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
Were Those UAP Just “Drones and Balloons?”
Understanding the nature of recent Unidentified Aerial Phenomena (UAP) sightings could impact national security assessments and public perceptions of military encounters.
- Recent videos of UAP sightings by U.S. naval vessels off the California coast have intensified speculation about their nature.
- Analyst Tyler Rogoway suggests many UAP reports may be attributed to foreign drones and balloons, rather than extraterrestrial origins.
- The current debate emphasizes the need for more witness testimony and physical evidence to clarify what UAPs may be.
UAPs are objects observed in the sky that cannot be identified as aircraft or known phenomena. Recent encounters, particularly involving U.S. Navy vessels, have sparked renewed interest and analysis in their origins. Ongoing investigations aim to obtain more evidence and expert testimonies to better understand these sightings.
Read the original ↗
New AI architecture delivers 100x faster reasoning than LLMs with just 1,000 training examples
The development of the Hierarchical Reasoning Model (HRM) could revolutionize AI by enabling faster and more efficient processing of complex reasoning tasks, reducing reliance on large datasets and computational resources.
- Sapient Intelligence, a startup in Singapore, has developed a new AI architecture called the Hierarchical Reasoning Model (HRM).
- The HRM architecture outperforms large language models (LLMs) in complex reasoning tasks while requiring only 1,000 training examples.
- The model operates through a hierarchical structure that mimics human brain functions, allowing for efficient problem-solving without excessive data or slow response times.
- Testing revealed that HRM achieved near-perfect accuracy on benchmark tests like Sudoku and maze-solving, where LLMs scored 0% in some cases.
- Sapient Intelligence is working on evolving HRM into a more general-purpose reasoning module for various applications, including healthcare and robotics.
The Hierarchical Reasoning Model (HRM) is an innovative AI architecture inspired by human thinking processes, which allows it to be smaller and faster than conventional large language models. Traditionally, LLMs use a method known as chain-of-thought prompting, which can be slow and data-intensive. The introduction of HRM may signal a shift away from merely increasing model size toward improving their structural efficiency and reasoning capabilities.
Read the original ↗
AI referrals to top websites were up 357% year-over-year in June, reaching 1.13B
The significant rise in AI referrals indicates changing dynamics in online traffic, potentially affecting how websites engage with audiences and their dependency on traditional search engines like Google.
- AI platforms generated 1.13 billion referrals to the top 1,000 websites globally in June 2025, marking a 357% increase year-over-year.
- Google Search still dominates web traffic, leading with 191 billion referrals in June 2025.
- In the news media category, AI referrals rose 770% since June 2024, with Yahoo receiving the most at 2.3 million referrals.
AI referral traffic comes from platforms like ChatGPT and others, counted by Similarweb. The trend shows a significant shift in online traffic management, particularly affecting news sites and traditional search dependence.
Read the original ↗
Meta names Shengjia Zhao as chief scientist of AI superintelligence unit
The appointment of Shengjia Zhao as chief scientist is significant as it strengthens Meta's position in the competitive AI landscape, particularly in developing advanced reasoning models.
- Shengjia Zhao has been appointed as the Chief Scientist of Meta's new AI unit, Meta Superintelligence Labs (MSL).
- Zhao, a former OpenAI researcher, contributed to major projects like ChatGPT and GPT-4, boosting MSL's research capabilities.
- Zhao will lead research efforts under Alexandr Wang, a non-researcher as the unit's manager, indicating a diverse leadership structure.
- Meta has recruited top researchers from various AI firms to bolster MSL's talent pool and focus on AI reasoning models.
- Meta is enhancing its cloud computing resources, with plans for a 1 gigawatt cluster to support extensive AI training needs by 2026.
Meta is expanding its AI efforts through the establishment of Meta Superintelligence Labs, designed to develop advanced AI models. This comes as the company aims to enhance its competitiveness against other leading AI organizations like OpenAI and Google. Zhao's expertise and the new leadership team are crucial for the lab's success in this rapidly evolving field.
Read the original ↗
CoSyn: The open-source tool that’s making GPT-4V-level vision AI accessible to everyone
The release of CoSyn democratizes access to advanced visual AI capabilities, enabling more equitable competition in AI development between open-source communities and major tech companies.
- CoSyn, developed by researchers at the University of Pennsylvania and Allen Institute, allows open-source AI to match or exceed the capabilities of proprietary models like GPT-4V.
- The tool generates synthetic training data using the coding abilities of language models, mitigating the issues of data scarcity and copyright.
- CoSyn-trained models outperformed proprietary models on seven benchmarks measuring text-rich image understanding, scoring an average of 80.9%.
- The technology is already being utilized in industries for applications like quality control and automated document processing.
- CoSyn provides a complete open-source codebase and dataset, fostering further research and development in the AI field.
CoSyn (Code-Guided Synthesis) addresses AI's long-standing challenges with data scarcity in training capable vision models. The tool employs code generation for creating synthetic training data and facilitates more effective AI learning. Its implications may significantly alter the competitive landscape of AI development.
Read the original ↗
Hypnosis Radically Alters How The Brain Processes Information, New Study Finds
The study provides new neuroscientific insight into how hypnosis alters brain function, potentially resolving longstanding debates about its nature and applications in psychological treatment.
- Researchers from the University of Turku found that hypnosis significantly disrupts normal brain information processing, with brain regions acting independently rather than in synchronization.
- The study utilized perturbational transcranial magnetic stimulation and electroencephalography to measure brain activity during hypnosis, observing marked differences compared to normal wakefulness.
- These findings suggest that altered mental states, like those induced by hypnosis, increase the complexity of brain activity, akin to prior results seen in psychedelic studies.
Hypnosis is an induced state of semi-consciousness where an individual is highly suggestible. This research builds on centuries of interest in hypnosis and its varying applications, from therapy to entertainment. Future studies may further explore the implications of these findings for psychological practices and understanding consciousness.
Read the original ↗
OpenAI’s most capable AI model, GPT-5, may be coming in August
The upcoming release of GPT-5 is significant as it may integrate advanced reasoning capabilities and unify OpenAI's technology, impacting the AI landscape substantially.
- OpenAI is reportedly preparing to launch GPT-5 as early as August, according to sources familiar with the company's plans.
- The new model aims to combine traditional large language model capabilities with simulated reasoning features.
- GPT-5 will include 'mini' and 'nano' versions accessible through the API, offering varied functionalities.
- OpenAI's CEO Sam Altman demonstrated the model's capabilities during a podcast, highlighting its advanced reasoning ability.
OpenAI, a leader in artificial intelligence, plans to unify its product offerings with the launch of GPT-5, a more capable model building on features from its previous releases. The company first hinted at GPT-5 in 2024, but several delays and development challenges postponed its release. Watching for the timeline of its launch and any additional safety testing of related models will be essential.
Read the original ↗
Delta’s AI spying to “jack up” prices must be banned, lawmakers say
The proposed law aims to protect consumers and workers from controversial AI practices that could lead to increased prices and reduced wages based on personal data surveillance.
- Legislators are advocating for a ban on AI-driven price and wage surges categorized as predatory surveillance pricing.
- The Stop AI Price Gouging and Wage Fixing Act would allow lawsuits against companies using AI for unfair pricing and wage-setting practices.
- Delta plans to implement AI for determining 20 percent of its flight prices by the end of the year, prompting legislative scrutiny.
The Stop AI Price Gouging and Wage Fixing Act is introduced amid growing concerns over AI's role in pricing strategies, which critics argue could exacerbate economic disparities. Lawmakers highlight Delta's AI pricing practices as a key example of the industry's potential harms. Future developments will depend on political support and the response from companies implicated in surveillance pricing.
Read the original ↗
Tesla is reportedly bringing a limited version of its robotaxi service to San Francisco
Tesla's expansion into robotaxi services could significantly impact regulatory discussions on autonomous vehicles in California amidst existing legal challenges and safety concerns.
- Tesla is launching a limited version of its robotaxi service in San Francisco this weekend.
- This follows an initial rollout in Austin, Texas, where the service is invite-only and has a human safety operator in the front seat.
- Tesla lacks the necessary permits for driverless testing and autonomous passenger services from California regulatory bodies and has not applied for them.
- If Tesla deploys its robotaxis with autonomous systems engaged, it may violate state regulations, even with a human in the vehicle.
- The California DMV is currently pursuing legal action against Tesla regarding its autonomous driving promises and related safety issues.
Tesla's robotaxi service is part of its efforts to develop fully autonomous vehicles. The company previously launched a limited service in Austin, but faces regulatory hurdles in California. Observers are watching to see how this expansion addresses legal compliance and operational safety.
Read the original ↗
Lyft to add autonomous shuttles in 2026 as Uber inks more self-driving deals
The introduction of autonomous shuttles by Lyft represents a significant step in the expansion of self-driving technology in public transportation, aiming to enhance mobility in urban settings.
- Lyft plans to add autonomous shuttles from Benteler Group to its network in late 2026.
- The shuttles will operate with no steering wheels or pedals, accommodating up to 15 passengers in a unique design.
- This announcement follows a series of robotaxi partnerships involving Uber with multiple companies such as Waymo and WeRide.
Lyft is a ride-hailing company currently exploring the use of autonomous vehicles in its fleet. As competitors like Uber expand their robotaxi offerings, Lyft aims to enhance its position in the transportation market with these new shuttles.
Read the original ↗
Sam Altman warns there’s no legal confidentiality when using ChatGPT as a therapist
The lack of legal confidentiality for ChatGPT conversations raises significant privacy concerns, especially for users seeking emotional support or therapy through AI applications.
- Sam Altman noted that using ChatGPT for therapy lacks legal confidentiality, which could expose personal conversations in legal situations.
- Currently, OpenAI is required to produce user conversations if legally mandated, creating privacy risks for users.
- OpenAI is appealing a court order that would compel it to retain chat data for legal discovery, raising further privacy concerns.
OpenAI's ChatGPT is an artificial intelligence tool used for various tasks, including providing emotional support. As the legal framework for AI evolves, there is increasing scrutiny on how user data is handled, particularly for sensitive interactions. Users are advised to consider privacy implications when using AI for personal matters.
Read the original ↗
Mistral’s new “environmental audit” shows how much AI is hurting the planet
AI's environmental impact is significant and quantifiable, raising questions about sustainability as usage increases globally.
- Mistral released an environmental audit quantifying the impacts of its large language models, focusing on CO2 emissions and water consumption.
- The audit found that over 85% of emissions occurred during model training and inference.
- Mistral estimated that its AI model emitted 20.4 kilotons of CO2 and consumed 281,000 cubic meters of water in its first 18 months.
- The marginal environmental impact of an average AI prompt was found to be 1.14 grams of CO2 and 45 milliliters of water consumed.
Mistral is a French company specializing in large language models (LLMs). This environmental audit, conducted in partnership with sustainability experts, seeks to provide greater transparency about AI's resource usage and emissions. Ongoing discussions in the tech world revolve around balancing the benefits of AI against its environmental footprint.
Read the original ↗
DARPA Develops AI Sarcasm Detector (Just What We Needed)
The development of an AI model to detect sarcasm in online communication is crucial for enhancing the accuracy of sentiment analysis in social media, which has implications for information warfare and security.
- DARPA has demonstrated an AI model that can detect sarcasm in textual communications, addressing challenges in sentiment analysis.
- Researchers from the University of Central Florida developed this model to identify sarcasm through patterns and cue words.
- The model has reportedly achieved nearly perfect sarcasm detection across five datasets from social media and online media.
DARPA, the U.S. agency responsible for advanced research projects, aims to improve understanding of online social behavior through their SocialSIM program. This sarcasm detection effort highlights challenges in interpreting nuanced human communication in digital environments and is part of broader initiatives to enhance AI systems' reliability.
Read the original ↗
It's Qwen's summer: new open source Qwen3-235B-A22B-Thinking-2507 tops OpenAI, Gemini reasoning models on key benchmarks
The Qwen3-235B-A22B-Thinking-2507 model's top performance on key benchmarks could significantly influence the competitive landscape of AI reasoning models, particularly in enterprise applications.
- Alibaba's Qwen Team released Qwen3-235B-A22B-Thinking-2507, a new reasoning large language model, which leads several AI benchmarks.
- The model scored 92.3 on the AIME25 benchmark, outperforming OpenAI's o4-mini and Google's Gemini-2.5 Pro.
- Qwen3-Thinking-2507 is licensed under Apache 2.0, allowing commercial use and modification without restrictions.
- The team also launched additional models designed for coding and multilingual tasks, enhancing the overall Qwen3 ecosystem.
Qwen is Alibaba's frontier AI research division focused on developing generative AI models. The recent release of specialized models reflects a strategic shift towards optimizing AI for distinct tasks. This trend is noteworthy as companies increasingly seek open-source alternatives to proprietary AI solutions.
Read the original ↗
ChatGPT Agent has Arrived and it’s so Cool!
The launch of the ChatGPT Agent represents a significant advancement in AI capabilities, allowing users to automate tasks previously requiring human intervention, which could enhance productivity across various sectors.
- OpenAI's ChatGPT Agent, released on July 17, 2025, transforms the chatbot into a digital assistant that autonomously performs tasks like research and planning.
- The Agent can browse websites, manipulate data, create presentations, and manage complex workflows for users.
- Pro subscribers can access the Agent first, with a limited monthly usage cap, and it will later be available to Plus and Team users.
OpenAI is a leading artificial intelligence company known for developing versatile AI models, including the popular ChatGPT. Previous versions of ChatGPT provided text-based assistance, but the new Agent version enhances its capabilities to independently complete tasks. Future developments will likely focus on expanding the Agent's functionalities and accessibility to more users.
Read the original ↗
A satellite just used AI to make its own decisions in space — and NASA’s stoked
This advancement in satellite technology can significantly enhance data collection efficiency in space, potentially improving responses to natural disasters and environmental monitoring.
- A satellite autonomously decided where and when to capture a scientific image using onboard AI for the first time, completing this in under 90 seconds.
- The technology, called Dynamic Targeting, was tested by NASA's Jet Propulsion Laboratory aboard a satellite from UK startup Open Cosmos.
- Ubotica's AI analyzed cloud cover to determine whether to take detailed images, optimizing data collection by skipping shots obscured by clouds.
NASA's Jet Propulsion Laboratory is actively developing technologies to improve satellite operations. The Dynamic Targeting technology builds on earlier collaborations with Ubotica and Open Cosmos that included AI cloud detection tests.
Read the original ↗
Boeing’s X-20 Dyna-Soar Was The Air Force’s ‘Spaceplane’ That Never Flew
The Boeing X-20 Dyna-Soar program highlights the historical challenges and debates in American aerospace development, illustrating how priorities and funding can reshape technological possibilities.
- The Boeing X-20 Dyna-Soar was a military spaceplane program conceived by the USAF from 1957 to 1963 that ultimately never flew.
- The program faced significant hurdles, including funding issues and doubts about its utility, which led to its cancellation before prototype construction was completed.
- The X-20 was designed for various military purposes but was ultimately deemed unnecessary as objectives for its orbital flight were not sufficiently defined.
The Boeing X-20 Dyna-Soar was part of early American efforts to develop reusable spacecraft, alongside the more recognized Space Shuttle. Its development was influenced by prior programs like the X-15 and the controversies over military spaceflight objectives. Current spaceplane developments, such as the Boeing X-37, can trace their lineage back to concepts like the X-20.
Read the original ↗
NASA’s “Hail Mary” Moment: Incredible Deep Space Surgery Saves JunoCam from Jupiter’s Fierce Radiation Fields
This successful remote repair technique enhances the potential for maintaining spacecraft instruments in hostile environments, which is crucial for future deep-space missions.
- NASA executed a successful rescue mission to repair the JunoCam camera aboard the Juno spacecraft in December 2023.
- Engineers used an experimental technique known as annealing to fix a damaged voltage regulator caused by radiation exposure.
- The JunoCam camera, which captures visible light images, resumed producing sharp images shortly after the repair, enabling new observations of Jupiter's moon Io.
The Juno spacecraft has been studying Jupiter since 2016, with JunoCam designed to capture high-resolution images. Previous concerns about radiation damage led to innovative repair techniques that could inform the design of future space instruments.
Read the original ↗
Anthropic deploys AI agents to audit models for safety
The deployment of AI agents for model auditing highlights the growing emphasis on safety in artificial intelligence development, a crucial factor in public trust and regulatory compliance.
- Anthropic has deployed AI agents to audit models for safety.
- This initiative aims to enhance the reliability of AI systems through improved oversight.
Anthropic is a company focused on developing safer AI models. The deployment of auditing tools marks a step toward addressing concerns about the ethical implications and potential risks of AI technology. Future updates may provide insights into the effectiveness of these safety measures and their impact on AI advancement.
Based on the headline only — the full article could not be fetched.
Read the original ↗
Alibaba’s new Qwen reasoning AI model sets open-source records
The achievement of Alibaba's Qwen reasoning AI model could significantly advance open-source developments in artificial intelligence, influencing future applications and innovations in the field.
- Alibaba has launched its new Qwen reasoning AI model.
- The model has set records in the open-source AI space.
Alibaba is a major Chinese tech company known for e-commerce and cloud computing. Open-source AI models promote collaboration and advancement in technology, allowing developers to access and improve upon existing AI frameworks. This development may lead to increased competition in the AI industry and further innovations.
Based on the headline only — the full article could not be fetched.
Read the original ↗