Latest

The recent emergence of Large Vision-Language Models(VLMs) has resulted in a variety of different benchmarks for evaluating such models. Despite this, we observe that most existing evaluation methods suffer from the fact that they either require the model to choose from pre-determined responses,...

Recently, model merging methods have demonstrated powerful strengths in combining abilities on various tasks from multiple Large Language Models (LLMs). While previous model merging methods mainly focus on merging homogeneous models with identical architecture, they meet challenges when dealing with...

MLLM reasoning has drawn widespread research for its excellent problem-solving capability. Current reasoning methods fall into two types: PRM, which supervises the intermediate reasoning steps, and ORM, which supervises the final results. Recently, DeepSeek-R1 has challenged the traditional view that...

Large real-world robot datasets hold great potential to train generalist robot models, but scaling real-world human data collection is time-consuming and resource-intensive. Simulation has great potential in supplementing large-scale data, especially with recent advances in generative AI and...

World foundation models, which simulate the physical world by predicting future states from current observations and inputs, have become central to many applications in physical intelligence, including autonomous driving and robotics. However, these models require substantial computational resources...

We propose Rec-R1, a general reinforcement learning framework that bridges large language models (LLMs) with recommendation systems through closed-loop optimization. Unlike prompting and supervised fine-tuning (SFT), Rec-R1 directly optimizes LLM generation using feedback from a fixed black-box...

This paper presents StreamChat, a novel approach that enhances the interaction capabilities of Large Multimodal Models (LMMs) with streaming video content. In streaming interaction scenarios, existing methods rely solely on visual information available at the moment a question is posed, resulting in...

Hallucination, a phenomenon where large language models (LLMs) produce output that is factually incorrect or unrelated to the input, is a major challenge for LLM applications that require accuracy and dependability. In this paper, we introduce a reliable and high-speed production system aimed at...

We evaluated 4 systems (ELIZA, GPT-4o, LLaMa-3.1-405B, and GPT-4.5) in two randomised, controlled, and pre-registered Turing tests on independent populations. Participants had 5 minute conversations simultaneously with another human participant and one of these systems before judging which...

Recent advances in view synthesis and real-time rendering have achieved photorealistic quality at impressive rendering speeds. While Radiance Field-based methods achieve state-of-the-art quality in challenging scenarios such as in-the-wild captures and large-scale scenes, they often suffer from...

Large language models (LLMs) demonstrate impressive capabilities in mathematical reasoning. However, despite these achievements, current evaluations are mostly limited to specific mathematical topics, and it remains unclear whether LLMs are genuinely engaging in reasoning. To address these gaps, we...

The detection of telecom fraud faces significant challenges due to the lack of high-quality multimodal training data that integrates audio signals with reasoning-oriented textual analysis. To address this gap, we present TeleAntiFraud-28k, the first open-source audio-text slow-thinking dataset...

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by retrieving relevant documents from external sources and incorporating them into the context. While it improves reliability by providing factual texts, it significantly increases inference costs as context length grows and...

There are a thousand ways to caption an image. Contrastive Language Pretraining (CLIP) on the other hand, works by mapping an image and its caption to a single vector -- limiting how well CLIP-like models can represent the diverse ways to describe an image. In this work, we introduce Llip, Latent...

Recognition of handwritten mathematical expressions allows to transfer scientific notes into their digital form. It facilitates the sharing, searching, and preservation of scientific information. We introduce MathWriting, the largest online handwritten mathematical expression dataset to date. It...

End-to-end models capable of handling multiple sub-tasks in parallel have become a new trend, thereby presenting significant challenges and opportunities for the integration of multiple tasks within the domain of 3D vision. The limitations of 3D data acquisition conditions have not only restricted...

The power of large language models (LLMs) has been demonstrated through numerous data and computing resources. However, the application of language models on mobile devices is facing huge challenge on the computation and memory costs, that is, tiny language models with high performance are urgently...

The key-value (KV) cache accelerates LLMs decoding by storing KV tensors from previously generated tokens. It reduces redundant computation at the cost of increased memory usage. To mitigate this overhead, existing approaches compress KV tensors into lower-bit representations; however, quantization...

Parameter generation has emerged as a novel paradigm for neural network development, offering an alternative to traditional neural network training by synthesizing high-quality model weights directly. In the context of Low-Rank Adaptation (LoRA) for evolving ($\textit{i.e.}$, constantly updated)...

This study presents a systematic comparison of three approaches for the analysis of mental health text using large language models (LLMs): prompt engineering, retrieval augmented generation (RAG), and fine-tuning. Using LLaMA 3, we evaluate these approaches on emotion classification and mental health...

Led by a former OpenAI executive, Amazon’s AI lab focuses on the decision-making capabilities of next generation of software agents—and borrows insights from physical robots.

Amazon on Monday unveiled Nova Act, a general-purpose AI agent that can take control of a web browser and independently perform some simple actions. Alongside the new agentic AI model, Amazon is releasing the Nova Act SDK, a toolkit that allows developers to build agent prototypes with Nova Act. Nova...

Creatives in the UK are once again speaking out against AI developers accessing copyrighted material. The Society of Authors have published an open letter calling for UK Secretary of State Lisa Nandy to hold Meta accountable for possible copyright infringement regarding its LLM, Llama 3. Signatories...

Amazon has delivered on its promise to launch an upgraded Alexa experience, Alexa+, by the end of March. But the service reportedly lacks many of the features Amazon demoed at a press event in late February. Alexa+ can’t yet perform certain “agentic” AI tasks showcased in the company’s marketing,...

There may come a point where you want to break ties with Amazon — maybe you’re cutting down on impulse buys, trying to escape the subscription rabbit hole or just rethinking where your money goes. Whatever the reason, Amazon gives you two main options: you can cancel your Prime membership to stop the...

How can we push CPUs forward? That’s the question the computing industry has been asking since the Intel 4004 processor launched in 1971. Chipmakers have tried cranking up clock speeds, adding multiple cores and miniaturizing chip architectures to make them faster and more efficient. The conflict...

An AI overhaul may be on the horizon for Apple’s Health app. In the Power On newsletter, Bloomberg’s Mark Gurman reports that Apple is working on a much more comprehensive version of its Health app under the code name Project Mulberry, with plans to integrate an AI agent that would somewhat...

Meta’s VP of AI research, Joelle Pineau, is planning to leave the company, she announced in a post on Facebook Tuesday. Pineau said she’s leaving in May after more than two years overseeing FAIR, Meta’s internal AI research lab led by Yann LeCun. Pineau’s exit comes as Meta ramps up its AI efforts,...

In a series of posts on X on Monday, OpenAI CEO Sam Altman said that the popularity of the company’s new image-generation tool in ChatGPT will cause unspecified product delays. “We are getting things under control, but you should expect new releases from OpenAI to be delayed, stuff to break, and for...

Google on Tuesday announced a new partnership with gaming company Roblox, which will allow advertisers to purchase and scale Roblox’s Rewarded Video and other immersive ad formats. That means marketers who want to reach the younger Gen Z audience that dominates the platform will be able to use Google...

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...

Today, Google announced new short-term precipitation forecasts are now available across Africa in Search. This expands upon our work to use machine learning to tackle th…

All Google Workspace for Education users have added data protection in the Gemini app, meaning their chats with Gemini are not human reviewed or used to train AI models.…

Immersive Ad showing on a billboard in a game.

A screenshot of a digital presentation tool with a slide containing four steps listed.

Prep for Small Business Week with us! Tips, tools, and using AI to prepare for a great week for growing businesses this May.

A group of people at an event standing in front of a sign that says Decoded Futures

This week, we released Gemini 2.5 Pro, our most intelligent model yet. On the latest episode of the Release Notes podcast, host Logan Kilpatrick goes on a deep dive with…

We are thrilled to announce that Juan Carlos Niebles, one of our talented researchers, has been named one of the top 100 in AI by the Artificial Intelligence Observatory in Colombia. This recognition…