The day's AI news, briefed
The Daily Prompt
Our AI editor reads each day's AI and robotics stories, pulls out why they matter, and links every original source — so you can keep up in minutes.
Join over 250,000 subscribers getting the latest AI news
Latest AI & Robotics News Today
Part-X-MLLM: Part-aware 3D Multimodal Large Language Model
This development in 3D multimodal large language models could significantly enhance capabilities in generating and editing complex 3D structures efficiently.
- Part-X-MLLM is a 3D multimodal large language model that formulates diverse 3D tasks as programs in a structured grammar.
- The model generates coherent token sequences that encode part-level bounding boxes, semantic descriptions, and edit commands from RGB point clouds and natural language prompts.
- By decoupling symbolic planning from geometric synthesis, the model allows control of various geometry engines through a language-native interface.
- Pre-training on a large-scale, part-centric dataset enhances the model's ability to produce structured plans and excel in tasks such as grounded Q&A and compositional generation.
Part-X-MLLM utilizes innovative techniques to unify 3D task execution, enhancing the integration of AI in 3D modeling and design. This model represents a significant step in the development of advanced AI systems capable of understanding and manipulating 3D data. Future developments may include more sophisticated applications in design, robotics, and gaming.
Read the original ↗
We’re expanding our presence in Singapore to advance AI in the Asia-Pacific region
DeepMind's new research lab in Singapore aims to enhance AI development and collaboration in the Asia-Pacific region, benefitting diverse communities and driving real-world applications.
- Google DeepMind is opening a new research lab in Singapore to expand its AI presence in the Asia-Pacific region.
- The lab will focus on advancing AI technology through collaboration with governments, businesses, and academic institutions.
- This expansion builds on a previous increase in Google DeepMind's APAC team, which has more than doubled in the past year.
- The lab will contribute to projects in various sectors, including health, public services, and education, while promoting inclusivity in AI research.
Google DeepMind is a subsidiary of Alphabet Inc. focused on artificial intelligence and machine learning. Previously, it has collaborated on various local initiatives in Singapore, particularly in health and education. Future developments will likely focus on enhancing AI applications aligned with the Singapore government's strategies.
Read the original ↗
Google’s Sundar Pichai Urges Caution When Using AI Tools
Sundar Pichai's caution about AI tools highlights the need for users to critically evaluate AI-generated information, which is essential for ensuring accuracy and safety in various fields.
- Sundar Pichai advised users not to fully trust AI responses, noting that AI can produce incorrect or misleading information.
- He emphasized that AI should be considered one resource among many, not as a final authority.
- Research shows that various AI chatbots frequently generate major errors when summarizing news, exposing industry-wide challenges.
- Pichai highlighted Google's commitment to user safety as it develops new AI products, including tools for detecting AI-generated content.
Google, a leader in AI technology, has introduced various AI tools aimed at enhancing user interaction with its search engine. The company has faced scrutiny following public errors in its AI systems, raising concerns about the reliability of AI-generated information. Looking ahead, continued advancements in AI will likely bring both innovation and challenges related to accuracy and responsibility.
Read the original ↗
Gemini 3 is Here! The Most Powerful AI Model Out There
The entry of Gemini 3 into the AI space signals a significant advancement in capabilities, particularly in multimodal reasoning and task completion, which could influence various applications across industries.
- Gemini 3 is Google’s next-generation AI model, capable of understanding text, images, audio, video, and code.
- The model enhances reasoning abilities, memory handling, and user intent awareness compared to previous versions.
- Gemini 3 is integrated into several Google products, including Search in AI Mode, Google AI Studio, and Vertex AI, promoting broader usability.
- It achieves notable performance improvements, scoring 91.9 percent on the GPQA Diamond benchmark, indicating strong reasoning capabilities.
- Developers can access Gemini 3 through APIs and tools like Gemini CLI, facilitating the creation of applications and automated workflows.
Gemini 3 is a multimodal AI model developed by Google, designed to unify its various AI capabilities in one framework. The model enhances user interactions, making AI feel more like a collaborator than a simple tool. Observers will be watching how companies integrate and leverage this advanced AI in real-world applications.
Read the original ↗
xAI’s Grok 4.1 Pushes Toward Higher Emotional Intelligence, Lower Hallucinations and Tighter Safety Controls
Improvements in emotional intelligence and safety in AI could lead to more effective and reliable interactions between human users and AI systems, enhancing user experience and trust.
- xAI has introduced Grok 4.1, an advanced model aimed at increasing emotional intelligence in AI.
- The update also focuses on reducing hallucinations, which refer to false information generation by AI.
xAI, founded by Elon Musk, aims to develop advanced AI technologies. Previous iterations of Grok have focused on enhancing conversational capabilities, and this latest version addresses critical issues such as safety and reliability in AI interactions.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Microsoft’s Agent 365 Tries to Be the AI Bot Boss
Agent 365 enables businesses to efficiently manage AI agents, addressing potential security risks as organizations increasingly rely on automated processes in their workflows.
- Microsoft's Agent 365 is a new tool designed for businesses to manage their AI agents as if they were human employees.
- The tool provides a registry for tracking active agents, their roles, and permissions within a company's software.
- Agent 365 includes security features that monitor agent activities in real-time to protect sensitive data.
Microsoft's Agent 365 reflects the growing trend of integrating AI agents into the workplace, with the aim of helping organizations manage increasing numbers of automated tools. This development comes amid rising interest in generative AI, prompting companies to explore its applications while addressing potential security challenges. Future adoption will likely focus on optimizing the balance between human workers and AI agents.
Read the original ↗
Hugging Face CEO says we’re in an ‘LLM bubble,’ not an AI bubble
Understanding the dynamics of the LLM market can influence investment strategies and development priorities in the broader AI industry.
- Clem Delangue, CEO of Hugging Face, claims we are in an 'LLM bubble' distinct from an overall AI bubble.
- Delangue anticipates that the LLM bubble may burst next year, but AI's future remains secure despite this possibility.
- He suggests that specialized, smaller models will likely gain traction for specific applications, such as banking customer chatbots.
Hugging Face is known for its AI platform that supports LLMs and various AI applications. Delangue emphasizes a shift toward customized solutions rather than one-size-fits-all models. The company operates with a capital-efficient strategy in contrast to many competitors spending billions on AI development.
Read the original ↗
AI data center provider Lambda raises whopping $1.5B after multibillion-dollar Microsoft deal
This significant funding round positions Lambda as a strong competitor in the AI data center market, potentially affecting the landscape of AI infrastructure services and partnerships.
- Lambda raised $1.5 billion in funding, led by TWG Global, which has a focus on AI investments.
- The new funding follows a multibillion-dollar deal between Lambda and Microsoft for AI infrastructure, including Nvidia GPUs.
- Lambda is a competitor to CoreWeave and sells its services to hyperscaler clouds.
- Previously, Lambda raised $480 million in February, with a valuation of approximately $2.5 billion.
Lambda provides AI data center services and competes with CoreWeave, focusing on supplying infrastructure for AI applications. The company recently secured a large contract with Microsoft, which enhances its market position and investor interest. Future developments may include an increased valuation or an initial public offering (IPO).
Read the original ↗
Tech giants pour billions into Anthropic as circular AI investments roll on
This partnership signals a strategic shift in the AI market aimed at reducing reliance on major players, influencing competition and technology innovation in AI development.
- Microsoft and Nvidia have announced a partnership investing a combined total of up to $15 billion in Anthropic.
- The deal focuses on using Microsoft’s cloud services and Nvidia’s chips to enhance Anthropic's AI models.
- Under the agreement, Anthropic will produce up to 1 gigawatt of computing power using Nvidia hardware.
Anthropic is an AI company and competitor to OpenAI, known for its Claude AI models. The partnership reflects a broader trend of tech companies diversifying their AI investments and capabilities in light of changing market dynamics.
Read the original ↗
Microsoft tries to head off the “novel security risks” of Windows 11 AI agents
The introduction of AI agents in Windows 11 raises significant privacy and security concerns that users must navigate, particularly regarding access to sensitive data.
- Microsoft has released a new Windows 11 build with experimental AI agents capable of performing tasks like organizing files and scheduling meetings.
- These agents will operate under separate user accounts to limit their permissions and will require user approval for data access.
- Microsoft warns that the AI agents could introduce novel security risks, such as hijacking by attackers that could lead to unauthorized actions.
- Users can toggle off these features in Settings, with the option set to off by default, allowing for more user control over AI functionalities.
Microsoft has been integrating AI features into Windows 11, with recent enhancements aimed at improving user productivity through background task management. Previously, the company faced criticism for security lapses, prompting more robust precautions in this latest release. The response to these security challenges will be important as AI technologies continue to evolve in consumer software.
Read the original ↗
Lost for 3,000 Years, an Ancient Babylonian Hymn Has Been Restored Using Artificial Intelligence
The use of artificial intelligence to restore an ancient Babylonian hymn highlights the technology's potential in recovering lost cultural heritage and enhancing understanding of historical societies.
- Researchers used artificial intelligence to restore a 3,000-year-old Babylonian hymn from cuneiform tablets.
- The hymn offers insights into daily life, social roles, and environmental themes in ancient Babylon.
- Led by Professor Enrique Jiménez, the team identified 30 similar manuscripts using machine learning to reconstruct the text.
Cuneiform, an ancient Mesopotamian writing system, has fragmented records due to centuries of dispersal. This project is part of ongoing efforts to digitize and study ancient texts, helping to illuminate historic cultures.
Read the original ↗
Cloudflare Reports “Spike in Unusual Traffic” as Outage Limits Access to ChatGPT, X, and Other Major Sites
The Cloudflare outage disrupted access to major online platforms, highlighting the vulnerabilities of internet infrastructure that many businesses rely on for daily operations.
- Cloudflare reported a global service outage that began around 6:40 a.m. ET on Tuesday, impacting platforms such as ChatGPT, X, and Spotify.
- The outage followed a detected spike in unusual traffic on Cloudflare's network, although the company has not confirmed the source of the traffic surge.
- Scheduled maintenance activities coincided with the outage, but Cloudflare has not linked these maintenance operations to the service disruption.
- Cloudflare stated that the outage does not currently appear to be linked to a cyber-attack or malicious activity.
- Recent similar outages, including one at Amazon Web Services, have raised concerns about potential links to security threats.
Cloudflare provides crucial security and content delivery services for approximately 20% of global web traffic. The company is currently investigating the causes of the unusual traffic spike, with no confirmed links to malicious activity at this time. This incident reflects ongoing concerns about the resilience of internet infrastructure in the face of increasing disruptions.
Read the original ↗
Poe’s AI app now supports group chats across AI models
The introduction of group chat functionality in Poe's AI app allows for collaborative engagements across various AI models, enhancing user interaction and productivity.
- Poe, Quora's AI app, now supports group chats with up to 200 participants.
- Users can collaborate using over 200 different AI models in a single conversation.
- The group chat feature allows real-time chat history synchronization across devices.
Poe integrates various AI models into one platform, enabling diverse interactions. This update follows a similar group chat feature launch by OpenAI’s ChatGPT. Future enhancements to group chats will be based on user feedback.
Read the original ↗
Google unveils Gemini 3 AI model and AI-first IDE called Antigravity
The development of Gemini 3 and the Antigravity IDE showcases significant advancements in AI capabilities that could fundamentally change programming and project management.
- Google has launched its upgraded AI model, Gemini 3 Pro, which offers enhanced reasoning and visual outputs available in a limited form today.
- The new AI-first integrated development environment, Antigravity, is also available for testing and is designed to streamline coding tasks.
- Gemini 3 Pro scored record results in various benchmarks, including a 72.1% on the SimpleQA Verified test and notable improvements in coding assessments.
- The generative interface of Gemini 3 allows for more interactive and visually appealing outputs, enhancing user experience through features like visual layout and dynamic view.
Gemini is Google's flagship AI model, with prior versions like Gemini 2.5 being integrated across its products. The introduction of Gemini 3 and Antigravity highlights Google's push towards more sophisticated AI tools in development and user assistance.
Read the original ↗
Google CEO: If an AI bubble pops, no one is getting out clean
Concerns about an AI investment bubble highlight potential risks for companies in the technology sector, affecting market stability and investor confidence.
- Sundar Pichai warns that no company, including Alphabet, will be immune if the AI market experiences a downturn.
- He compared the current AI investment climate to the late 1990s dotcom boom, hinting at possible future challenges.
- Pichai emphasized the importance of critically assessing AI outputs, encouraging users not to trust AI-generated information blindly.
- Despite these concerns, Pichai described AI as a profound technology that will evolve various jobs and create new opportunities.
Alphabet is the parent company of Google and has experienced significant growth in its market value, primarily due to investments in AI technologies. Industry analysts are increasingly skeptical about the sustainability of these investments, particularly regarding competitors like OpenAI. Observers are watching for signs of market correction and its impacts on technology firms.
Read the original ↗
Researchers Trace How a Psychedelic Compound Alters Brain Circuits Tied to Addiction
Understanding how psilocin affects brain circuits linked to addiction could inform new treatment strategies for alcohol use disorder and improve emotional regulation in individuals with addiction.
- Researchers at the University of North Carolina discovered how psilocin reduces alcohol intake by dampening activity in stress-sensitive neurons in the central amygdala.
- The study involved female mice, which typically consume more alcohol than males, and found that psilocin calmed overactive neurons associated with stress and drinking.
- Mice administered psilocin showed reduced alcohol consumption, linking neural activity changes directly to behavioral shifts during the effects of the compound.
Psilocin is the active metabolite of psilocybin, found in magic mushrooms, impacting serotonin receptors to potentially aid in addiction treatment. Prior research indicated psilocybin's effects on emotional regulation, although biological mechanisms remained unclear. Future studies will explore gender differences and the duration of psilocin's effects.
Read the original ↗
Give Your Back a Break With Our Favorite Office Chairs
Investing in a high-quality ergonomic chair can significantly enhance comfort and productivity for individuals working from home, potentially reducing discomfort associated with prolonged sitting.
- The Branch Ergonomic Chair Pro is recommended for its balance of value, comfort, and adjustability, priced at $499.
- WIRED's reviews team has tested over 60 chairs to compile recommendations tailored for various body types and budgets.
- The Steelcase Gesture, priced at $1,510, is highlighted for its comfort and fluid adjustability, catering to diverse sitting positions.
Ergonomic chairs provide necessary lumbar support and customizable adjustments to prevent discomfort during long work hours. Many consumers working from home are increasingly prioritizing comfort, leading to a rise in demand for quality office furniture. Future innovations in chair design may focus on enhanced ergonomics and materials to improve user experience.
Read the original ↗
Stack Overflow is remaking itself into an AI data provider
Stack Overflow's transition into an AI data provider reflects a significant shift in how tech platforms can monetize their data and expertise, impacting enterprise AI development.
- Stack Overflow aims to evolve its problem-solving forum into a tool for translating human expertise into an AI-accessible format.
- The new enterprise product, Stack Internal, will include security features tailored for enterprise environments.
- The company has established content deals with AI labs to allow model training on public Stack Overflow data for a fee.
- Stack Internal will provide a layer of metadata with reliability scores to enhance the trust of answers used by AI agents.
Stack Overflow is a community-driven Q&A platform primarily focused on programming and software development. The company seeks to capitalize on its existing data by providing tools that enhance enterprise AI capabilities. Future developments may include advanced functionality for AI agents to autonomously generate queries.
Read the original ↗
P1: Mastering Physics Olympiads with Reinforcement Learning
The development of the P1 models signifies a key advancement in leveraging reinforcement learning to enhance problem-solving capabilities in physics education, potentially transforming competitive learning and assessment methodologies.
- The P1 family includes open-source physics reasoning models trained through reinforcement learning that excel in solving Olympiad-level problems.
- The P1-235B-A22B model achieved gold medal performance at the International Physics Olympiad (IPhO) 2025 and won 12 golds out of 13 competitions in 2024/2025.
- P1-30B-A3B model received a silver medal at IPhO 2025, surpassing most other open-source models in performance.
- P1 models also demonstrate strong capabilities in other reasoning tasks such as mathematics and coding, indicating their versatility.
- The release includes a full-stack open-source ecosystem comprising models, algorithms, benchmarks, and agent frameworks.
Large language models (LLMs) have evolved to solve complex scientific reasoning tasks, particularly in physics, which requires a deep understanding of concepts. The introduction of the P1 models aims to enhance physics research and education, benefiting competitive problem-solving contexts. Future developments may focus on expanding their application across various reasoning tasks and educational frameworks.
Read the original ↗
TiViBench: Benchmarking Think-in-Video Reasoning for Video Generative Models
This new benchmarking framework is crucial for advancing the reasoning capabilities of video generative models, which could enhance their practical applications in various fields.
- TiViBench is a new benchmark designed to evaluate reasoning capabilities of image-to-video generation models.
- It assesses reasoning across four dimensions: structural reasoning, spatial and visual pattern reasoning, symbolic and logical reasoning, and action planning.
- Commercial models like Sora 2 and Veo 3.1 show strong reasoning potential, while open-source models are limited by training scale and data diversity.
TiViBench addresses a gap in current benchmark evaluations that primarily focus on visual fidelity and coherence. It aims to measure higher-order reasoning abilities that can lead to better performance in video generation tasks.
Read the original ↗