Can ChatGPT diagnose Alzheimer's Disease (AD)? AD is a devastating neurodegenerative condition that affects approximately 1 in 9 individuals aged 65 and older, profoundly impairing memory and cognitive function. This paper utilises 9300 electronic health records (EHRs) with data from Magnetic Resonance Imaging (MRI) and cognitive tests to address an intriguing question: As a general-purpose task solver, can ChatGPT accurately detect AD using EHRs? We present an in-depth evaluation of ChatGPT using a black-box approach with zero-shot and multi-shot methods. This study unlocks ChatGPT's capability to analyse MRI and cognitive test results, as well as its potential as a diagnostic tool for AD. By automating aspects of the diagnostic process, this research opens a transformative approach for the healthcare system, particularly in addressing disparities in resource-limited regions where AD specialists are scarce. Hence, it offers a foundation for a promising method for early detection, supporting individuals with timely interventions, which is paramount for Quality of Life (QoL).

The increasing volume of drug combinations in modern therapeutic regimens needs reliable methods for predicting drug-drug interactions (DDIs). While Large Language Models (LLMs) have revolutionized various domains, their potential in pharmaceutical research, particularly in DDI prediction, remains largely unexplored. This study thoroughly investigates LLMs' capabilities in predicting DDIs by uniquely processing molecular structures (SMILES), target organisms, and gene interaction data as raw text input from the latest DrugBank dataset. We evaluated 18 different LLMs, including proprietary models (GPT-4, Claude, Gemini) and open-source variants (from 1.5B to 72B parameters), first assessing their zero-shot capabilities in DDI prediction. We then fine-tuned selected models (GPT-4, Phi-3.5 2.7B, Qwen-2.5 3B, Gemma-2 9B, and Deepseek R1 distilled Qwen 1.5B) to optimize their performance. Our comprehensive evaluation framework included validation across 13 external DDI datasets, comparing against traditional approaches such as l2-regularized logistic regression. Fine-tuned LLMs demonstrated superior performance, with Phi-3.5 2.7B achieving a sensitivity of 0.978 in DDI prediction, with an accuracy of 0.919 on balanced datasets (50% positive, 50% negative cases). This result represents an improvement over both zero-shot predictions and state-of-the-art machine-learning methods used for DDI prediction. Our analysis reveals that LLMs can effectively capture complex molecular interaction patterns and cases where drug pairs target common genes, making them valuable tools for practical applications in pharmaceutical research and clinical settings.

We investigate in-context temporal biases in attention heads and transformer outputs. Using cognitive science methodologies, we analyze attention scores and outputs of the GPT-2 models of varying sizes. Across attention heads, we observe effects characteristic of human episodic memory, including temporal contiguity, primacy and recency. Transformer outputs demonstrate a tendency toward in-context serial recall. Importantly, this effect is eliminated after the ablation of the induction heads, which are the driving force behind the contiguity effect. Our findings offer insights into how transformers organize information temporally during in-context learning, shedding light on their similarities and differences with human memory and learning.

Jailbreak backdoor attacks on LLMs have garnered attention for their effectiveness and stealth. However, existing methods rely on the crafting of poisoned datasets and the time-consuming process of fine-tuning. In this work, we propose JailbreakEdit, a novel jailbreak backdoor injection method that exploits model editing techniques to inject a universal jailbreak backdoor into safety-aligned LLMs with minimal intervention in minutes. JailbreakEdit integrates a multi-node target estimation to estimate the jailbreak space, thus creating shortcuts from the backdoor to this estimated jailbreak space that induce jailbreak actions. Our attack effectively shifts the models' attention by attaching strong semantics to the backdoor, enabling it to bypass internal safety mechanisms. Experimental results show that JailbreakEdit achieves a high jailbreak success rate on jailbreak prompts while preserving generation quality, and safe performance on normal queries. Our findings underscore the effectiveness, stealthiness, and explainability of JailbreakEdit, emphasizing the need for more advanced defense mechanisms in LLMs.

Drug optimization has become increasingly crucial in light of fast-mutating virus strains and drug-resistant cancer cells. Nevertheless, it remains challenging as it necessitates retaining the beneficial properties of the original drug while simultaneously enhancing desired attributes beyond its scope. In this work, we aim to tackle this challenge by introducing ScaffoldGPT, a novel Large Language Model (LLM) designed for drug optimization based on molecular scaffolds. Our work comprises three key components: (1) A three-stage drug optimization approach that integrates pretraining, finetuning, and decoding optimization. (2) A uniquely designed two-phase incremental training approach for pre-training the drug optimization LLM-based generator on molecule scaffold with enhanced performance. (3) A token-level decoding optimization strategy, TOP-N, that enabling controlled, reward-guided generation using pretrained/finetuned LLMs. Finally, by conducting a comprehensive evaluation on COVID and cancer benchmarks, we demonstrate that SCAFFOLDGPT outperforms the competing baselines in drug optimization benchmarks, while excelling in preserving the original functional scaffold and enhancing desired properties.

Refusal training on Large Language Models (LLMs) prevents harmful outputs, yet this defense remains vulnerable to both automated and human-crafted jailbreaks. We present a novel LLM-as-red-teamer approach in which a human jailbreaks a refusal-trained LLM to make it willing to jailbreak itself or other LLMs. We refer to the jailbroken LLMs as $J_2$ attackers, which can systematically evaluate target models using various red teaming strategies and improve its performance via in-context learning from the previous failures. Our experiments demonstrate that Sonnet 3.5 and Gemini 1.5 pro outperform other LLMs as $J_2$, achieving 93.0% and 91.0% attack success rates (ASRs) respectively against GPT-4o (and similar results across other capable LLMs) on Harmbench. Our work not only introduces a scalable approach to strategic red teaming, drawing inspiration from human red teamers, but also highlights jailbreaking-to-jailbreak as an overlooked failure mode of the safeguard. Specifically, an LLM can bypass its own safeguards by employing a jailbroken version of itself that is willing to assist in further jailbreaking. To prevent any direct misuse with $J_2$, while advancing research in AI safety, we publicly share our methodology while keeping specific prompting details private.

We developed the Enhanced Transformer for Health Outcome Simulation (ETHOS), an AI model that tokenizes patient health timelines (PHTs) from EHRs. ETHOS predicts future PHTs using transformer-based architectures. The Adaptive Risk Estimation System (ARES) employs ETHOS to compute dynamic and personalized risk probabilities for clinician-defined critical events. ARES incorporates a personalized explainability module that identifies key clinical factors influencing risk estimates for individual patients. ARES was evaluated on the MIMIC-IV v2.2 dataset in emergency department (ED) settings, benchmarking its performance against traditional early warning systems and machine learning models. We processed 299,721 unique patients from MIMIC-IV into 285,622 PHTs, with 60% including hospital admissions. The dataset contained over 357 million tokens. ETHOS outperformed benchmark models in predicting hospital admissions, ICU admissions, and prolonged hospital stays, achieving superior AUC scores. ETHOS-based risk estimates demonstrated robustness across demographic subgroups with strong model reliability, confirmed via calibration curves. The personalized explainability module provides insights into patient-specific factors contributing to risk. ARES, powered by ETHOS, advances predictive healthcare AI by providing dynamic, real-time, and personalized risk estimation with patient-specific explainability to enhance clinician trust. Its adaptability and superior accuracy position it as a transformative tool for clinical decision-making, potentially improving patient outcomes and resource allocation in emergency and inpatient settings. We release the full code at github.com/ipolharvard/ethos-ares to facilitate future research.

Can we derive computational metrics to quantify visual creativity in drawings across intelligent agents, while accounting for inherent differences in technical skill and style? To answer this, we curate a novel dataset consisting of 1338 drawings by children, adults and AI on a creative drawing task. We characterize two aspects of the drawings -- (1) style and (2) content. For style, we define measures of ink density, ink distribution and number of elements. For content, we use expert-annotated categories to study conceptual diversity, and image and text embeddings to compute distance measures. We compare the style, content and creativity of children, adults and AI drawings and build simple models to predict expert and automated creativity scores. We find significant differences in style and content in the groups -- children's drawings had more components, AI drawings had greater ink density, and adult drawings revealed maximum conceptual diversity. Notably, we highlight a misalignment between creativity judgments obtained through expert and automated ratings and discuss its implications. Through these efforts, our work provides, to the best of our knowledge, the first framework for studying human and artificial creativity beyond the textual modality, and attempts to arrive at the domain-agnostic principles underlying creativity. Our data and scripts are available on GitHub.

Large language models (LLMs) have facilitated a wide range of applications with distinct service-level objectives (SLOs), from latency-sensitive online tasks like interactive chatbots to throughput-oriented offline workloads like document summarization. The existing deployment model, which dedicates machines to each workload, simplifies SLO management but often leads to poor resource utilization. This paper introduces HyGen, an interference-aware LLM serving system that enables efficient co-location of online and offline workloads while preserving latency requirements. HyGen incorporates two key innovations: (1) performance control mechanisms, including a latency predictor to estimate batch execution time and an SLO-aware profiler to quantify latency interference, and (2) SLO-aware offline scheduling policies that maximize serving throughput and prevent starvation, without compromising online serving latency. Our evaluation on production workloads shows that HyGen achieves up to 3.87x overall throughput and 5.84x offline throughput gains over online and hybrid serving baselines, respectively, while strictly satisfying latency SLOs.

Understanding temporal dynamics is critical for conversational agents, enabling effective content analysis and informed decision-making. However, time-aware datasets, particularly for persona-grounded conversations, are still limited, which narrows their scope and diminishes their complexity. To address this gap, we introduce MTPChat, a multimodal, time-aware persona dialogue dataset that integrates linguistic, visual, and temporal elements within dialogue and persona memory. Leveraging MTPChat, we propose two time-sensitive tasks: Temporal Next Response Prediction (TNRP) and Temporal Grounding Memory Prediction (TGMP), both designed to assess a model's ability to understand implicit temporal cues and dynamic interactions. Additionally, we present an innovative framework featuring an adaptive temporal module to effectively integrate multimodal streams and capture temporal dependencies. Experimental results validate the challenges posed by MTPChat and demonstrate the effectiveness of our framework in multimodal time-sensitive scenarios.