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
A former physician has launched Robyn, an empathetic AI companion
Robyn aims to address emotional disconnection by providing an empathetic AI assistant, which could reshape how users engage with technology for personal support.
- Robyn is an AI assistant developed by former physician Jenny Shao, designed to offer empathic and emotionally intelligent support.
- The app is not intended to replace therapists or friends but aims to assist users in understanding themselves better.
- Robyn employs a unique emotional memory system to track user interactions and offer personalized insights based on user responses.
- The app costs $19.99 monthly or $179 annually and has raised $5.5 million in seed funding.
- Robyn includes safety features, such as directing users to crisis resources if self-harm is mentioned.
Robyn represents a new approach to AI companionship, focusing on emotional intelligence rather than therapy or friendship. Previous AI companions have faced scrutiny over potential risks, including mental health crises. The app is now available in the U.S. after months of testing with select users.
Read the original ↗
Robot Learning from a Physical World Model
This development could streamline robotic training by enabling accurate manipulation based solely on generated videos, thus reducing reliance on physical data collection.
- PhysWorld is a framework that integrates video generation with physical world modeling for robotic learning.
- It generates task-conditioned videos from an image and command, reconstructing the physical world in the process.
- The method eliminates the need for real robot data and enables zero-shot generalizable manipulation.
- Experiments show that PhysWorld significantly enhances manipulation accuracy over previous methods.
PhysWorld represents a significant advancement in robotics by combining machine learning with physical world simulations. Previous approaches often required extensive real-world data collection, limiting training efficiency. Future developments will likely focus on expanding these capabilities for more complex tasks.
Read the original ↗
Meta AI Releases Omnilingual ASR: A Suite of Open-Source Multilingual Speech Recognition Models for 1600+ Languages
The release of open-source multilingual speech recognition models potentially enhances global communication and technology accessibility across diverse languages.
- Meta AI has released Omnilingual ASR, a suite of open-source multilingual speech recognition models.
- These models support over 1600 languages, offering a wider range for speech recognition tasks.
Meta is a technology company known for its innovations in AI. The Omnilingual ASR release represents a significant advancement in speech recognition technology, which has traditionally focused on fewer, more widely spoken languages. Future developments may include improvements to model accuracy and usability across various applications.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Chinese AI startup Moonshot outperforms GPT-5 and Claude Sonnet 4.5: What you need to know
The emergence of Moonshot as a competitor to established AI models could shift the landscape of artificial intelligence innovation and market dynamics.
- Chinese AI startup Moonshot has reportedly outperformed GPT-5 and Claude Sonnet 4.5.
- This development highlights the growing advancements in AI technology from startups worldwide.
Moonshot is a startup in China's rapidly evolving artificial intelligence sector, competing against advanced models like OpenAI's GPT-5 and Anthropic's Claude. The success of such startups may influence future technological developments and investments in AI.
Based on the headline only — the full article could not be fetched.
Read the original ↗
A Coding Implementation to Build and Train Advanced Architectures with Residual Connections, Self-Attention, and Adaptive Optimization Using JAX, Flax, and Optax
Advanced coding implementations can enhance the development of neural networks by improving their performance, which has implications for AI application efficiency.
- The article discusses a coding implementation for building and training advanced neural network architectures.
- It emphasizes features like residual connections, self-attention mechanisms, and adaptive optimization techniques.
JAX is a numerical computing library that facilitates high-performance machine learning. Flax is a neural network library built on JAX, and Optax is an optimization library for machine learning. This implementation is notable for leveraging these technologies to improve AI model training efficacy.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Lovable says it’s nearing 8 million users as the year-old AI coding startup eyes more corporate employees
The rapid increase in users for Lovable highlights the growing demand for accessible AI coding platforms, reflecting broader trends in the tech industry's approach to software development.
- Lovable, a Stockholm-based AI coding platform, is nearing 8 million users, up from 2.3 million active users reported in July.
- The company claims to facilitate the creation of 100,000 new products daily and has raised a total of $228 million in funding.
- More than half of Fortune 500 companies are currently using Lovable, which aims to enhance creativity in software development.
- Lovable achieved $100 million in annual recurring revenue in June, but recent reports indicate a 40% decline in traffic since September.
- The startup is expanding its workforce and focusing on enhancing security measures following concerns in the vibe coding sector.
Lovable was founded a year ago and emerged from GPT Engineer, an open-source tool for developers. The company aims to simplify software creation for non-coders, with a user base that includes many large corporations. Future developments to watch include potential fundraising and the evolution of user engagement amid fluctuating traffic.
Read the original ↗
Researchers isolate memorization from reasoning in AI neural networks
This research clarifies how AI models process memorization and reasoning separately, which could lead to advancements in AI training and memory management techniques.
- Researchers from Goodfire.ai found that AI models use distinct neural pathways for memorization and reasoning functions.
- Removing memorization pathways from AI models led to a 97% drop in their ability to recite training data, while reasoning capabilities remained largely unaffected.
- Arithmetic operations were found to share the same pathways as memorization, resulting in a 66% reduction in mathematical performance when memorization was removed.
Goodfire.ai is exploring the architecture of AI language models, which often combines memorization and reasoning. This study is part of ongoing research aimed at improving AI performance in logical reasoning and memory management.
Read the original ↗
Moonshot AI Releases Kosong: The LLM Abstraction Layer that Powers Kimi...
The release of Kosong signals a significant advancement in AI technology by providing a new framework for leveraging large language models, which could enhance various applications.
- Moonshot AI has launched Kosong, an abstraction layer designed for large language models (LLMs).
- Kosong serves as the foundational technology that powers Kimi, a command-line interface (CLI) application.
Moonshot AI is focused on developing innovative AI technologies. The launch of Kosong illustrates a trend toward creating frameworks that streamline the use of complex AI models in practical applications. Future developments may include more applications utilizing this layer.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Kaltura acquires eSelf, founded by creator of Snap’s AI, in $27M deal
This acquisition highlights Kaltura's strategic shift towards integrating advanced AI technologies to enhance their video offerings, potentially transforming customer and employee experiences in various sectors.
- Kaltura is acquiring Israel-based startup eSelf.ai for approximately $27 million to integrate conversational AI into its video platform.
- eSelf is known for creating photorealistic digital avatars that can engage users in real-time conversations across more than 30 languages.
- All eSelf employees will join Kaltura, with its founders overseeing the integration of their technology into Kaltura's existing systems.
Kaltura, headquartered in New York, provides cloud-based video solutions and has over 800 enterprise customers. The acquisition of eSelf is part of Kaltura's strategy to evolve its offerings beyond video management into interactive, AI-driven user experiences.
Read the original ↗
A better way of thinking about the AI bubble
The dynamics of AI investments highlight potential risks and mismatches in technology deployment, which could impact future market stability and innovation.
- Investors are facing uncertainties with the rapid development of AI software versus the slow construction of data centers.
- Oracle's data center campus in New Mexico secured $18 billion in credit from 20 banks for AI infrastructure.
- Oracle has contracted $300 billion in cloud services with OpenAI to develop a $500 billion AI infrastructure project in collaboration with SoftBank.
- Meta has committed to spending $600 billion on AI infrastructure over the next three years.
- A recent McKinsey survey indicates that while AI is widely adopted, few companies are using it at scale, leading to uncertainty in demand.
The article examines the complexities of the AI sector, discussing the potential consequences of over-investment amid uncertain future demand. As companies like Oracle and Meta invest heavily in AI infrastructure, understanding the balance of supply and demand remains crucial. Future developments in AI service usage and infrastructure capacity will shape the industry's landscape.
Read the original ↗
6sense founder Amanda Kahlow raises $30 million for new human-replacement AI sales startup 1mind
The funding raised by 1mind highlights significant investment in AI technologies aimed at transforming sales processes and potentially reducing human roles in sales operations.
- 1mind, co-founded by Amanda Kahlow, raised $30 million in a Series A funding round, increasing its total funding to $40 million.
- The startup's AI sales agent, Mindy, is designed to handle inbound sales and assist in closing deals, augmenting human sales efforts.
- 1mind's technology is currently utilized by over 30 companies, including HubSpot and LinkedIn, with annual contracts averaging six figures.
1mind is a startup focused on employing AI in sales, particularly in managing inbound sales interactions. Founded by Amanda Kahlow, a notable figure in sales technology from her previous work at 6sense, the company aims to enhance sales efficiency. Future developments may lead to greater reliance on AI for tasks traditionally handled by human sales representatives.
Read the original ↗
Wikipedia urges AI companies to use its paid API, and stop scraping
Wikipedia's decision to promote its paid API and regulate AI access aims to protect its content and ensure sustainable support for its nonprofit mission amid declining traffic.
- Wikipedia urges AI developers to use its paid API, Wikimedia Enterprise, instead of scraping content from its website.
- The paid API allows companies to access Wikipedia's content responsibly while supporting its nonprofit mission.
- Recent updates revealed that AI bots were scraping Wikipedia at high volumes, contributing to an 8% decline in human page views year-over-year.
The Wikimedia Foundation operates Wikipedia, an online encyclopedia that has seen declining traffic due to AI content practices. Wikipedia previously proposed AI strategies for enhancing editor workflows, emphasizing attribution for content use by AI developers.
Read the original ↗
Artificial Intelligence (370)
This report highlights trends in AI education, which could inform policy decisions and education strategies nationwide.
- The report discusses the AI education surge in the United States.
- It identifies which states and schools are leading in AI educational initiatives for 2025.
AI education is becoming increasingly important as technology evolves. Understanding which states and schools excel may shape future educational frameworks and policies. Monitoring these trends can help stakeholders adapt to the changing landscape.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Google Maps releases new AI tools that let you create interactive projects
Google Maps' new AI tools facilitate enhanced project development, broadening accessibility for users and developers to create customized interactive experiences with geographic data.
- Google Maps is introducing AI tools including a builder agent and an MCP server to assist users and developers in creating interactive projects.
- The builder agent generates code based on user descriptions, allowing for projects like real-time weather maps or themed street tours.
- Google is implementing a feature called Grounding Lite, enabling AI assistants to connect with external data for more contextual responses.
- A Contextual View component will provide visual answers to user queries in a low-code format, enhancing interaction with the map.
- Additionally, Google is rolling out a code assistant toolkit that connects with Google Maps' documentation to aid developers.
Google Maps is a mapping service that now integrates AI to enhance user experience and developer capabilities. The recent introduction of features builds on ongoing advancements in AI, particularly with Google's Gemini models, to improve project interactivity. Future developments may include further integration of AI tools to strengthen consumer engagement and problem-solving capabilities in mapping.
Read the original ↗
EVTAR: End-to-End Try on with Additional Unpaired Visual Reference
This advancement in virtual try-on technology facilitates a more accessible and realistic consumer experience in online shopping, potentially increasing user engagement and satisfaction.
- EVTAR is an end-to-end virtual try-on model that improves accuracy by using additional reference images.
- The model simplifies the inference process, eliminating the need for complex inputs like human poses or segmentation maps.
- EVTAR employs a two-stage training strategy, requiring only the source image and the target garment for generating try-on results.
EVTAR represents a shift in virtual try-on technology by simplifying input requirements, which could lead to its practical application in e-commerce. Previous virtual try-on methods often depended on complex data, limiting usability. Stakeholders should monitor its performance on existing benchmarks and its integration into commercial platforms.
Read the original ↗
How AI is giving Northern Ireland teachers time back
This pilot program demonstrates how AI tools can enhance teaching efficiency, enabling educators to use saved time for more impactful engagement with students and professional development.
- A six-month pilot program in Northern Ireland involved 100 teachers using AI tools like Gemini and Google Workspace.
- Participating teachers reported saving an average of 10 hours per week, which they redirected to student engagement and professional development.
- The project identified over 600 unique uses for Gemini, including lesson creation and administrative task automation.
The C2k initiative is aimed at integrating technology in education across Northern Ireland. Prior to this pilot, AI's role in education was limited, with increased interest in enhancing teaching practices through technology. Future developments may involve broader AI tool training for more teachers through C2k's initiatives.
Read the original ↗
Amazon’s House of David Used Over 350 AI Shots in Season 2. Its Creator Isn’t Sorry
The increased use of AI in filmmaking raises questions about creative integrity and the potential impact of technology on traditional production methods, particularly in the entertainment industry.
- House of David season 2 utilized between 350 and 400 AI-generated shots, significantly up from over 70 in season 1.
- Showrunner Jon Erwin stated that AI tools were used to create complex scenes affordably, combining multiple AI technologies.
- The series has resonated with evangelical audiences, selling over half a million subscriptions in three weeks since its release.
House of David is an Amazon Prime series exploring the biblical story of King David. This season reportedly used AI technology to produce visuals more cost-effectively amidst industry debates on AI's impact on creative jobs and artistic quality.
Read the original ↗
Scribe hits $1.3B valuation as it moves to show where AI will actually pay off
Scribe's substantial valuation reflects growing business interest in identifying realistic applications for AI and automation, critical for optimizing productivity in enterprises.
- Scribe has raised $75 million at a $1.3 billion valuation to launch Scribe Optimize, a platform designed to help organizations identify effective AI applications.
- The Series C funding round was led by StepStone with participation from multiple existing investors, providing financial backing to accelerate Scribe Optimize's rollout.
- Scribe's existing product, Scribe Capture, automates the documentation of workflows, reportedly saving users significant hours and speeding up onboarding by 40%.
- Since its founding in 2019, Scribe has documented over 10 million workflows and currently serves more than 78,000 paid customers, including teams at 94% of Fortune 500 companies.
Scribe is a San Francisco-based startup focused on workflow documentation and automation. Founded in 2019, the company has seen rapid growth, with plans to double its workforce within a year. The market for process documentation includes competitors like Tango and UserGuiding.
Read the original ↗
7 Crazy Things You Can Do with Canva Creative Operating System
The launch of Canva's Creative Operating System has the potential to streamline design and marketing processes for a wide range of users, reducing complexity and enhancing productivity.
- Canva launched its AI-powered Creative Operating System, integrating design, marketing, and automation tools into a single platform.
- The platform features a new AI design model that allows users to create editable design files rather than flat images.
- Canva Grow enables users to create and publish ads directly to social media, improving integration between design and marketing efforts.
- New tools like Canva Forms and Canva Code simplify web design without needing coding expertise, enhancing interactivity and data management.
- The Brand Kit system helps maintain brand consistency by centralizing assets and guidelines for teams.
Canva is a graphic design platform that allows users to create visual content. The Creative Operating System marks a significant development in integrating advanced AI capabilities with design tools, aiming to simplify the creative process. Watch for further advancements in AI-driven design functionalities and their impact on marketing automation.
Read the original ↗
DeepEyesV2: Toward Agentic Multimodal Model
The development of DeepEyesV2 represents a significant advancement in multimodal AI models that aim to integrate tool use into reasoning, enhancing real-world task performance.
- DeepEyesV2 is an agentic multimodal model utilizing a two-stage training pipeline for effective tool use integration.
- The model performs well on real-world reasoning tasks, such as mathematical reasoning and search-intensive operations.
- A comprehensive benchmark called RealX-Bench was introduced to evaluate the model's real-world multimodal reasoning capabilities.
DeepEyesV2 builds upon previous vision-language models by integrating text and image reasoning with external tool usage. The research focuses on improving the effectiveness of these models in practical applications through innovative training and evaluation methods. Future developments may include enhancements in model integration and performance across various reasoning tasks.
Read the original ↗