With the wide adoption of large language models (LLMs) in information assistance, it is essential to examine their alignment with human communication styles and values. We situate this study within the context of fact-checking health information, given the critical challenge of rectifying conceptions and building trust. Recent studies have explored the potential of LLM for health communication, but style differences between LLMs and human experts and associated reader perceptions remain under-explored. In this light, our study evaluates the communication styles of LLMs, focusing on how their explanations differ from those of humans in three core components of health communication: information, sender, and receiver. We compiled a dataset of 1498 health misinformation explanations from authoritative fact-checking organizations and generated LLM responses to inaccurate health information. Drawing from health communication theory, we evaluate communication styles across three key dimensions of information linguistic features, sender persuasive strategies, and receiver value alignments. We further assessed human perceptions through a blinded evaluation with 99 participants. Our findings reveal that LLM-generated articles showed significantly lower scores in persuasive strategies, certainty expressions, and alignment with social values and moral foundations. However, human evaluation demonstrated a strong preference for LLM content, with over 60% responses favoring LLM articles for clarity, completeness, and persuasiveness. Our results suggest that LLMs' structured approach to presenting information may be more effective at engaging readers despite scoring lower on traditional measures of quality in fact-checking and health communication.

Flowcharts are indispensable tools in software design and business-process analysis, yet current Vision Language Models (VLMs) frequently misinterpret the directional arrows and graph topology that set these diagrams apart from natural images. This paper introduces a seven-stage pipeline, grouped into three broader processes--(1) arrow-aware detection of nodes and arrow endpoints; (2) Optical Character Recognition (OCR) to extract node text; and (3) construction of a structured prompt that guides the VLMs. Tested on a 90-question benchmark distilled from 30 annotated flowcharts, our method raises overall accuracy from 80% to 89% (+9 pp), a sizeable and statistically significant gain achieved without task-specific fine-tuning of the VLMs. The benefit is most pronounced for next-step queries (25/30 30/30; 100%, +17 pp); branch-result questions improve more modestly, and before-step queries remain difficult. A parallel evaluation with an LLM-as-a-Judge protocol shows the same trends, reinforcing the advantage of explicit arrow encoding. Limitations include dependence on detector and OCR precision, the small evaluation set, and residual errors at nodes with multiple incoming edges. Future work will enlarge the benchmark with synthetic and handwritten flowcharts and assess the approach on Business Process Model and Notation (BPMN) and Unified Modeling Language (UML).

The escalating demand for efficient decoding in large language models (LLMs) is particularly critical for reasoning-intensive architectures like OpenAI-o3 and DeepSeek-R1, which depend on extended chain-of-thought reasoning. This study investigates speculative decoding techniques through dense LLM architectures to establish foundational insights for accelerating reasoning tasks. While speculative decoding methods leveraging parallel draft-verification cycles have emerged as promising acceleration techniques, the scaling laws governing decoding efficiency remain under-explored compared to conventional backbone LLMs developed through Pretraining->SFT->RLHF training paradigms. In this work, we discover Log-linear Scaling Laws (Theorem 1.1, 1.2 and 1.3) governing draft model acceptance rate (or decoding speed) across three dimensions: pretraining token volume, draft model capacity, and decoding batch size. Building on these laws, we achieve Scylla, which coordinates multi-dimensional scaling for popular LLMs (Llama2/3, Qwen2.5). Empirical validation shows Scylla achieves 1.5-2.2 higher acceptance rate than EAGLE2 and 0.3 higher than EAGLE3 at temperature T = 0, with peak performance gains on summarization and QA tasks (Figure 2). Industrial inference engine deployments demonstrate 2X decoding throughput improvements over EAGLE2 (Table 5), validating the transformative potential of systematic scaling for efficient LLM inference. Code will be released later.

On a Wednesday morning last month, I thought, just for a second, that AI was going to kill me. I had hailed a self-driving Waymo to bring me to a hacker house in Nob Hill, San Francisco. Just a few blocks from arrival, the car lurched toward the other lane--which was, thankfully, empty--and immediately jerked back. That sense of peril felt right for the moment. As I stepped into the cab, Federal Reserve Chair Jerome Powell was delivering a speech criticizing President Donald Trump's economic policies, and in particular the administration's sweeping on-again, off-again tariffs. A day earlier, the White House had claimed that Chinese goods would be subject to overall levies as high as 245 percent when accounting for preexisting tariffs, and the AI giant Nvidia's stock had plummeted after the company reported that it expected to take a quarterly hit of more than 5 billion for selling to China.

As Brian Chesky tells it, the reinvention of Airbnb started with the coup at OpenAI. On November 17, 2023, the board of OpenAI fired company CEO Sam Altman. His friend Chesky leapt into action--publicly defending his pal on X, getting on the phone with Microsoft's CEO, and throwing himself into the thick of Altman's battle to retake OpenAI. Five days later Altman prevailed, and Chesky--"I was so jacked up," he says--turned his buzzing mind to his own company, Airbnb. The Chesky extended family had already held their turkey get-together a week earlier, and the Airbnb CEO had no holiday plan.

After publishing my third book in early April, I kept encountering headlines that made me feel like the protagonist of some Black Mirror episode. "Vauhini Vara consulted ChatGPT to help craft her new book'Searches,'" one of them read. "To tell her own story, this acclaimed novelist turned to ChatGPT," said another. "Vauhini Vara examines selfhood with assistance from ChatGPT," went a third. The publications describing Searches this way were reputable and fact-based.

A few weeks ago, OpenAI pulled off one of the greatest corporate promotions in recent memory. Whereas the initial launch of ChatGPT, back in 2022, was "one of the craziest viral moments i'd ever seen," CEO Sam Altman wrote on social media, the response to a new upgrade was, in his words, "biblical": 1 million users supposedly signed up to use the chatbot in just one hour, Altman reported, thanks to a new, more permissive image-generating capability that could imitate the styles of various art and design studios. Altman called it "a new high-water mark for us in allowing creative freedom." Almost immediately, images began to flood the internet. The most popular style, by a long shot, was that of Studio Ghibli, the Japanese animation studio co-founded by Hayao Miyazaki and widely beloved for films such as Spirited Away and Princess Mononoke.

The United States has reportedly been investigating reports that Nvidia GPUs have landed illegally in China to be used by Chinese LLMs like DeepSeek, and one US lawmaker will be introducing a new bill that aims to track the locations of AI chips--like the ones made by Nvidia--after they're sold, reports Reuters and Neowin. The smuggling of CPUs and GPUs is nothing new. PC components have often been smuggled across the ocean to countries like China and other East Asian countries for years. But with the rising power of AI and the implications of AI on technological prowess, it's not unusual that the US government wouldn't want that tech falling into rival hands. The proposed legislation would oblige US authorities to develop regulations for location verification of AI chips.

SoftBank Group reported a 124% jump in quarterly profit on resilient AI demand that's supporting startup valuations and chip unit sales, a boost to its aggressive data center investment plans. The Tokyo-based company reported net income of 517.18 billion ( 3.5 billion) in its fiscal fourth quarter. It was helped by the Vision Fund, which swung to a profit of 26.1 billion. The earnings come at a critical juncture for SoftBank as it plans to invest 30 billion in OpenAI while leading a 100 billion foray into building AI hardware in the U.S. Maintaining a healthy cash flow and balance sheet is key to securing the billions of dollars needed at minimum cost.

The rise of large language models (LLMs), such as ChatGPT, DeepSeek, and Grok-3, has reshaped the artificial intelligence landscape. As prominent examples of foundational models (FMs) built on LLMs, these models exhibit remarkable capabilities in generating human-like content, bringing us closer to achieving artificial general intelligence (AGI). However, their large-scale nature, sensitivity to privacy concerns, and substantial computational demands present significant challenges to personalized customization for end users. To bridge this gap, this paper presents the vision of artificial personalized intelligence (API), focusing on adapting these powerful models to meet the specific needs and preferences of users while maintaining privacy and efficiency. Specifically, this paper proposes personalized federated intelligence (PFI), which integrates the privacy-preserving advantages of federated learning (FL) with the zero-shot generalization capabilities of FMs, enabling personalized, efficient, and privacy-protective deployment at the edge. We first review recent advances in both FL and FMs, and discuss the potential of leveraging FMs to enhance federated systems. We then present the key motivations behind realizing PFI and explore promising opportunities in this space, including efficient PFI, trustworthy PFI, and PFI empowered by retrieval-augmented generation (RAG). Finally, we outline key challenges and future research directions for deploying FM-powered FL systems at the edge with improved personalization, computational efficiency, and privacy guarantees. Overall, this survey aims to lay the groundwork for the development of API as a complement to AGI, with a particular focus on PFI as a key enabling technique.