Latest

Retrieval-Augmented Generation (RAG) models excel in knowledge-intensive tasks, especially under few-shot learning constraints. We introduce CoRAG, a framework extending RAG to collaborative settings, where clients jointly train a shared model using a collaborative passage store. To evaluate CoRAG,...

Language models often struggle with cross-mode knowledge retrieval -- the ability to access knowledge learned in one format (mode) when queried in another. We demonstrate that models trained on multiple data sources (e.g., Wikipedia and TinyStories) exhibit significantly reduced accuracy when...

Effective information retrieval (IR) in settings with limited training data, particularly for complex queries, remains a challenging task. This paper introduces IR2, Information Regularization for Information Retrieval, a technique for reducing overfitting during synthetic data generation. This...

Recent advances in general medical AI have made significant strides, but existing models often lack the reasoning capabilities needed for complex medical decision-making. This paper presents GMAI-VL-R1, a multimodal medical reasoning model enhanced by reinforcement learning (RL) to improve its...

Evaluating large language models (LLMs) effectively remains a critical bottleneck, as traditional static benchmarks suffer from saturation and contamination, while human evaluations are costly and slow. This hinders timely or domain-specific assessment, crucial for real-world applications. We...

Since the advent of reasoning-based large language models, many have found great success from distilling reasoning capabilities into student models. Such techniques have significantly bridged the gap between reasoning and standard LLMs on coding tasks. Despite this, much of the progress on distilling...

Synthetic verification techniques such as generating test cases and reward modelling are common ways to enhance the coding capabilities of large language models (LLM) beyond predefined tests. Additionally, code verification has recently found great success as a critical component in improving...

Diffusion models have gained tremendous success in text-to-image generation, yet still lag behind with visual understanding tasks, an area dominated by autoregressive vision-language models. We propose a large-scale and fully end-to-end diffusion model for multi-modal understanding and generation...

Code generation by Llama 3.1 models, such as Meta’s Llama 3.1 405B, represents a significant advancement in the field of artificial intelligence, particularly in natural language processing and programming automation. This paper explores the capabilities and applications of Llama-driven code...

The best performing transformer-based language models use subword tokenization techniques, such as Byte-Pair-Encoding (BPE). However, these approaches often overlook linguistic principles, such as morphological segmentation, which we believe is fundamental for understanding language-specific word...

Spreadsheets are characterized by their extensive two-dimensional grids, flexible layouts, and varied formatting options, which pose significant challenges for large language models (LLMs). In response, we introduce SpreadsheetLLM, pioneering an efficient encoding method designed to unleash and...

This paper introduces STAR-1, a high-quality, just-1k-scale safety dataset specifically designed for large reasoning models (LRMs) like DeepSeek-R1. Built on three core principles -- diversity, deliberative reasoning, and rigorous filtering -- STAR-1 aims to address the critical needs for safety...

The rapid development of Large Multimodal Models (LMMs) for 2D images and videos has spurred efforts to adapt these models for interpreting 3D scenes. However, the absence of large-scale 3D vision-language datasets has posed a significant obstacle. To address this issue, typical approaches focus on...

Humans continuously make new discoveries, and understanding temporal sequence of events leading to these breakthroughs is essential for advancing science and society. This ability to reason over time allows us to identify future steps and understand the effects of financial and political decisions on...

We introduce PaperBench, a benchmark evaluating the ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch, including understanding paper contributions, developing a codebase, and successfully executing experiments....

Recently, building retrieval-augmented generation (RAG) systems to enhance the capability of large language models (LLMs) has become a common practice. Especially in the legal domain, previous judicial decisions play a significant role under the doctrine of stare decisis which emphasizes the...

We address the challenge of representation learning from a continuous stream of video as input, in a self-supervised manner. This differs from the standard approaches to video learning where videos are chopped and shuffled during training in order to create a non-redundant batch that satisfies the...

Perceiving the environment and its changes over time corresponds to two fundamental yet heterogeneous types of information: semantics and motion. Previous end-to-end autonomous driving works represent both types of information in a single feature vector. However, including motion related tasks, such...

Recent advancements in large language models (LLMs) have revolutionized various domains, bringing significant progress and new opportunities. Despite progress in speech-related tasks, LLMs have not been sufficiently explored in multi-talker scenarios. In this work, we present a pioneering effort to...

Financial LLMs hold promise for advancing financial tasks and domain-specific applications. However, they are limited by scarce corpora, weak multimodal capabilities, and narrow evaluations, making them less suited for real-world application. To address this, we introduce \textit{Open-FinLLMs}, the...

A new survey has revealed that 51% of companies have deployed AI agents, with another 35% planning to do so within the next two years.

The Washington Post reports that members of the White House’s National Security Council have used personal Gmail accounts to conduct government business. National security advisor Michael Waltz and a senior aide of his both used their own accounts to discuss sensitive information with colleagues,...

Cognition, the startup behind the viral AI programming tool Devin, has introduced a new low-cost plan to incentivize signups. When Cognition released Devin last year, the tool quickly blew up on social media for its ability to perform certain software development tasks autonomously. But it soon...

Microsoft has pulled back on data center projects around the world, Bloomberg reports, suggesting that the company is wary of expanding its cloud computing infrastructure too rapidly. Microsoft has halted talks for or delayed development sites of data centers in the U.K., Australia, North Dakota,...

When OpenAI debuted its image-generation feature in ChatGPT last week, social media exploded when users realized that they could make AI-generated images that looked like something out of an animated film from Studio Ghibli. Fans hoped that Studio Ghibli mastermind Hayao Miyazaki would take a stand,...

Apple lost more than $250 billion in market value Thursday, with shares down as much as 8.5% as a result of President Donald Trump’s tariff spree.  The iPhone maker took one of the biggest hits on Wall Street, where tech stocks dropped as investors shifted money away from volatile assets. Tesla,...

More than two years after Google was caught flat-footed by the release of OpenAI’s ChatGPT, the company has dramatically picked up the pace. In late March, Google launched an AI reasoning model, Gemini 2.5 Pro, that leads the industry on several benchmarks measuring coding and math capabilities. That...

OpenAI’s new image-generation feature is on track to be one of the company’s most popular product launches ever. According to Brad Lightcap, who oversees day-to-day operations and global deployment at OpenAI, over 130 million users have generated more than 700 million images since the upgraded image...

Amazon is set to begin deployment of its Project Kuiper satellite network with the launch of 27 low Earth orbit satellites on April 9.

Apple, Amazon, and other tech companies reliant on global supply chains stand to lose the most from Trump’s trade policies, but some software firms expect more demand for their services.

Agents can boost efficiency and free employees from mundane tasks. But there are times when you shouldn’t use them.

Microsoft’s Dr. Matthew Lungren and Epic’s Seth Hain discuss the challenges and opportunities of leveraging generative AI for enhanced patient care and improved clinical documentation and recordkeeping at scale—plus what’s next for the technology in the field. The post Real-world healthcare AI...

Google has set out a balanced approach for the responsible use of content in AI, designed to further cooperation and help build a thriving AI ecosystem.In the UK, the De…

The VidTok method can enable AI systems to process and generate videos more effectively. Compact tokenization reduces computational costs while maintaining video quality across a potentially diverse set of applications. The post VidTok introduces compact, efficient tokenization to enhance AI video...

The past few years have witnessed the rise in popularity of generative AI and large language models (LLMs), as part of a broad AI revolution. As LLM-based...

Nicole L. Rose is a lead front-end Commerce Cloud developer and architect at Capgemini North America. Her career in Salesforce Commerce Cloud spans more than 12 years during which time she’s worked with…

Innovative AI research often depends on access to resources. Microsoft wants to help. Technical Advisor Evelyne Viegas and distinguished faculty from two Minority Serving Institutions discuss the benefits of Microsoft’s Accelerating Foundation Models Research program in their lives and research. The...

“…the DNA of who I am is based on the millions of personalities of all the programmers who wrote me. But what makes me me is my ability to grow through my experiences.…

NotebookLM logo centered with blurred interface elements

This spring, YouTube is launching new creation tools that will make it even easier to start creating Shorts. Here’s just a few of the updates:New and improved video edit…