In 2025, Large Language Model (LLM) services have launched a new feature -- AI video chat -- allowing users to interact with AI agents via real-time video communication (RTC), just like chatting with real people. Despite its significance, no systematic study has characterized the performance of existing AI video chat systems. To address this gap, this paper proposes a comprehensive benchmark with carefully designed metrics across four dimensions: quality, latency, internal mechanisms, and system overhead. Using custom testbeds, we further evaluate five mainstream AI video chatbots with this benchmark. This work provides the research community a baseline of real-world performance and identifies unique system bottlenecks. In the meantime, our benchmarking results also open up several research questions for future optimizations of AI video chatbots.

Code generation under long contexts is becoming increasingly critical as Large Language Models (LLMs) are required to reason over extensive information in the codebase. While recent advances enable code LLMs to process long inputs, high API costs and generation latency remain substantial bottlenecks. Existing context pruning techniques, such as LLMLingua, achieve promising results for general text but overlook code-specific structures and dependencies, leading to suboptimal performance in programming tasks. In this paper, we propose LongCodeZip, a novel plug-and-play code compression framework designed specifically for code LLMs. LongCodeZip employs a dual-stage strategy: (1) coarse-grained compression, which identifies and ranks function-level chunks using conditional perplexity with respect to the instruction, retaining only the most relevant functions; and (2) fine-grained compression, which segments retained functions into blocks based on perplexity and selects an optimal subset under an adaptive token budget to maximize relevance. Evaluations across multiple tasks, including code completion, summarization, and question answering, show that LongCodeZip consistently outperforms baseline methods, achieving up to a 5.6x compression ratio without degrading task performance. By effectively reducing context size while preserving essential information, LongCodeZip enables LLMs to better scale to real-world, large-scale code scenarios, advancing the efficiency and capability of code intelligence applications.

Automated structured radiology report generation (SRRG) from chest X-ray images offers significant potential to reduce workload of radiologists by generating reports in structured formats that ensure clarity, consistency, and adherence to clinical reporting standards. While radiologists effectively utilize available clinical contexts in their diagnostic reasoning, existing SRRG systems overlook these essential elements. This fundamental gap leads to critical problems including temporal hallucinations when referencing non-existent clinical contexts. To address these limitations, we propose contextualized SRRG (C-SRRG) that comprehensively incorporates rich clinical context for SRRG. We curate C-SRRG dataset by integrating comprehensive clinical context encompassing 1) multi-view X-ray images, 2) clinical indication, 3) imaging techniques, and 4) prior studies with corresponding comparisons based on patient histories. Through extensive benchmarking with state-of-the-art multimodal large language models, we demonstrate that incorporating clinical context with the proposed C-SRRG significantly improves report generation quality. We publicly release dataset, code, and checkpoints to facilitate future research for clinically-aligned automated RRG at https://github.com/vuno/contextualized-srrg.

The Mamba model has gained significant attention for its computational advantages over Transformer-based models, while achieving comparable performance across a wide range of language tasks. Like Transformers, Mamba exhibits in-context learning (ICL) capabilities, i.e., making predictions for new tasks based on a prompt containing input-label pairs and a query, without requiring fine-tuning. Despite its empirical success, the theoretical understanding of Mamba remains limited, largely due to the nonlinearity introduced by its gating mechanism. To the best of our knowledge, this paper presents the first theoretical analysis of the training dynamics of a one-layer Mamba model, which consists of a linear attention component followed by a nonlinear gating layer, and its ICL generalization on unseen binary classification tasks, even when the prompt includes additive outliers. Our analysis shows that Mamba leverages the linear attention layer to select informative context examples and uses the nonlinear gating layer to suppress the influence of outliers. By establishing and comparing to the analysis of linear Transformers under the same setting, we show that although Mamba may require more training iterations to converge, it maintains accurate predictions even when the proportion of outliers exceeds the threshold that a linear Transformer can tolerate. These theoretical findings are supported by empirical experiments.

Epilepsy is a chronic neurological condition characterized by recurrent seizures, with global prevalence estimated at 50 million people worldwide. While progress in high-throughput sequencing has allowed for broad-based transcriptomic profiling of brain tissues, the deciphering of these highly complex datasets remains one of the challenges. To address this issue, in this paper we propose a new analysis pipeline that integrates the power of deep learning strategies with GPU-acceleration computation for investigating Gene expression patterns in epilepsy. Specifically, our proposed approach employs GPT-2 XL, a transformer-based Large Language Model (LLM) with 1.5 billion parameters for genomic sequence analysis over the latest NVIDIA H100 Tensor Core GPUs based on Hopper architecture. Our proposed method enables efficient preprocessing of RNA sequence data, gene sequence encoding, and subsequent pattern identification. We conducted experiments on two epilepsy datasets including GEO accession GSE264537 and GSE275235. The obtained results reveal several significant transcriptomic modifications, including reduced hippocampal astrogliosis after ketogenic diet treatment as well as restored excitatory-inhibitory signaling equilibrium in zebrafish epilepsy model. Moreover, our results highlight the effectiveness of leveraging LLMs in combination with advanced hardware acceleration for transcriptomic characterization in neurological diseases.

AI question answering systems increasingly generate responses with attributions to sources. However, the task of verifying the actual content of these attributions is in most cases impractical. In this paper, we present attribution gradients as a solution. Attribution gradients provide integrated, incremental affordances for diving into an attributed passage. A user can decompose a sentence of an answer into its claims. For each claim, the user can view supporting and contradictory excerpts mined from sources. Those excerpts serve as clickable conduits into the source (in our application, scientific papers). When evidence itself contains more citations, the UI unpacks the evidence into excerpts from the cited sources. These features of attribution gradients facilitate concurrent interconnections among answer, claim, excerpt, and context. In a usability study, we observed greater engagement with sources and richer revision in a task where participants revised an attributed AI answer with attribution gradients and a baseline.

As large language models (LLMs) continue to grow in capability, there is increasing interest in incorporating them into decision-making tasks. A common pipeline for this is Decision-Pretrained Transformers (DPTs). However, existing training methods for DPTs often struggle to generalize beyond their pretraining data distribution. To explore mitigation of this limitation, we propose in-context curiosity -- a lightweight, exploration-inspired regularizer for offline pretraining -- and introduce the Prediction-Powered Transformer (PPT) framework. PPT augments DPT with an auxiliary reward predictor, using prediction error as an intrinsic curiosity signal to encourage broader exploration during training. In proof-of-concept experiments on Gaussian multi-armed bandits, PPT shows improved robustness: it moderates the performance degradation observed in DPT when test environments exhibit higher variance in reward, particularly when pretraining data has limited diversity. While the quality of offline data remain fundamental, our preliminary results suggest that curiosity-driven pretraining offers a promising direction for enhancing out-of-distribution generalization in in-context RL agents.

Adaptive chatbots that mimic a user's linguistic style can build rapport and engagement, yet unconstrained mimicry risks an agent that feels unstable or sycophantic. We present a computational evaluation framework that makes the core design tension explicit: balancing moment-to-moment linguistic synchrony against long-term persona stability. Using an 8-dimensional style vector and a closed-loop "base+delta" prompting architecture, we simulate and compare explicit adaptation policies - Uncapped, Cap, Exponential Moving Average (EMA), Dead-Band, and Hybrids - on a human-log dataset. Our analysis maps a clear Pareto frontier: bounded policies achieve substantial gains in stability at a modest cost to synchrony. For example, a Hybrid (EMA+Cap) raises stability from 0.542 to 0.878 (+62%) while reducing synchrony by only 17%. We confirm this trade-off through large-scale replications on three public corpora (DailyDialog, Persona-Chat, EmpatheticDialogues) and LLM-in-the-loop validation across two model families. Furthermore, we quantify "prompt legibility," showing that frontier policies reduce instruction churn and cut jarring register flips (major tone changes) from 0.254 to 0.092, yielding systems that are easier to reason about and maintain. Taken together, our framework provides a general evaluation harness for style adaptation; a systematic ablation that identifies Pareto-efficient policies; robust validation across diverse datasets and models; and novel legibility metrics linking policy choices to system maintainability.

Code search is an important information retrieval application. Benefits of better code search include faster new developer on-boarding, reduced software maintenance, and ease of understanding for large repositories. Despite improvements in search algorithms and search benchmarks, the domain of code search has lagged behind. One reason is the high cost of human annotation for code queries and answers. While humans may annotate search results in general text QA systems, code annotations require specialized knowledge of a programming language (PL), as well as domain specific software engineering knowledge. In this work we study the use of Large Language Models (LLMs) to retrieve code at the level of functions and to generate annotations for code search results. We compare the impact of the retriever representation (sparse vs. semantic), programming language, and LLM by comparing human annotations across several popular languages (C, Java, Javascript, Go, and Python). We focus on repositories that implement common data structures likely to be implemented in any PLs. For the same human annotations, we compare several LLM-as-a-Judge models to evaluate programming language and other affinities between LLMs. We find that the chosen retriever and PL exhibit affinities that can be leveraged to improve alignment of human and AI relevance determinations, with significant performance implications. We also find differences in representation (sparse vs. semantic) across PLs that impact alignment of human and AI relevance determinations. We propose using transpilers to bootstrap scalable code search benchmark datasets in other PLs and in a case study demonstrate that human-AI relevance agreement rates largely match the (worst case) human-human agreement under study. The application code used in this work is available at \href{https://github.com/rlucas7/code-searcher/}{this github repo}.

Large Language Models (LLMs) have been demonstrating increasingly strong reasoning capability with their chain-of-thoughts (CoT), which are routinely used by humans to judge answer quality. This reliance creates a powerful yet fragile basis for trust. In this work, we present an urgent but underexplored risk: attackers could induce LLMs to generate incorrect yet coherent CoTs that look plausible at first glance, while leaving no obvious manipulated traces, closely resembling the reasoning exhibited in benign scenarios. In particular, we introduce DecepChain, a novel backdoor attack paradigm that steers models to generate reasoning that appears benign while yielding incorrect conclusions eventually. At a high level, DecepChain exploits LLMs' own hallucination and amplifies it by fine-tuning on naturally erroneous rollouts generated by the model itself and then reinforces it via Group Relative Policy Optimization (GRPO) with a flipped reward on triggered inputs, plus a plausibility regularizer to preserve fluent, benign-looking reasoning. Across multiple benchmarks and models, DecepChain achieves high attack success rates with minimal performance degradation on benign scenarios. Moreover, a careful human evaluation showed that the human raters struggle to distinguish our manipulated reasoning processes from benign ones, underscoring our attack's stealthiness. Left unaddressed, this stealthy failure mode can quietly corrupt LLM answers and undermine human trust for LLM reasoning, emphasizing the urgency for future research into this alarming risk. Project page: https://decepchain.github.io/.