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
Former TikTok execs built an app that uses AI to teach you how to pose for a photo
This app addresses a common problem related to photography, using AI to enhance users' posing skills and improve their photo quality.
- Superpose, a new iOS app developed by former TikTok employees, uses AI to teach users how to pose for photos.
- The app generates four potential poses based on selfies or photos taken by the user, offering a unique approach to portrait photography.
- Launched in July, Superpose has been downloaded over 22,000 times and users have generated more than 190,000 poses.
- The app allows five free pose generations daily, with additional poses available for purchase at $2.99 for five or $9.99 for twenty.
- Superpose has raised $2.2 million in funding from Khosla Ventures, Meitu, and OVTR VC.
Superpose is built by Melody Chu and Jing Liu, who previously worked at TikTok, aiming to improve the art of posing for photographs using insights from computer vision and generative AI. This initiative aligns with broader trends in AI-driven photography tools, as seen in previous releases by Google and Adobe. Future improvements are expected to enhance personalization and generation accuracy in pose suggestions.
Read the original ↗
AEO startup Profound hits unicorn valuation, raises $180M Series D 7 months after last round
Profound's significant funding round reflects growing investment in companies leveraging AI for marketing, underscoring shifts in how brands engage with consumers in digital spaces.
- Profound has raised $180 million in a Series D funding round, achieving a valuation of $1.8 billion.
- This funding comes less than seven months after its $96 million Series C round.
- Sequoia and Kleiner Perkins led the latest funding round, with participation from existing investors including Lightspeed Venture Partners and Khosla Ventures.
- Profound, founded two years ago, initially focused on analytics but now helps businesses optimize their marketing strategies for AI-driven consumer searches.
- The company reports a threefold revenue increase in the past six months and serves over 1,000 enterprise customers, including Comcast and Walmart.
Profound develops marketing software aimed at enhancing brand visibility in AI search results. It operates in the generative engine optimization (GEO) and answer engine optimization (AEO) field, where businesses strive to adapt to AI tools increasingly used in consumer searches. Observers should monitor how this investment trend affects market competition and innovation in the AI marketing space.
Read the original ↗
OpenAI, Anthropic, Google have been in talks on AI safety for weeks
Collaboration among leading AI companies on safety measures is critical to address potential risks and ensure responsible development of AI technologies, especially amid ongoing regulatory debates.
- OpenAI, Anthropic, and Google DeepMind have been in talks for several weeks regarding AI safety issues.
- The discussions follow the call for collective action against potential catastrophic risks made by Anthropic CEO Dario Amodei.
- OpenAI's policy chief Chris Lehane acknowledged these talks while addressing U.S. lawmakers about the risks associated with AI.
OpenAI is a prominent artificial intelligence research lab that develops advanced AI models. Recent calls for regulation and safety in the AI industry have gained momentum, prompting discussions among leading tech firms. Future developments are expected regarding the formation of an industry standards body for safety monitoring.
Read the original ↗
BVB: Benchmarking Agentic Video Understanding via Programmatic Reconstruction in Blender
This benchmark addresses the limitations of current video understanding models by evaluating their ability to reconstruct real-world videos, highlighting significant gaps in spatiotemporal accuracy.
- BVB, or Blender-VideoBench, requires agents to programmatically reconstruct real-world videos as animated scenes in Blender.
- Current models show high perceptual similarity to source videos but struggle to maintain factual accuracy regarding spatiotemporal details.
- The benchmark evaluates performance using two key measures: Dual VQA for preserving spatiotemporal facts and Latent Similarity for perceptual matching.
- 51 configurations from 10 different model families were evaluated, with the top model achieving an 88.6 score in Latent Similarity but only 53.7% retention of source facts.
Blender is a popular open-source 3D creation suite used for various types of animation and video rendering. The traditional evaluation of video understanding models has often relied on question-answering metrics rather than on reconstruction ability. The introduction of benchmarks like BVB is intended to enhance understanding of how agents process video content and improve their reasoning skills.
Read the original ↗
PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models
The introduction of PhysBrain 1.5 represents a significant advancement in embodied AI, enhancing capabilities in physical environment understanding and action generation, which could impact various robotics applications.
- PhysBrain 1.5 combines physical environment understanding, action generation, and future state prediction into a single model.
- It utilizes autoregressive training on sequences of vision-language, motion, and visual targets, achieving state-of-the-art open-source performance.
- The model does well across 28 benchmarks, achieving an average score of 72.5, on par with proprietary models like GPT-6-Astra.
- PhysBrain 1.5 is trained using human interaction videos and is fine-tuned with various demonstrations and simulations.
PhysBrain 1.5 is an embodied foundation model designed for robotics, focusing on how machines understand and interact with their physical environments. Previous models lacked this comprehensive approach, limiting their ability to predict future interactions. This model's performance sets a new benchmark for open-source projects in the field.
Read the original ↗
ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search
ZGCM-1's development represents a significant advancement in open-source foundation models, emphasizing efficiency and autonomous operations, which could influence how AI tools are developed and utilized in the future.
- ZGCM-1 is a fully open 7 billion parameter dense foundation model trained to enhance reasoning and efficiency through innovative design and training methods.
- The model demonstrates competitive performance against much larger architectures in mathematical reasoning and agentic search tasks.
- ZGCM-1 employs an advanced training strategy including progressive context scaling and the use of agent swarms for streamlined operations and data curation.
- Extensive evaluations indicate that ZGCM-1 achieves a 4.2 times improvement in efficiency during pre-training phases compared to prior techniques.
ZGCM-1 is part of a growing interest in enhanced artificial intelligence models that integrate reasoning with tool use. Foundation models are increasingly seen as crucial for broad AI applications, and ZGCM-1's open-source approach may facilitate further research and development in this field. Future developments will likely focus on enhancing the model's capabilities and efficiency.
Read the original ↗
Atria Dawn: The Dawn of Agentic Superintelligence
The development of agentic language models like Atria Dawn could fundamentally transform how human researchers collaborate with AI, enhancing productivity and shaping the future of scientific research.
- Atria Dawn Preview is a foundation agentic language model aimed at improving scientific research and engineering workflows.
- The model was trained through verified tool interactions and is competitive across 16 benchmarks, achieving the highest scores on five.
- Human researchers rated one-third of AI-assisted tasks as infeasible without AI support, indicating a shift to project-level collaboration.
- Agents not only executed tasks but also proposed methods and revisions, while humans maintained final decision-making authority.
Atria Dawn is an advance in AI language models focusing on collaborative capabilities in research environments. The model represents ongoing efforts to integrate AI agents more actively into scientific workflows, highlighting a trend toward deeper human-AI partnerships.
Read the original ↗
OpenEvidence and Penn Medicine Launch Global Clinical AI Partnership
This partnership aims to enhance clinical decision-making for over 10,000 clinicians, potentially improving patient care in both the U.S. and low- to middle-income countries.
- OpenEvidence and Penn Medicine have launched a global partnership to provide AI-powered clinical decision support to clinicians across Penn Medicine's seven-hospital health system.
- The partnership also extends support to low- and middle-income countries through the Botswana-UPenn Partnership, aiming to deliver personalized, evidence-based insights from medical literature.
- OpenEvidence's tools will integrate with existing programs at Penn Medicine, enhancing access to specialized care, particularly for underserved communities.
OpenEvidence is a widely used medical AI tool for U.S. physicians, while Penn Medicine operates a health system with significant outreach, especially into global health initiatives. The Botswana-UPenn Partnership, established 25 years ago, focuses on enhancing healthcare in Botswana through training and research efforts.
Read the original ↗
Anthropic Releases Salesforce in Claude Plugin With 37 Sales Skills
The release of the Salesforce in Claude plugin streamlines sales processes for organizations, allowing sales teams to efficiently manage accounts and opportunities, thereby enhancing productivity.
- Anthropic announced the beta release of the Salesforce in Claude plugin, which integrates sales data into the Claude AI assistant.
- The plugin features 37 sales skills that aid account executives with tasks like call preparation and pipeline management.
- Sellers can summarize account histories, log calls, and create follow-up tasks seamlessly within Salesforce using the plugin.
- Claude generates personalized daily briefs that include meetings and at-risk opportunities, improving task management for sellers.
- The plugin is currently in beta and available for organizations approved by Salesforce.
Anthropic's Salesforce in Claude is designed to integrate Salesforce's customer relationship management system with AI capabilities. Previously, sales technology faced challenges with data fragmentation across platforms. This release is part of a strategic partnership between Anthropic and Salesforce, announced in August 2026, which aims to enhance sales productivity with AI tools.
Read the original ↗
Crusoe Signs Multi-Year Deal to Power Perplexity Training and Inference
The partnership enhances AI model training and inference efficiency, which could significantly impact companies relying on real-time AI insights.
- Crusoe has entered a multi-year partnership with Perplexity to run its model lifecycle on Crusoe Cloud.
- Perplexity will utilize dedicated NVIDIA GB300 NVL72 clusters for training frontier models and production serving.
- Crusoe’s Managed Inference service, launched on November 20, 2025, supports the high-performance requirements of Perplexity's operations.
Crusoe provides cloud computing infrastructure optimized for AI workloads, while Perplexity offers AI-driven information retrieval services. This partnership is expected to address the growing demand for efficient AI model training and real-time inference capabilities.
Read the original ↗
Cisco Brings Splunk AI On-Premises With New NVIDIA-Accelerated AI POD
The launch of Cisco AI POD for Splunk enables companies to implement AI solutions on-premises, addressing security and operational challenges associated with deploying enterprise AI.
- Cisco has introduced the AI POD for Splunk, which integrates self-managed Splunk AI into on-premises environments, leveraging an expanded partnership with NVIDIA.
- The AI POD uses a pre-validated architecture that combines Cisco infrastructure and NVIDIA computing for optimal Splunk AI workload management.
- Cisco's AI Tier software is designed for private cloud and air-gapped environments, allowing customers to maintain control over their data and operations.
- New functionalities include the Splunk AI Assistant and Agent Launchpad, facilitating ad-hoc investigations and custom agent creation for various use cases.
- A partnership with AWS aims to co-develop security solutions to combat AI-driven threats, enhancing operations for security analysts.
Cisco is a major player in networking technology, while Splunk provides data analytics software. Their collaboration looks to simplify AI deployment, aiming to address concerns over security and usability in enterprise environments. Observers should note the potential for these developments to influence AI applications in business operations.
Read the original ↗
Meta Launches Meta One Subscriptions Bundling AI Usage Across Its Apps
The introduction of Meta One marks a significant shift in how Meta monetizes its apps, providing new tiered AI functionalities that prioritize increased user engagement and creative capabilities.
- Meta launched Meta One, a subscription service offering paid tiers across Instagram, Facebook, WhatsApp, and Meta AI, as of September 15, 2026.
- The service includes over 50 features, with more than 15 million subscriptions and trials reported to date.
- Core bundles are priced at $7.99 per month and Premium at $19.99 per month for enhanced AI tools and content generation capabilities.
- Single-product plans are available, including Instagram Plus for $3.99 and WhatsApp Plus for $2.99 per month, with customization features to enhance user experience.
- Plans for creators and businesses start at $14.99 per month, providing professional tools integrated within existing apps without requiring separate accounts.
Meta is a technology company that owns and manages multiple social media platforms, including Instagram and Facebook. The Meta One subscription service represents its strategy to generate revenue by offering enhanced features leveraging AI technologies. Future developments may include more integrations and features for AI capabilities within their apps.
Read the original ↗
Physical AI Can See More Than Ever. Why Are Manufacturers Still Making Slow Decisions?
Manufacturers risk falling behind if they do not leverage advancements in Physical AI technology to improve decision-making processes across their operations.
- Physical AI technology enables real-time defect detection with cameras and identifies early equipment failure through sensors.
- Current manufacturing systems often operate independently, leading to fragmented data and delayed decision-making.
- A cohesive institutional intelligence layer is necessary to connect operational signals and facilitate faster, coordinated responses across systems.
Physical AI technology includes tools like cameras and sensors that enhance visibility in manufacturing environments. Despite having these technologies, manufacturers struggle to make timely decisions due to disconnected systems. Future efforts should focus on integrating these technologies to streamline decision-making processes.
Read the original ↗
From Uptime to Experience: The AI-Driven Shift in Modern Observability
The shift from traditional monitoring to AI-driven observability enhances system management effectiveness, potentially preventing larger IT failures by detecting and addressing issues proactively.
- IBM's 2001 manifesto on autonomic computing identified self-management as key, but technology had not yet evolved to support it.
- Observability, initially a concept from systems engineering, was adapted by Twitter for modern IT environments needing specialized monitoring tools.
- Experience Level Objectives (XLOs) focus on the quality of user experiences, unlike traditional Service Level Agreements (SLAs) which centered on uptime.
- AI tools can significantly reduce the number of alerts from thousands per day to only the most critical, improving response efficiency for IT teams.
- AI-driven observability platforms are positioned to fulfill the long-standing vision of autonomous IT management that past attempts struggled to achieve.
Observability refers to the ability to measure a system's internal state based on its outputs. This technology has evolved from early IT monitoring tools to sophisticated AI systems capable of managing large-scale architectures. The ongoing integration of AI in observability aims to enhance user experience and operational efficiency in IT environments.
Read the original ↗
IBM Research Proves Quantum Circuits Outperform LLMs on Two Problems
This research highlights the potential advantages of quantum circuits over large language models for specific computational tasks, signaling significant developments in comparing quantum and classical computing capabilities.
- IBM Research has demonstrated that shallow quantum circuits can theoretically outperform large language models (LLMs) on two types of problems: functional and sampling.
- The study focuses on an iterated index function problem, which shallow quantum circuits can solve with near-constant depth, unlike LLMs that require substantial computational resources.
- In the sampling category, quantum circuits efficiently predict the parity of unknown strings, outperforming diffusion language models even when enhanced with chain-of-thought capabilities.
Quantum circuits use quantum bits (qubits) to perform computations. This research builds upon earlier findings showing advantages of quantum circuits in specific tasks, while acknowledging current limitations in quantum technology compared to classical computing resources.
Read the original ↗
AI models need more data about biology, and OpenAI is paying to create it
This initiative by OpenAI underscores the increasing recognition that enhanced data access can significantly advance AI's ability to impact healthcare and drug development.
- The OpenAI Foundation is funding a project called Public Data for Health to create high-quality scientific datasets for medical AI.
- Ruxandra Teslo's proposal, which aims to use data from failed biotech companies, has received a $500,000 grant to better inform AI in the drug approval process.
- OpenAI will also provide $40 million for cancer vaccine data collection at the University of North Carolina and support drug effect prediction competitions.
- The foundation aims to donate $1 billion by year-end, focusing on grants to nonprofits and research institutions.
- 1Day Sooner, the organization pursuing Teslo's idea, has already acquired some datasets from bankrupt companies for AI training.
OpenAI, initially a nonprofit, has transitioned towards a for-profit model while maintaining a charitable foundation aimed at benefiting humanity. This new funding effort reflects a strategic push to overcome data limitations in medical AI, which some experts identify as a major hurdle to breakthroughs in drug development.
Read the original ↗
Artificial Intelligence (516)
Accessing free educational resources can significantly boost skills in data science and artificial intelligence, fostering career development and innovation in these fields.
- Microsoft is offering five free GitHub courses focused on data science and artificial intelligence.
- These courses aim to enhance learning in key areas of artificial intelligence and data science.
Microsoft, a leading technology company, provides educational resources to equip individuals with skills for the evolving job market. Previous initiatives have included various free online courses to expand knowledge in technology.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear
Salesforce's new Koa model addresses enterprise AI needs with enhanced reasoning capabilities, potentially shifting the competitive landscape in AI solutions for businesses.
- Salesforce introduced Koa, its first reasoning model, built on Nvidia's Nemotron model, designed for sales, marketing, and customer support.
- Koa offers an alternative to proprietary AI models, focusing on specific work tasks and ensuring no actual customer data is processed.
- The model reduces AI spending by using fewer tokens for tasks compared to leading models like Claude and ChatGPT.
- Post-training involved synthetic data to simulate customer scenarios, enhancing Koa's task performance and cost-efficiency.
- Salesforce remains partnered with Anthropic for additional AI solutions while emphasizing Koa's distinct capabilities.
Salesforce is a leading customer relationship management platform that now seeks to enhance its AI offerings with the introduction of Koa, a reasoning model. Previously reliant on external AI models for complex tasks, Salesforce's new model aims to fill this need with a proprietary solution. The demand for tailored AI applications in enterprises is rising, prompting further developments in this area.
Read the original ↗
Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents
As AI technology continues to advance, ensuring the safety and reliability of AI agents is crucial to prevent potential risks and abuses in various sectors.
- Artificial Intelligence Underwriting Company (AIUC) raised $40 million in a Series A funding round led by Ribbit Capital, bringing its total funding to $55 million.
- AIUC aims to apply a cybersecurity model to mitigate risks associated with rogue AI agents in enterprises, developing a certification standard called AIUC-1.
- The startup tests AI agents against 5,000 scenarios to assess their behavior, producing detailed reports on their safety and reliability.
Artificial Intelligence Underwriting Company was founded by Rune Kvist and Rajiv Dattani, who are addressing the rising concerns of AI safety in enterprises. The move follows increased calls from industry leaders for standardized evaluations to prevent AI misuse and accidents. Future developments may center around expanding the use of independent audits in AI deployment processes.
Read the original ↗
Exclusive: Paying for frontier AI models buys 4-month head start at 5x the cost
The accelerating adoption of cheaper open AI models can shift operational costs and efficiency for many businesses as competition among AI technology providers intensifies.
- The performance gap between US frontier AI models and Chinese open models has closed to 4.4 months.
- Open models, like Moonshot AI's Kimi K3, are significantly cheaper, costing only 30% of closed models like Anthropic's Fable 5.
- Many organizations are now opting for open models for routine tasks due to their cost-effectiveness and narrowing performance gap.
Frontier AI models are cutting-edge AI technologies from US companies that offer proprietary solutions, while open-weights models allow anyone to download and use the core components. The recent report from Mozilla highlights a significant trend: businesses are increasingly recognizing the value of open models as they replicate the capabilities of closed models at a much lower cost. Future developments will likely focus on maintaining competitiveness between US and Chinese AI offerings.
Read the original ↗