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
CHARM: Control-point-based 3D Anime Hairstyle Auto-Regressive Modeling
The development of CHARM could significantly improve the efficiency and quality of generating anime hairstyles, catering to both artists and computational learning methods.
- CHARM is a new generative framework designed for efficiently modeling anime hairstyles.
- It employs a compact control-point-based parameterization, encoding hair geometry with just five parameters per point.
- The framework uses an autoregressive transformer to capture hairstyle features from input images or point clouds, enhancing generation quality.
- A dataset called AnimeHair, consisting of 37,000 high-quality anime hairstyles, has been created to support training and evaluation of the model.
CHARM addresses limitations in traditional hair modeling approaches by offering a method suited for stylized anime hair. Previous methods often struggled with efficiency and scalability due to their reliance on complex meshes or manual designs. Future developments will likely focus on refining the generative capabilities and expanding applications of CHARM within the animation industry.
Read the original ↗
MMR1: Enhancing Multimodal Reasoning with Variance-Aware Sampling and Open Resources
By enhancing multimodal reasoning models through innovative data strategies and addressing algorithmic instability, this research aims to improve AI's ability to process and understand complex information.
- The research introduces Variance-Aware Sampling (VAS) to enhance reinforcement learning fine-tuning and stabilize policy optimization.
- It releases ~1.6 million high-quality long chain-of-thought data points and ~15,000 reinforcement learning question-answer pairs for community use.
- The study provides open-source multimodal reasoning models across multiple scales to establish standardized benchmarks for future research.
Multimodal reasoning involves AI's ability to integrate and process information from different modes, such as text and visuals. The previous research faced challenges in the availability of robust data and stable training methods. Future work will likely build on these resources to further advance AI capabilities in reasoning tasks.
Read the original ↗
Tree Search for LLM Agent Reinforcement Learning
This development in reinforcement learning could enhance the performance of large language models in complex, long-term tasks, addressing prior limitations of sparse feedback.
- Tree-based Group Relative Policy Optimization (Tree-GRPO) improves reinforcement learning for large language models using tree search to enhance rollouts.
- The method is designed to address the limitations of sparse supervision in long-term agent tasks.
- Experiments show Tree-GRPO outperforms traditional chain-based reinforcement learning methods across 11 datasets and 3 types of question and answer tasks.
Reinforcement learning enables AI models to improve through interaction with their environment, with recent advancements significantly benefitting large language models. Previous methods struggled with generating meaningful feedback due to sparse rewards. Future research may explore further improvements in reinforcement learning techniques for AI applications.
Read the original ↗
SciReasoner: Laying the Scientific Reasoning Ground Across Disciplines
This development in scientific reasoning models could significantly enhance the efficiency and accuracy of research across various scientific fields by improving data handling and interpretation.
- The SciReasoner is a foundation model pre-trained on a 206 billion-token corpus of scientific data.
- It supports over 100 tasks across multiple workflows, including translation, extraction, and property prediction.
- The model utilizes specialized training techniques such as reinforcement learning and alignment on diverse scientific instructions.
SciReasoner is a model designed to improve scientific reasoning by aligning natural language with various scientific representations. This foundation model is part of a growing trend towards using AI to enhance data processing and interpretation in research. Future developments may focus on refining its applications and expanding its compatibility with various scientific disciplines.
Read the original ↗
ChatGPT Pulse delivers morning updates based on your chat history
The launch of ChatGPT Pulse highlights the growing trend of proactive AI features, which can provide users with timely updates and enhance personal productivity.
- OpenAI debuted ChatGPT Pulse, a feature that delivers personalized daily updates based on users' chat history and connected apps, available to Pro subscribers now.
- The feature analyzes users' conversations overnight and presents updates on relevant topics through visual 'cards' the next morning.
- Users can provide feedback on updates and connect Gmail and Google Calendar for more contextual insights, enhancing the relevance of suggestions.
ChatGPT is a conversational AI model developed by OpenAI, which has focused on making interactions more proactive. Prior to Pulse, OpenAI introduced Tasks for user-directed scheduling. Future updates may expand its capabilities and broaden app integration.
Read the original ↗
OpenAI Introduces GDPval: A New Evaluation Suite that Measures AI on Real-World Economically Valuable Tasks
The introduction of GDPval could significantly impact how AI systems are evaluated for their economic contribution, influencing future development and deployment strategies.
- OpenAI has launched GDPval, an evaluation suite designed to assess AI on tasks that have real-world economic value.
- This suite aims to measure the performance of AI in contexts relevant to economic productivity.
OpenAI is a leader in artificial intelligence research and development. The introduction of GDPval represents a shift towards quantifying AI’s impact on the economy. Future developments may focus on integrating these evaluations into AI training and application.
Based on the headline only — the full article could not be fetched.
Read the original ↗
OpenAI Releases ChatGPT ‘Pulse’: Proactive, Personalized Daily Briefings for Pro Users
The introduction of proactive, personalized briefings by OpenAI enhances the user experience for professionals, potentially improving information retention and decision-making processes.
- OpenAI has launched ChatGPT 'Pulse' for Pro users.
- The feature provides proactive and personalized daily briefings.
OpenAI is known for its advancements in artificial intelligence and natural language processing. Previous iterations of ChatGPT focused on reactive interactions, while 'Pulse' represents a shift towards more tailored user engagement. Observers should watch for user feedback and potential updates.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Breaking Physics? Scientists Defy Heisenberg Uncertainty Principle in Landmark Experiment
This breakthrough allows for new methods of measurement that could significantly enhance technologies in fields like navigation, medical imaging, and quantum sensing.
- Researchers developed a method to measure a particle's location and momentum simultaneously without violating the Heisenberg Uncertainty Principle.
- The approach utilizes quantum methods to redistribute uncertainty, enabling measurements with precision beyond classical limits.
- This research could lead to advancements in ultra-precise sensing across various applications, including GPS-denied navigation and medical technologies.
The Heisenberg Uncertainty Principle, proposed in 1927, states that measuring certain particle properties simultaneously is impossible. Recent experiments have reaffirmed its claims, but this team from the University of Sydney has now shown a way to circumvent that limitation, offering promising new avenues for scientific and technological developments.
Read the original ↗
Experts urge caution about using ChatGPT to pick stocks
As retail investors increasingly turn to AI tools for stock picking, experts caution about the dangers of relying on these models without understanding market risks, especially in a fluctuating economy.
- At least 1 in 10 retail investors now consult AI chatbots like ChatGPT for stock-picking advice, according to a recent survey.
- US stocks are at record highs, making it easier for any stock-picking strategy to appear successful, experts say.
- There are concerns that AI models can misquote data and fail to provide real-time market information, posing risks for investors.
AI chatbots are becoming popular among retail investors, who previously relied on human experts for stock advice. Despite some reports of AI-selected portfolios outperforming traditional funds, experts warn of the potential risks involved in using AI for investment decisions without sufficient market knowledge.
Read the original ↗
Elon Musk’s xAI offers Grok to federal government for 42 cents
The government deal for Grok indicates an emerging competition in AI technology supply for federal agencies, potentially influencing pricing and innovation in the sector.
- xAI has reached an agreement with the U.S. General Services Administration (GSA) to sell its AI chatbot Grok for 42 cents over 18 months.
- The price is significantly lower than offerings from OpenAI and Anthropic, which charge $1 for their AI services to federal agencies.
- The deal includes support from xAI engineers for technology integration within government agencies.
xAI, founded by Elon Musk, develops AI technologies, including the Grok chatbot. The company faced delays earlier this year due to controversy but is now moving forward with the GSA partnership. Future developments may reveal how Grok performs and competes in government sectors against established AI providers.
Read the original ↗
OpenAI says GPT-5 stacks up to humans in a wide range of jobs
As AI technology advances, understanding its capabilities relative to human workers could shape workforce dynamics and impact job markets across various industries.
- OpenAI's new benchmark, GDPval, tests AI performance against human professionals across nine industries.
- The GPT-5 model was ranked better than or on par with human experts 40.6% of the time.
- Anthropic's Claude Opus 4.1 model scored better than or on par with experts in 49% of tasks.
- The test covers 44 occupations, including healthcare, finance, and manufacturing, assessing tasks like report generation.
- OpenAI acknowledges the need for more comprehensive testing beyond current capabilities to accurately measure AI performance.
OpenAI aims to develop artificial general intelligence (AGI) and GDPval is part of its assessment strategy. Previous AI models had lower performance, but recent improvements indicate that AI could assist in various job-related tasks. Future tests are planned to evaluate more complex workflows and industries.
Read the original ↗
Steph Curry’s VC firm just backed an AI startup that wants to fix food supply chains
Improving food supply chains through AI could significantly enhance efficiency in a trillion-dollar industry that struggles with outdated technology and processes.
- Burnt, a Y Combinator startup, received $3.8 million in seed funding, led by Steph Curry's Penny Jar Capital, to automate supply chain tasks in the food industry.
- The startup claims its AI agents can handle up to 80% of workflows currently managed by legacy software systems, which are often inefficient.
- Since its launch in January, Burnt has processed over $10 million in monthly orders and is generating six-figure revenue with steady growth.
Burnt aims to improve the food supply chain efficiency by using AI to automate back-office tasks traditionally managed by humans. The company was founded by Joseph Jacob, who has deep family ties to the food industry, and it seeks to restore trust in technology adoption after decades of failed implementations in the area.
Read the original ↗
Juicebox raises $30M from Sequoia to revolutionize hiring with LLM-powered search
Juicebox's innovative use of LLMs for recruitment illustrates a significant advancement in hiring technology, potentially reshaping how companies find and evaluate talent.
- Juicebox has raised $30 million in a Series A round led by Sequoia, bringing total funding to $36 million.
- The AI search engine, PeopleGPT, analyzes professional profiles and websites to identify qualified candidates more efficiently than traditional methods.
- Juicebox serves over 2,500 customers, achieving over $10 million in annual recurring revenue shortly after launching their product.
Juicebox is a startup that provides AI-powered recruitment tools to streamline candidate searches. The company emerged from Y Combinator and has quickly gained traction among various customers, including startups and large firms. The growing demand for faster hiring solutions in tech industries makes Juicebox's services particularly relevant.
Read the original ↗
OpenAI launches ChatGPT Pulse to proactively write you morning briefs
ChatGPT Pulse aims to enhance user experience by providing curated reports while users sleep, signaling a shift towards more proactive AI tools in consumer technology.
- OpenAI has introduced ChatGPT Pulse, a feature that generates personalized morning briefs for users.
- Pulse will initially be available to subscribers of the $200-a-month Pro plan before potential rollout to all users.
- The feature can summarize news or provide contextual information based on user preferences, enhancing its utility as a personal assistant.
ChatGPT is a conversational AI developed by OpenAI. Pulse is part of OpenAI's strategy to create tools that operate asynchronously, shifting from reactive to proactive interactions. Upcoming improvements are expected to enhance Pulse's capabilities and accessibility.
Read the original ↗
Google DeepMind unveils its first “thinking” robotics AI
The development of AI that can reason before acting significantly enhances the adaptability and functionality of robots, potentially transforming their applications in various fields.
- Google DeepMind has introduced two new models, Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, that enable robots to think and act more autonomously.
- Gemini Robotics-ER 1.5 demonstrates simulated reasoning abilities, allowing it to generate instructions based on visual and text inputs for task completion.
- Gemini Robotics 1.5 executes the generated instructions to perform tasks while considering various factors through its own processing.
Google DeepMind is a leader in AI research focused on developing technologies that mimic human-like reasoning. The Gemini Robotics project aims to create robots that can adjust to new environments without extensive retraining. The introduction of these models marks a significant step in making robotics more versatile and capable of complex tasks.
Read the original ↗
Love, lies, and algorithms: Is AI really helping us find ‘the one’? Live at TechCrunch Disrupt 2025
Understanding the role of AI in dating and relationships is crucial as it shapes social interactions and emotional connections in a technology-driven world.
- Leaders from Tinder, Replika, and Kinsey Institute will discuss AI's impact on love and relationships during TechCrunch Disrupt 2025.
- Panelists include Eugenia Kuyda from Replika, Mark Kantor from Tinder, and Dr. Amanda Gesselman from the Kinsey Institute.
- Topics will cover the effects of recommendation engines, the psychological impact of AI companionship, and ethical concerns in digital intimacy.
TechCrunch Disrupt 2025 is an annual technology conference that features discussions on advances in AI and their implications. The event will take place in San Francisco from October 27 to 29, 2025, and aims to explore the intersection of technology and human relationships.
Read the original ↗
Clarifai’s new reasoning engine makes AI models faster and less expensive
Optimizing AI model performance can reduce costs and improve operational efficiency, addressing the growing demand for computational resources in AI development.
- Clarifai has launched a new reasoning engine that claims to make AI models twice as fast and 40% less expensive.
- The reasoning engine is adaptable to various models and cloud hosts, optimizing hardware performance.
- Third-party verification showed industry-best records in throughput and latency for the new engine.
Clarifai, originally a computer vision service, has shifted focus to compute orchestration in response to increasing demand for GPU resources. The new reasoning engine targets multi-step reasoning models and aims to optimize existing computational infrastructure, amidst rising demand for AI capabilities.
Read the original ↗
What top VCs want from AI founders: Inside the investor lens with Jon McNeill, Aileen Lee, and Steve Jang at TechCrunch Disrupt 2025
Understanding investor expectations can enhance the success of AI startups as they navigate a competitive landscape and seek funding opportunities.
- Top VCs Jon McNeill, Aileen Lee, and Steve Jang will discuss investor insights at TechCrunch Disrupt 2025 in San Francisco from October 27-29.
- The discussion will focus on what makes AI startups defensible and the signals investors look for before making investments.
- Founders will learn how to stand out in a crowded AI startup environment and what they need to demonstrate to attract funding.
TechCrunch Disrupt is an annual conference that brings together startup founders and investors, showcasing emerging technologies. Investors are increasingly interested in AI startups, given their rapid growth and transformative potential. What investors prioritize in AI startups is crucial for founders seeking to secure funding.
Read the original ↗
Two Wins for Google's AI Co-Scientist
These studies indicate that AI can significantly accelerate scientific discovery, potentially transforming how researchers generate and test hypotheses in various fields.
- Google's AI-powered scientific-research assistant generated novel ideas in two studies, potentially shortening years of lab work.
- In one study, the AI identified drugs for treating liver fibrosis that showed promising results in lab tests.
- Another study revealed a hypothesis about bacterial DNA transfer that matched unpublished research, demonstrating the AI's effectiveness.
The AI co-scientist is part of an ongoing effort to leverage AI for scientific research, aiming to assist researchers by generating testable hypotheses. Previous skepticism about AI's creative capabilities has been addressed by these new findings, which show its potential for genuine scientific contributions. Future developments could expand access to this technology, enhancing its role in academic settings.
Read the original ↗
ARE: Scaling Up Agent Environments and Evaluations
New platforms and benchmarks like ARE and Gaia2 aim to enhance the capability and evaluation of AI agents, which is crucial for advancing artificial intelligence in real-world applications.
- Meta Agents Research Environments (ARE) facilitate the creation of complex environments for agent research.
- Gaia2, a benchmark developed on ARE, evaluates agent capabilities in dynamic environments and requires collaboration, handling ambiguities, and adapting to changes.
- Gaia2 runs asynchronously, revealing new failure modes that prior benchmarks did not detect.
Meta Agents Research Environments (ARE) provide tools for scalable environment creation and execution for AI agents. Previously, benchmarks were largely static, limiting evaluations of agent capabilities. Future development will focus on adaptive architectures and continuous extensions of benchmarks.
Read the original ↗