Developing autonomous agents that quickly explore an environment and adapt their behavior online is a canonical challenge in robotics and machine learning. While humans are able to achieve such fast online exploration and adaptation, often acquiring new information and skills in only a handful of interactions, existing algorithmic approaches tend to rely on random exploration and slow, gradient-based behavior updates. How can we endow autonomous agents with such capabilities on par with humans? Taking inspiration from recent progress on both in-context learning and large-scale behavioral cloning, in this work we propose behavioral exploration: training agents to internalize what it means to explore and adapt in-context over the space of ``expert'' behaviors. To achieve this, given access to a dataset of expert demonstrations, we train a long-context generative model to predict expert actions conditioned on a context of past observations and a measure of how ``exploratory'' the expert's behaviors are relative to this context. This enables the model to not only mimic the behavior of an expert, but also, by feeding its past history of interactions into its context, to select different expert behaviors than what have been previously selected, thereby allowing for fast online adaptation and targeted, ``expert-like'' exploration. We demonstrate the effectiveness of our method in both simulated locomotion and manipulation settings, as well as on real-world robotic manipulation tasks, illustrating its ability to learn adaptive, exploratory behavior.

Large language models (LLMs) are increasingly being used as decision aids. However, users have diverse values and preferences that can affect their decision-making, which requires novel methods for LLM alignment and personalization. Existing LLM comparison tools largely focus on benchmarking tasks, such as knowledge-based question answering. In contrast, our proposed ALIGN system focuses on dynamic personalization of LLM-based decision-makers through prompt-based alignment to a set of fine-grained attributes. Key features of our system include robust configuration management, structured output generation with reasoning, and several algorithm implementations with swappable LLM backbones, enabling different types of analyses. Our user interface enables a qualitative, side-by-side comparison of LLMs and their alignment to various attributes, with a modular backend for easy algorithm integration. Additionally, we perform a quantitative analysis comparing alignment approaches in two different domains: demographic alignment for public opinion surveys and value alignment for medical triage decision-making. The entire ALIGN framework is open source and will enable new research on reliable, responsible, and personalized LLM-based decision-makers.

The rapid evolution of Large Language Model (LLM) inference systems has yielded significant efficiency improvements. However, our systematic analysis reveals that current evaluation methodologies frequently exhibit fundamental flaws, often manifesting as common evaluation anti-patterns that obscure true performance characteristics and impede scientific progress. Through a comprehensive examination of recent systems, we identify recurring anti-patterns across three key dimensions: Baseline Fairness, Evaluation Setup, and Metric Design. These anti-patterns are uniquely problematic for LLM inference due to its dual-phase nature combining distinct prefill and decode operations, its handling of highly heterogeneous workloads, and its strict temporal requirements for interactive use. We demonstrate how common anti-patterns -- such as inadequate baseline comparisons that conflate engineering effort with algorithmic novelty, workload selections that fail to represent production scenarios, and metric normalizations that hide substantial performance variability like generation stalls-lead to misleading conclusions. To address these challenges, we provide a comprehensive checklist derived from our analysis, establishing a framework for recognizing and avoiding these anti-patterns in favor of robust LLM inference evaluation. To demonstrate the practical application of our framework, we present a case study analyzing speculative decoding, a technique whose bursty, non-uniform token generation is easily misinterpreted when evaluated using approaches characteristic of these anti-patterns. Our work establishes a rigorous foundation for evaluation methodology, enabling meaningful comparisons, ensuring reproducible results, and ultimately accelerating genuine progress in LLM inference systems by moving beyond common anti-patterns to align evaluation with real-world requirements.

A rapidly alternating red and black display known as Ganzflicker induces visual hallucinations that reflect the generative capacity of the visual system. Recent proposals regarding the imagery spectrum, that is, differences in the visual system of individuals with absent imagery, typical imagery, and vivid imagery, suggest these differences should impact the complexity of other internally generated visual experiences. Here, we used tools from natural language processing to analyze free-text descriptions of hallucinations from over 4,000 participants, asking whether people with different imagery phenotypes see different things in their mind's eye during Ganzflicker-induced hallucinations. Strong imagers described complex, naturalistic content, while weak imagers reported simple geometric patterns. Embeddings from vision language models better captured these differences than text-only language models, and participants with stronger imagery used language with richer sensorimotor associations. These findings may reflect individual variation in coordination between early visual areas and higher-order regions relevant for the imagery spectrum.

Edge inference for large language models (LLM) offers secure, low-latency, and cost-effective inference solutions. We emphasize that an edge accelerator should achieve high area efficiency and minimize external memory access (EMA) during the memory-bound decode stage, while maintaining high energy efficiency during the compute intensive prefill stage. This paper proposes an edge LLM inference accelerator featuring a hybrid systolic array (HSA) architecture that optimizes inference efficiency in both stages. To further reduce EMA, we adopt MXINT4 weight quantization and propose an optimized dataflow tailored for HSA, ensuring negligible dequantization overhead and achieving 100% hardware utilization with minimal accuracy loss under edge DRAM bandwidth constraints. For non-linear operations, we incorporate optimized root mean square normalization (RMSNorm) and rotary position embedding (RoPE) units, reducing their latency, area, and memory access overhead while enabling end-to-end inference on our accelerator. Our solution achieves 247/117 (token/s/mm2) while running a 1.3B LLM on long-input/long-output scenarios, providing >2.45x/13.5x improvement over existing approaches, while maintaining superior energy efficiency in token generation.

Source code segmentation, dividing code into functionally coherent segments, is crucial for knowledge retrieval and maintenance in software development. While enabling efficient navigation and comprehension of large codebases, manual and syntactic analysis approaches have become impractical as repositories grow, especially for low-resource languages like R and their research domains (e.g., social sciences, psychology).This paper introduces an automated, domain-specific approach for research R code segmentation using Large and Small Language Models (LLMs/SLMs). It presents two novel approaches and a human-annotated dataset, StatCodeSeg. We explore two distinct approaches: line-by-line analysis with context and range-based segment determination. We experiment with LLMs and fine-tuned SLMs. To support the generalizability of our approaches, we also include experiments on Python code from the computer science domain.Our results show that context-based line-by-line analysis is superior over range-based segmentation.Using smaller language models like CodeBERT and an encoder-only version of CodeT5+ are better than their LLM counterparts. Most notably, these two best-performing models did not see R code during pre-training versus the LLMs but were only fine-tuned on 4,130 lines of manually annotated code.

While poisoning attacks on machine learning models have been extensively studied, the mechanisms by which adversaries can distribute poisoned models at scale remain largely unexplored. In this paper, we shed light on how model leaderboards -- ranked platforms for model discovery and evaluation -- can serve as a powerful channel for adversaries for stealthy large-scale distribution of poisoned models. We present TrojanClimb, a general framework that enables injection of malicious behaviors while maintaining competitive leaderboard performance. We demonstrate its effectiveness across four diverse modalities: text-embedding, text-generation, text-to-speech and text-to-image, showing that adversaries can successfully achieve high leaderboard rankings while embedding arbitrary harmful functionalities, from backdoors to bias injection. Our findings reveal a significant vulnerability in the machine learning ecosystem, highlighting the urgent need to redesign leaderboard evaluation mechanisms to detect and filter malicious (e.g., poisoned) models, while exposing broader security implications for the machine learning community regarding the risks of adopting models from unverified sources.

--Recent years have witnessed remarkable progress in developing Vision-Language Models (VLMs) capable of processing both textual and visual inputs. These models have demonstrated impressive performance, leading to their widespread adoption in various applications. However, this widespread raises serious concerns regarding user privacy, particularly when models inadvertently process or expose private visual information. We propose a novel attack strategy that selectively conceals information within designated Region Of Interests (ROIs) in an image, effectively preventing VLMs from accessing sensitive content while preserving the semantic integrity of the remaining image. Unlike conventional adversarial attacks that often disrupt the entire image, our method maintains high coherence in unmasked areas. Experimental results across three state-of-the-art VLMs namely LLaV A, Instruct-BLIP, and BLIP2-T5 demonstrate up to 98% reduction in detecting targeted ROIs, while maintaining global image semantics intact, as confirmed by high similarity scores between clean and adversarial outputs. We believe that this work contributes to a more privacy-conscious use of multimodal models and offers a practical tool for further research, with the source code publicly available at https://github.com/hbrachemi/Vlm Vision-Language Models (VLMs) have emerged as a powerful paradigm in artificial intelligence, seamlessly integrating visual and textual information to achieve remarkable performance in various tasks such as image captioning, visual question answering, and document understanding [1]-[5]. This led to their rapid adoption in numerous applications, including content creation, customer service, and information retrieval. However, the widespread use of VLMs raises critical concerns regarding the privacy and security of user data.

W e introduce Projection-based Reduction of Implicit Spurious bias in vision-language Models (PRISM), a new data-free and task-agnostic solution for bias mitigation in VLMs like CLIP . VLMs often inherit and amplify biases in their training data, leading to skewed predictions. PRISM is designed to debias VLMs without relying on predefined bias categories or additional external data. It operates in two stages: first, an LLM is prompted with simple class prompts to generate scene descriptions that contain spurious correlations. Next, PRISM uses our novel contrastive-style debi-asing loss to learn a projection that maps the embeddings onto a latent space that minimizes spurious correlations while preserving the alignment between image and text em-beddings. Extensive experiments demonstrate that PRISM outperforms current debiasing methods on the commonly used W aterbirds and CelebA datasets W e make our code public at: https://github.com/MahdiyarMM/

Model steering represents a powerful technique that dynamically aligns large language models (LLMs) with human preferences during inference. However, conventional model-steering methods rely heavily on externally annotated data, not only limiting their adaptability to varying contexts but also tethering their effectiveness to annotation quality. In this paper, we present SIMS, the first self-improving model-steering framework that operates without relying on external supervision. At its core, SIMS autonomously generates and refines contrastive samples through iterative self-improvement cycles, enabling adaptive, context-specific steering. Additionally, SIMS employs novel strategies, including prompt ranking and contrast sampling, to further enhance steering efficacy. Extensive evaluation across diverse LLMs and benchmarks demonstrates that SIMS substantially outperforms existing methods in steering effectiveness and adaptability, highlighting self-improving model steering as a promising direction for future research on inference-time LLM alignment.