to Deconfound the Release Interval Bias (LDRI). LDRI enables jointly learning of the matching model and the video recency sensitivity perceptron. In the inference stage, we apply a backdoor adjustment, effectively blocking the backdoor path by intervening on each video. Extensive experiments on two benchmarks demonstrate that LDRI consistently outperforms backbone models and exhibits superior performance against state-of-the-art models. Additional comprehensive analyses confirm the deconfounding capability of LDRI.

Open-domain Implicit Format Control for Large Language Model Generation

Controlling the format of outputs generated by large language models (LLMs) is a critical functionality in various applications. Current methods typically employ constrained decoding with rule-based automata or fine-tuning with manually crafted format instructions, both of which struggle with open-domain format requirements. To address this limitation, we introduce a novel framework for controlled generation in LLMs, leveraging user-provided, one-shot QA pairs. This study investigates LLMs' capabilities to follow open-domain, one-shot constraints and replicate the format of the example answers. We observe that this is a non-trivial problem for current LLMs. We also develop a dataset collection methodology for supervised fine-tuning that enhances the open-domain format control of LLMs without degrading output quality, as well as a benchmark on which we evaluate both the helpfulness and format correctness of LLM outputs. The resulting datasets, named OIFC-SFT, along with the related code, will be made publicly available at https://github.com/cofe-ai/OIFC.