OpenAI CEO Sam Altman speaks in Washington in July. OpenAI signed a multiyear partnership Monday with chipmaker Advanced Micro Devices as the ChatGPT-maker continues an investment spree to secure massive amounts of computing power for rolling out generative artificial intelligence. The companies announced the plan to develop AI data centers that the chipmaker said would bring in tens of billions of dollars in new revenue over the next five years. AMD's share price surged 35% when markets opened on news of the agreement that would see the company deliver six gigawatts worth of chips to the ChatGPT-maker. In a time of both misinformation and too much information, quality journalism is more crucial than ever.

Large Language Models (LLMs) have become critical to modern software development, but their reliance on uncurated web-scale datasets for training introduces a significant security risk: the absorption and reproduction of malicious content. To systematically evaluate this risk, we introduce Scam2Prompt, a scalable automated auditing framework that identifies the underlying intent of a scam site and then synthesizes innocuous, developer-style prompts that mirror this intent, allowing us to test whether an LLM will generate malicious code in response to these innocuous prompts. In a large-scale study of four production LLMs (GPT -4o, GPT -4o-mini, Llama-4-Scout, and DeepSeek-V3), we found that Scam2Prompt's innocuous prompts triggered malicious URL generation in 4.24% of cases. To test the persistence of this security risk, we constructed Innoc2Scam-bench, a benchmark of 1,559 innocuous prompts that consistently elicited malicious code from all four initial LLMs. When applied to seven additional production LLMs released in 2025, we found the vulnerability is not only present but severe, with malicious code generation rates ranging from 12.7% to 43.8%. Furthermore, existing safety measures like state-of-the-art guardrails proved insufficient to prevent this behavior, with an overall detection rate of less than 0.3%.

Persona prompting is increasingly used in large language models (LLMs) to simulate views of various sociodemographic groups. However, how a persona prompt is formulated can significantly affect outcomes, raising concerns about the fidelity of such simulations. Using five open-source LLMs, we systematically examine how different persona prompt strategies, specifically role adoption formats and demographic priming strategies, influence LLM simulations across 15 intersectional demographic groups in both open- and closed-ended tasks. Our findings show that LLMs struggle to simulate marginalized groups but that the choice of demographic priming and role adoption strategy significantly impacts their portrayal. Specifically, we find that prompting in an interview-style format and name-based priming can help reduce stereotyping and improve alignment. Surprisingly, smaller models like OLMo-2-7B outperform larger ones such as Llama-3.3-70B. Our findings offer actionable guidance for designing sociodemographic persona prompts in LLM-based simulation studies.

LLM inference often generates a batch of candidates for a prompt and selects one via strategies like majority voting or Best-of- N (BoN). For difficult tasks, this single-shot selection often underperforms. Consequently, evaluations commonly report Pass@$k$: the agent may submit up to $k$ responses, and only the best of them is used when computing regret. Motivated by this, we study inference scaling in the more general Pass@$k$ inference setting, and prove that neither majority voting nor BoN exhibits the desirable scaling with $k$ and the sampling budget $N$. Combining the advantages of majority voting and BoN, we propose a new inference strategy called Best-of-Majority (BoM), with a pivotal step that restricts the candidates to the responses with high frequency in the $N$ samples before selecting the top-$k$ rewards. We prove that when the sampling budget is $N=\tildeΩ(C^*)$, the regret of BoM is $O(ε_{\mathrm{opt}}+\sqrt{ε_{\mathrm{RM}}^2C^*/k})$, where $C^*$ is the coverage coefficient, $ε_{\mathrm{RM}}$ is the estimation error of the reward model, and $ε_{\mathrm{opt}}$ is the estimation error of reward at the optimal response. We further establish a matching lower bound, certifying that our algorithm is minimax optimal. Beyond optimality, BoM has a key advantage: unlike majority voting and BoN, its performance does not degrade when increasing $N$. Experimental results of inference on math problems show BoM outperforming both majority voting and BoN.

Training modern machine learning models requires a careful choice of hyperparameters. A common practice for setting the learning rate (LR) is to linearly increase the LR in the beginning (warm-up stage) [Goyal et al., 2017, Vaswani et al., 2017] and gradually decrease at the end of the training (decay stage) [Loshchilov and Hutter, 2016, Vaswani et al., 2017, Hoffmann et al., 2022b, Zhang et al., 2023, Dremov et al., 2025]. Decaying the LR is a classical requirement in the theoretical analysis of SGD, ensuring convergence under broad conditions [Defazio et al., 2023, Gower et al., 2021], and it has been consistently observed to improve empirical performance [Loshchilov and Hutter, 2016, Hu et al., 2024, Hägele et al., 2024]. Recent work further demonstrates that decaying step sizes can improve theoretical guarantees by yielding tighter bounds [Schaipp et al., 2025]. By contrast, the practice of linearly increasing the LR at the start of training (warm-up phase) has become nearly ubiquitous in modern deep learning [He et al., 2016, Hu et al., 2024, Hägele et al., 2024], yet a clear theoretical understanding of why it helps optimization remains elusive. This raises the central question we address in this paper: Why does LR warm-up improve training, and under what conditions can its benefits be theoretically justified?

Automating clinical documentation with large language models requires precise alignment with priorities such as completeness and factual grounding. We present an evaluation-integrated reinforcement learning framework for long-form clinical text generation that couples Group Relative Policy Optimization (GRPO) with DocLens, a claim-level evaluator that provides deterministic, dialogue-grounded rewards. Our method directly optimizes factual grounding and completeness without training a separate reward model or relying on human-authored references. Empirically, the approach improves clinical note quality and reduces training cost via a simple reward-gating strategy. An independent GPT-5 qualitative evaluation further supports these gains, showing higher preference for GRPO outputs in factuality, completeness, and brevity, with fewer omissions and hallucinations. Because the benchmarks are relatively clean and the base model already well aligned, these improvements likely represent a conservative lower bound. The framework is scalable to real-world settings and can incorporate custom objectives such as guideline adherence or billing preferences.

Spoken language understanding (SLU) tasks involve diverse skills that probe the information extraction, classification and/or generation capabilities of models. In this setting, task-specific training data may not always be available. While traditional task-specific SLU models are unable to cater to such requirements, the speech-text large language models (LLMs) offer a promising alternative with emergent abilities. However, out of-the-box, our evaluations indicate that the zero/few-shot performance of prominent open-source speech-text LLMs on SLU tasks are not up to the mark. In this paper, we introduce a novel approach to robust task-agnostic fine-tuning using randomized class labels. With this proposed fine-tuning, we illustrate that the performance of the speech-text LLMs on an unseen task is significantly improved over standard approaches. Critically, the proposed approach avoids the requirement of task-specific data annotations for enabling new tasks in speech-text LLMs.

Accurately evaluating machine-translated text remains a long-standing challenge, particularly for long documents. Recent work has shown that large language models (LLMs) can serve as reliable and interpretable sentence-level translation evaluators via MQM error span annotations. With modern LLMs supporting larger context windows, a natural question arises: can we feed entire document translations into an LLM for quality assessment? Ideally, evaluation should be invariant to text length, producing consistent error spans regardless of input granularity. However, our analysis shows that text length significantly impacts evaluation: longer texts lead to fewer error spans and reduced system ranking accuracy. To address this limitation, we evaluate several strategies, including granularity-aligned prompting, Focus Sentence Prompting (FSP), and a fine-tuning approach to better align LLMs with the evaluation task. The latter two methods largely mitigate this length bias, making LLMs more reliable for long-form translation evaluation.

Reinforcement learning from human or AI feedback (RLHF/RLAIF) has become the standard paradigm for aligning large language models (LLMs). However, most pipelines rely on a single reward model (RM), limiting alignment quality and risking overfitting. Recent work explores RM routing--dynamically selecting an RM from a candidate pool to exploit complementary strengths while maintaining O(1) RM calls--but existing methods suffer from cold-start and insufficient exploration. We propose BayesianRouter, a hybrid routing framework that combines offline RM strengths learning with online Bayesian selection. In the offline stage, a multi-task router is trained on preference data to estimate per-RM reliability. In the online stage, a Bayesian Thompson sampling router performs per-query RM selection, initializing RM-specific weight vectors with offline embeddings as Gaussian priors and adaptively updating their posteriors with online rewards to adapt to the evolving policy distribution. Extensive experiments on instruction-following (AlpacaEval-2, Arena-Hard, MT -Bench) and reasoning (GSM8K, MMLU) benchmarks show that BayesianRouter consistently outperforms individual RMs, RM ensembling, and existing routing methods. Large language models (LLMs) have revolutionized artificial intelligence, demonstrating substantial capabilities in language understanding, reasoning, and open-ended text generation across diverse domains (Guo et al., 2025; Achiam et al., 2023). To safely and effectively deploy LLMs, recent post-training techniques, particularly reinforcement learning from human feedback (RLHF) and its AI-augmented variant (RLAIF) (Wang et al., 2024b), aimed at aligning LLMs with human values and preferences. In standard RLHF, a reward model (RM) provides the feedback signal for optimizing the policy LLM; for example, in PPO, the RM provides a scalar reward to directly increase the probability of preferred responses (Ziegler et al., 2019), while in direct preference optimization (DPO), the RM compares two candidate responses to determine which is better (Dong et al., 2024).

Recent progress in retrieval-augmented generation (RAG) has led to more accurate and interpretable multi-hop question answering (QA). Yet, challenges persist in integrating iterative reasoning steps with external knowledge retrieval. To address this, we introduce StepChain GraphRAG, a framework that unites question decomposition with a Breadth-First Search (BFS) Reasoning Flow for enhanced multi-hop QA. Our approach first builds a global index over the corpus; at inference time, only retrieved passages are parsed on-the-fly into a knowledge graph, and the complex query is split into sub-questions. For each sub-question, a BFS-based traversal dynamically expands along relevant edges, assembling explicit evidence chains without overwhelming the language model with superfluous context. Experiments on MuSiQue, 2WikiMultiHopQA, and HotpotQA show that StepChain GraphRAG achieves state-of-the-art Exact Match and F1 scores. StepChain GraphRAG lifts average EM by 2.57% and F1 by 2.13% over the SOTA method, achieving the largest gain on HotpotQA (+4.70% EM, +3.44% F1). StepChain GraphRAG also fosters enhanced explainability by preserving the chain-of-thought across intermediate retrieval steps. We conclude by discussing how future work can mitigate the computational overhead and address potential hallucinations from large language models to refine efficiency and reliability in multi-hop QA.