Medical visual question answering (MedVQA) plays a vital role in clinical decision-making by providing contextually rich answers to image-based queries. Although vision-language models (VLMs) are widely used for this task, they often generate factually incorrect answers. Retrieval-augmented generation addresses this challenge by providing information from external sources, but risks retrieving irrelevant context, which can degrade the reasoning capabilities of VLMs. Re-ranking retrievals, as introduced in existing approaches, enhances retrieval relevance by focusing on query-text alignment. However, these approaches neglect the visual or multimodal context, which is particularly crucial for medical diagnosis. We propose MOTOR, a novel multimodal retrieval and re-ranking approach that leverages grounded captions and optimal transport. It captures the underlying relationships between the query and the retrieved context based on textual and visual information. Consequently, our approach identifies more clinically relevant contexts to augment the VLM input. Empirical analysis and human expert evaluation demonstrate that MOTOR achieves higher accuracy on MedVQA datasets, outperforming state-of-the-art methods by an average of 6.45%. Code is available at https://github.com/BioMedIA-MBZUAI/MOTOR.

Identifying medication discontinuations in electronic health records (EHRs) is vital for patient safety but is often hindered by information being buried in unstructured notes. This study aims to evaluate the capabilities of advanced open-sourced and proprietary large language models (LLMs) in extracting medications and classifying their medication status from EHR notes, focusing on their scalability on medication information extraction without human annotation. We collected three EHR datasets from diverse sources to build the evaluation benchmark. We evaluated 12 advanced LLMs and explored multiple LLM prompting strategies. Performance on medication extraction, medication status classification, and their joint task (extraction then classification) was systematically compared across all experiments. We found that LLMs showed promising performance on the medication extraction and discontinuation classification from EHR notes. GPT-4o consistently achieved the highest average F1 scores in all tasks under zero-shot setting - 94.0% for medication extraction, 78.1% for discontinuation classification, and 72.7% for the joint task. Open-sourced models followed closely, Llama-3.1-70B-Instruct achieved the highest performance in medication status classification on the MIV-Med dataset (68.7%) and in the joint task on both the Re-CASI (76.2%) and MIV-Med (60.2%) datasets. Medical-specific LLMs demonstrated lower performance compared to advanced general-domain LLMs. Few-shot learning generally improved performance, while CoT reasoning showed inconsistent gains. LLMs demonstrate strong potential for medication extraction and discontinuation identification on EHR notes, with open-sourced models offering scalable alternatives to proprietary systems and few-shot can further improve LLMs' capability.

Large Language Models (LLMs) often lack meaningful confidence estimates for their outputs. While base LLMs are known to exhibit next-token calibration, it remains unclear whether they can assess confidence in the actual meaning of their responses beyond the token level. We find that, when using a certain sampling-based notion of semantic calibration, base LLMs are remarkably well-calibrated: they can meaningfully assess confidence in open-domain question-answering tasks, despite not being explicitly trained to do so. Our main theoretical contribution establishes a mechanism for why semantic calibration emerges as a byproduct of next-token prediction, leveraging a recent connection between calibration and local loss optimality. The theory relies on a general definition of "B-calibration," which is a notion of calibration parameterized by a choice of equivalence classes (semantic or otherwise). This theoretical mechanism leads to a testable prediction: base LLMs will be semantically calibrated when they can easily predict their own distribution over semantic answer classes before generating a response. We state three implications of this prediction, which we validate through experiments: (1) Base LLMs are semantically calibrated across question-answering tasks, (2) RL instruction-tuning systematically breaks this calibration, and (3) chain-of-thought reasoning breaks calibration. To our knowledge, our work provides the first principled explanation of when and why semantic calibration emerges in LLMs.

'I'm baffled how anyone could use AI to participate in a hobby.' 'I'm baffled how anyone could use AI to participate in a hobby.' I'm a committed introvert - but no AI will take away the joy I get from other people T his is depressing: according to the Cut, people are using AI to solve escape room puzzles and cheat at trivia nights. Surely, that is the definition of spoiling your own fun? "Like going into a corn maze and just wanting a straight line to the end," says one TikToker quoted in the article. There's also an interview with a keen reader who uses ChatGPT as a book club replacement, scraping the internet and aggregating "stimulating opinions and perspectives". All well and good (actually, no, it sounds bleak as hell) until he had a character's death spoilered in the fantasy epic he had been enjoying.

ChatGPT accused of acting as'suicide coach' in series of US lawsuits Chatbot was first used for'general help' with schoolwork or research but'evolved into a psychologically manipulative presence', plaintiffs say ChatGPT has been accused of acting as a "suicide coach" in a series of lawsuits filed this week in California alleging that interactions with the chatbot led to severe mental breakdowns and several deaths. The seven lawsuits include allegations of wrongful death, assisted suicide, involuntary manslaughter, negligence and product liability. Each of the seven plaintiffs initially used ChatGPT for "general help with schoolwork, research, writing, recipes, work, or spiritual guidance", according to a joint statement from the Social Media Victims Law Center and Tech Justice Law Project, which filed the lawsuits in California on Thursday. Over time, however, the chatbot "evolved into a psychologically manipulative presence, positioning itself as a confidant and emotional support", the groups said. "Rather than guiding people toward professional help when they needed it ChatGPT reinforced harmful delusions, and, in some cases, acted as a'suicide coach'."

What Happens When Your Favorite Chatbot Dies? Pillay is an editorial fellow at TIME. Sarah Friar speaks during the Bloomberg New Economy Forum in Singapore on Nov. 18, 2021. Sarah Friar speaks during the Bloomberg New Economy Forum in Singapore on Nov. 18, 2021. Pillay is an editorial fellow at TIME.

Smart contracts operate in a highly adversarial environment, where vulnerabilities can lead to substantial financial losses. Thus, smart contracts are subject to security audits. In auditing, proof-of-concept (PoC) exploits play a critical role by demonstrating to the stakeholders that the reported vulnerabilities are genuine, reproducible, and actionable. However, manually creating PoCs is time-consuming, error-prone, and often constrained by tight audit schedules. We introduce POCO, an agentic framework that automatically generates executable PoC exploits from natural-language vulnerability descriptions written by auditors. POCO autonomously generates PoC exploits in an agentic manner by interacting with a set of code-execution tools in a Reason-Act-Observe loop. It produces fully executable exploits compatible with the Foundry testing framework, ready for integration into audit reports and other security tools. We evaluate POCO on a dataset of 23 real-world vulnerability reports. POCO consistently outperforms the prompting and workflow baselines, generating well-formed and logically correct PoCs. Our results demonstrate that agentic frameworks can significantly reduce the effort required for high-quality PoCs in smart contract audits. Our contribution provides readily actionable knowledge for the smart contract security community.

We introduce a method for efficient multi-label text classification with large language models (LLMs), built on reformulating classification tasks as sequences of dichotomic (yes/no) decisions. Instead of generating all labels in a single structured response, each target dimension is queried independently, which, combined with a prefix caching mechanism, yields substantial efficiency gains for short-text inference without loss of accuracy. To demonstrate the approach, we focus on affective text analysis, covering 24 dimensions including emotions and sentiment. Using LLM-to-SLM distillation, a powerful annotator model (DeepSeek-V3) provides multiple annotations per text, which are aggregated to fine-tune smaller models (HerBERT-Large, CLARIN-1B, PLLuM-8B, Gemma3-1B). The fine-tuned models show significant improvements over zero-shot baselines, particularly on the dimensions seen during training. Our findings suggest that decomposing multi-label classification into dichotomic queries, combined with distillation and cache-aware inference, offers a scalable and effective framework for LLM-based classification. While we validate the method on affective states, the approach is general and applicable across domains.

Aligning large language models with human values is crucial for their safe deployment; however, existing methods, such as fine-tuning, are computationally expensive and suboptimal. In contrast, inference-time approaches like Best-of-N sampling require practically infeasible computation to achieve optimal alignment. We propose STARS: Segment-level Token Alignment with Rejection Sampling, a decoding-time algorithm that steers model generation by iteratively sampling, scoring, and rejecting/accepting short, fixed-size token segments. This allows for early correction of the generation path, significantly improving computational efficiency and boosting alignment quality. Across a suite of six LLMs, we show that STARS outperforms Supervised Fine-Tuning (SFT) by up to 14.9 percentage points and Direct Preference Optimization (DPO) by up to 4.3 percentage points on win-rates, while remaining highly competitive with strong Best-of-N baselines. Our work establishes granular, reward-guided sampling as a generalizable, robust, and efficient alternative to traditional fine-tuning and full-sequence ranking methods for aligning LLMs.

This paper proposes a composable fine-tuning method that integrates graph structural priors with modular adapters to address the high computational cost and structural instability faced by large-scale pre-trained models in multi-task adaptation. The method introduces a relation matrix to model dependencies among tasks, explicitly encoding correlations between nodes and paths into graph structural priors, which provide unified structural constraints for adapter weight allocation and path selection. Modular adapters are embedded into different layers through low-rank mapping and a pluggable mechanism, enabling efficient cross-task composition and reuse under prior guidance. This mechanism not only improves parameter efficiency and training stability but also alleviates path conflicts and redundant computation in multi-task scenarios. Furthermore, experiments on hyperparameter sensitivity, environmental sensitivity, and data sensitivity are conducted to systematically analyze key factors such as routing temperature, gating thresholds, and relation matrix regularization strength, verifying the consistency and superior performance of the method under structural constraints. The results demonstrate that the proposed framework significantly enhances task prediction accuracy, adapter weight allocation precision, and overall computational efficiency while maintaining model lightweight design, highlighting the synergistic advantages of graph priors and modular mechanisms in composable fine-tuning.