We present a novel approach to personalized sleep health management using few-shot Chain-of-Thought (CoT) distillation, enabling small-scale language models (> 2B parameters) to rival the performance of large language models (LLMs) in specialized health domains. Our method simultaneously distills problem-solving strategies, long-tail expert knowledge, and personalized recommendation capabilities from larger models into more efficient, compact models. Unlike existing systems, our approach offers three key functionalities: generating personalized sleep health recommendations, supporting user-specific follow-up inquiries, and providing responses to domain-specific knowledge questions. We focus on sleep health due to its measurability via wearable devices and its impact on overall well-being. Our experimental setup, involving GPT-4o for data synthesis, Qwen-max for instruction set creation, and Qwen2.5 1.5B for model distillation, demonstrates significant improvements over baseline small-scale models in penalization, reasoning, and knowledge application. Experiments using 100 simulated sleep reports and 1,000 domain-specific questions shows our model achieves comparable performance to larger models while maintaining efficiency for real-world deployment. This research not only advances AI-driven health management but also provides a novel approach to leveraging LLM capabilities in resource-constrained environments, potentially enhancing the accessibility of personalized healthcare solutions.

Large language models (LLM) have been attracting much attention from the community recently, due to their remarkable performance in all kinds of downstream tasks. According to the well-known scaling law, scaling up a dense LLM enhances its capabilities, but also significantly increases the computational complexity. Mixture-of-Experts (MoE) models address that by allowing the model size to grow without substantially raising training or inference costs. Yet MoE models face challenges regarding knowledge sharing among experts, making their performance somehow sensitive to routing accuracy. To tackle that, previous works introduced shared experts and combined their outputs with those of the top $K$ routed experts in an ``addition'' manner. In this paper, inspired by collective matrix factorization to learn shared knowledge among data, we propose CartesianMoE, which implements more effective knowledge sharing among experts in more like a ``multiplication'' manner. Extensive experimental results indicate that CartesianMoE outperforms previous MoE models for building LLMs, in terms of both perplexity and downstream task performance. And we also find that CartesianMoE achieves better expert routing robustness.

There is an increasing body of work using Large Language Models (LLMs) as agents for orchestrating workflows and making decisions in domains that require planning and multi-step reasoning. As a result, it is imperative to evaluate LLMs on core skills required for planning. In this work, we present ACPBench, a benchmark for evaluating the reasoning tasks in the field of planning. The benchmark consists of 7 reasoning tasks over 13 planning domains. The collection is constructed from planning domains described in a formal language. This allows us to synthesize problems with provably correct solutions across many tasks and domains. Further, it allows us the luxury of scale without additional human effort, i.e., many additional problems can be created automatically. Our extensive evaluation of 22 LLMs and OpenAI o1 reasoning models highlights the significant gap in the reasoning capability of the LLMs. Our findings with OpenAI o1, a multi-turn reasoning model, reveal significant gains in performance on multiple-choice questions, yet surprisingly, no notable progress is made on boolean questions. The ACPBench collection is available at https://ibm.github.io/ACPBench.

For the two investigated workflows, there are significant difference in the prediction of variants terms and additional phenotypic filtering terms. An intuitive comparison between the parallel and multistep extraction model is that, in the parallel workflow the models' instructions are rather simple, where the model is asked to predict only variants specific fields (variants extractor template) and other fields (filter extractor template) not concerning about the presence of the fields in the Beacon schema. Not all extracted terms in this extractor chain are valid for Beacon. A further validator template is further required here to filter out the terms that are not related to Beacon. In contrast, in the multistep workflow, both the variants and phenotypic terms are extracted only when they match with the beacon schema without the necessity of the validation prompt. Thus, although these models are predicting less terms, the extracted terms are aligned with the schema with less hallucination than the Parallel schema, as seen in previous section.

"This suit is brought by news publishers who seek redress for Perplexity's brazen scheme to compete for readers while simultaneously freeriding on the valuable content the publishers produce," according to the lawsuit filed in the southern district of New York by the Wall Street Journal parent Dow Jones and the New York Post. Perplexity did not immediately respond to emails from Reuters seeking comment. The AI company is among the leading startups attempting to uproot the search engine market dominated by Alphabet's Google. It assembles information from webpages it deems to be authoritative, then provides a summary directly within Perplexity's own tool. Perplexity uses a variety of large language models (LLMs) to generate its summaries, from OpenAI to Meta's open-source model Llama.

Artificial intelligence (AI) captured the world's attention in 2023 with the emergence of pre-trained models such as GPT, on which the conversational AI system ChatGPT is based. For the first time, we can converse with an entity, however imperfectly, about anything, as we do with other humans. This new capability provided by pre-trained models has created a paradigm shift in AI, transforming it from an application to a general-purpose technology that is configurable to specific uses. Whereas historically an AI model was trained to do one thing well, it is now usable for a variety of tasks such as general conversations; assistance; decision making; and the generation of documents, code, and video--for which it was not explicitly trained. The scientific history of AI provides a backdrop for evaluating and discussing the capabilities and limitations of this new technology, and the challenges that lie ahead.

This infrastructure gap – the difference between funding and construction – is vast. And while governments and companies everywhere are feeling the strain of constructing an energy efficient and sustainable built environment, it's proving more than humans can do alone. To redress this imbalance, many organizations are turning to various forms of AI, including large language models (LLMs) and machine learning (ML). Collectively, they are not yet able to fix all current infrastructure problems but they are already helping to reduce costs, risks, and increase efficiency. A shortage of skilled engineering and construction labor is a major problem.

The owner of TikTok has sacked an intern for allegedly sabotaging an internal artificial intelligence project. ByteDance said it had dismissed the person in August after they "maliciously interfered" with the training of artificial intelligence (AI) models used in a research project. Thanks to the video-sharing app TikTok and its Chinese counterpart, Douyin, which rank among the world's most popular mobile apps, ByteDance has risen to become one of the world's most important social media companies. Like other big players in the tech sector, ByteDance has raced to embrace generative AI. Its Doubao chatbot earlier this year took over from the competitor Baidu's Ernie in the race to produce a Chinese rival to OpenAI's ChatGPT.

Large Language Models (LLMs) have shown remarkable performance improvements and are rapidly gaining adoption in industry. However, the methods for improving LLMs are still designed by humans, which restricts the invention of new model-improving algorithms to human expertise and imagination. To address this, we propose the Self-Developing framework, which enables LLMs to autonomously generate and learn model-improvement algorithms. In this framework, the seed model generates, applies, and learns model-improving algorithms, continuously improving both the seed model and the algorithms themselves. In mathematical reasoning tasks, Self-Developing not only creates models that surpass the seed model but also consistently outperforms models created using human-designed algorithms. Additionally, these LLM-discovered algorithms demonstrate strong effectiveness, including transferability to out-of-domain models.

We introduce the Seismic Language Model (SeisLM), a foundational model designed to analyze seismic waveforms -- signals generated by Earth's vibrations such as the ones originating from earthquakes. SeisLM is pretrained on a large collection of open-source seismic datasets using a self-supervised contrastive loss, akin to BERT in language modeling. This approach allows the model to learn general seismic waveform patterns from unlabeled data without being tied to specific downstream tasks. When fine-tuned, SeisLM excels in seismological tasks like event detection, phase-picking, onset time regression, and foreshock-aftershock classification. The code has been made publicly available on https://github.com/liutianlin0121/seisLM.