Lifelong knowledge editing enables continuous, precise updates to outdated knowledge in large language models (LLMs) without computationally expensive full retraining. However, existing methods often accumulate errors throughout the editing process, causing a gradual decline in both editing accuracy and generalization. To tackle this problem, we propose Neuron-Specific Masked Knowledge Editing (NMKE), a novel fine-grained editing framework that combines neuron-level attribution with dynamic sparse masking. Leveraging neuron functional attribution, we identify two key types of knowledge neurons, with knowledge-general neurons activating consistently across prompts and knowledge-specific neurons activating to specific prompts. NMKE further introduces an entropy-guided dynamic sparse mask, locating relevant neurons to the target knowledge. This strategy enables precise neuron-level knowledge editing with fewer parameter modifications. Experimental results from thousands of sequential edits demonstrate that NMKE outperforms existing methods in maintaining high editing success rates and preserving model general capabilities in lifelong editing.

Pretrained vision-language models such as CLIP achieve strong zero-shot generalization but remain vulnerable to distribution shifts caused by input corruptions. In this work, we investigate how corruptions affect CLIP's image embeddings and uncover a consistent phenomenon we term as embedding variance collapse, where both intra-class and inter-class variances shrink as corruption severity increases. We find that this collapse is closely tied to performance degradation, with inter-class variance strongly correlated with classification accuracy. To explain this phenomenon, we analyze how corruptions alter the structure of the embedding space. Our theoretical results suggest that the visual encoder tends to encode corruption-related signals, which dilute class-discriminative features and compress the representation geometry. We further show that maximizing inter-class variance, even when estimated from pseudo-labels, can provably enhance embedding quality. Based on this insight, we propose Mint, a simple test-time adaptation method that maximizes pseudo-label-based inter-class variance on the fly using a mean accumulator and a gradient accumulator. Mint operates effectively with small batch sizes and consistently improves performance across multiple corruption benchmarks and CLIP architectures. Our code is available at https://github.com/baowenxuan/Mint .

Most approaches to long-context processing increase the complexity of the transformer's internal architecture by integrating mechanisms such as recurrence or auxiliary memory modules. In this work, we introduce an alternative approach that modifies the input representation itself, rather than the transformer architecture. Inspired by cognitive models of human memory, our method applies a scale-invariant logarithmic compression to the input tokens. The resulting compressed representation is processed by a standard, unmodified transformer, preserving architectural simplicity. We evaluate this approach on the WikiText-103 and PG-19 language modeling benchmarks, showing a reduction in perplexity compared to uncompressed baselines. Moreover, performance improves consistently with longer compressed temporal contexts, showing that input-level logarithmic compression is a simple and effective way to extend a transformer's long-range memory.

Multimodal large language models (MLLMs) excel at vision-language tasks such as VQA and document understanding, yet precise coordinate prediction remains challenging. High-resolution inputs exacerbate this difficulty by producing long token sequences that weaken positional encodings and introduce directional biases in coordinate outputs. We investigate this phenomenon by analyzing how MLLMs behave when visual positional encodings (VPEs) are deliberately perturbed through shuffling. Our analysis reveals that such perturbations induce predictable, non-random coordinate biases rather than random errors, suggesting that models rely on internal positional priors when spatial grounding signals are degraded. Crucially, we observe similar directional error patterns in natural high-resolution datasets, indicating that positional encoding failures are a key bottleneck for accurate coordinate prediction at scale. To address this issue, we propose Vision-PE Shuffle Guidance (VPSG), a training-free test-time method that leverages the directional nature of these biases for correction. VPSG runs auxiliary decoding with shuffled VPEs to isolate position-unconditioned tendencies, then uses this as negative evidence to guide digit prediction while preserving coordinate format through a lightweight finite-state machine. Experiments on ScreenSpot-Pro demonstrate reliable improvements, highlighting positional encoding robustness as a critical factor for spatial reasoning in MLLMs.

Large Language Models (LLMs) have demonstrated impressive quality when applied to predictive tasks such as relevance ranking and semantic search. However, deployment of such LLMs remains prohibitively expensive for industry applications with strict latency and throughput requirements. In this work, we present lessons and efficiency insights from developing a purely text-based decoder-only Small Language Model (SLM) for a semantic search application at LinkedIn. Particularly, we discuss model compression techniques such as pruning that allow us to reduce the model size by up to $40\%$ while maintaining the accuracy. Additionally, we present context compression techniques that allow us to reduce the input context length by up to $10$x with minimal loss of accuracy. Finally, we present practical lessons from optimizing the serving infrastructure for deploying such a system on GPUs at scale, serving millions of requests per second. Taken together, this allows us to increase our system's throughput by $10$x in a real-world deployment, while meeting our quality bar.

Large Language Models (LLMs) exhibit a troubling duality, capable of both remarkable generalization and brittle, verbatim memorization of their training data. This unpredictability undermines their reliability in high-stakes applications. In this work, we propose a unified framework to understand, identify, and control these distinct reasoning modes. First, we introduce a theoretical model based on the Information Bottleneck (IB) principle, formalizing generalization as the learning of a compressed, task-relevant representation and memorization as a failure to compress. Building on this theory, we develop Dynamic Mode Steering (DMS), a novel inference-time algorithm which comprises two components: (1) a lightweight, causally-grounded linear probe that identifies the model's instantaneous reliance on memorization, and (2) a dynamic activation steering mechanism that nudges the model's computation towards pre-identified generalization circuits. We frame DMS as a form of adaptive, self-contrastive decoding. Experiments on reasoning and faithfulness tasks demonstrate that DMS significantly improves logical consistency and factual accuracy, thereby offering a principled approach to enhancing LLM reliability.

Brain-Computer Interfaces (BCIs) offer a direct communication pathway between the human brain and external devices, holding significant promise for individuals with severe neurological impairments. However, their widespread adoption is hindered by critical limitations, such as low information transfer rates and extensive user-specific calibration. To overcome these challenges, recent research has explored the integration of Large Language Models (LLMs), extending the focus from simple command decoding to understanding complex cognitive states. Despite these advancements, deploying agentic AI faces technical hurdles and ethical concerns. Due to the lack of comprehensive discussion on this emerging direction, this position paper argues that the field is poised for a paradigm extension from BCI to Brain-Agent Collaboration (BAC). We emphasize reframing agents as active and collaborative partners for intelligent assistance rather than passive brain signal data processors, demanding a focus on ethical data handling, model reliability, and a robust human-agent collaboration framework to ensure these systems are safe, trustworthy, and effective.

The field of computer architecture, which bridges high-level software abstractions and low-level hardware implementations, remains absent from current large language model (LLM) evaluations. To this end, we present QuArch (pronounced 'quark'), the first benchmark designed to facilitate the development and evaluation of LLM knowledge and reasoning capabilities specifically in computer architecture. QuArch provides a comprehensive collection of 2,671 expert-validated question-answer (QA) pairs covering various aspects of computer architecture, including processor design, memory systems, and interconnection networks. Our evaluation reveals that while frontier models possess domain-specific knowledge, they struggle with skills that require higher-order thinking in computer architecture. Frontier model accuracies vary widely (from 34% to 72%) on these advanced questions, highlighting persistent gaps in architectural reasoning across analysis, design, and implementation QAs. By holistically assessing fundamental skills, QuArch provides a foundation for building and measuring LLM capabilities that can accelerate innovation in computing systems. With over 140 contributors from 40 institutions, this benchmark represents a community effort to set the standard for architectural reasoning in LLM evaluation.

Large language models (LLMs) remain vulnerable to sophisticated prompt engineering attacks that exploit contextual framing to bypass safety mechanisms, posing significant risks in cybersecurity applications. We introduce Jailbreak Mimicry, a systematic methodology for training compact attacker models to automatically generate narrative-based jailbreak prompts in a one-shot manner. Our approach transforms adversarial prompt discovery from manual craftsmanship into a reproducible scientific process, enabling proactive vulnerability assessment in AI-driven security systems. Developed for the OpenAI GPT-OSS-20B Red-Teaming Challenge, we use parameter-efficient fine-tuning (LoRA) on Mistral-7B with a curated dataset derived from AdvBench, achieving an 81.0% Attack Success Rate (ASR) against GPT-OSS-20B on a held-out test set of 200 items. Cross-model evaluation reveals significant variation in vulnerability patterns: our attacks achieve 66.5% ASR against GPT-4, 79.5% on Llama-3 and 33.0% against Gemini 2.5 Flash, demonstrating both broad applicability and model-specific defensive strengths in cybersecurity contexts. This represents a 54x improvement over direct prompting (1.5% ASR) and demonstrates systematic vulnerabilities in current safety alignment approaches. Our analysis reveals that technical domains (Cybersecurity: 93% ASR) and deception-based attacks (Fraud: 87.8% ASR) are particularly vulnerable, highlighting threats to AI-integrated threat detection, malware analysis, and secure systems, while physical harm categories show greater resistance (55.6% ASR). We employ automated harmfulness evaluation using Claude Sonnet 4, cross-validated with human expert assessment, ensuring reliable and scalable evaluation for cybersecurity red-teaming. Finally, we analyze failure mechanisms and discuss defensive strategies to mitigate these vulnerabilities in AI for cybersecurity.

Large Language Models (LLMs) still produce gender-stereotyped language even in occupation-neutral contexts that reflect deep societal biases (Rudinger et al., 2018). To address this, prior work has proposed prompting, constrained decoding (Dathathri et al., 2020; Zhou et al., 2024), post-processing, and fine-tuning-based alignment (Rafailov et al., 2023; Ravfogel et al., 2022). However, the comparative efficacy and learning dynamics remain little understood. We report a comparative analysis of six control techniques for bias mitigation: prompt-only, generate-and-filter, DFA-based Ctrl-G decoding, Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Iterative Nullspace Projection (INLP). We evaluate each method on a compositional constraint task. This task requires generating sentences that contain at least one agentic and one communal descriptor for each of the twenty Winogender-derived occupations. We quantify trade-offs between control strength and naturalness with evaluations of constraint compliance, lexical diversity, and fluency. Our results reveal key contrasts among the methods: SFT achieves 99.87 +- 0.15% compliance and high lexical diversity, while DPO, despite similar training stability, fails at 4.53 +- 0.82%. Ctrl-G guarantees perfect compliance, but at the cost of severely reduced fluency and diversity. Preference-based learning fundamentally differs: it cannot satisfy compositional constraints, as binary preference signals encode ranking, not logical conjunctions. Only explicit positive supervision enables mitigation of compositional biases; preference-based alignment fails to generalize logical structures, underscoring the limitations of preference learning and the necessity of explicit supervision for fair and fluent controlled generation.