Ai · Top stories
GPT-5.6 Sol Completes Factorio Peaceful Mode in 13.5 Hours
GPT-5.6 Sol, running in Codex, successfully completed a peaceful mode run of the game Factorio in 13.5 hours of game time. This achievement demonstrates progress in AI agents' ability to manage complex, programmatic tasks, building on previous attempts where models struggled with basic game mechanics.
Language Model Runs from PC Boot Sector, Generates Fake Linux Boot Logs
A neural network has been developed that operates directly from a PC's boot sector, loading an 11.5 KB int8 model to generate simulated Linux boot logs. This project demonstrates running a language model in 16-bit real mode on a 386-compatible system without an operating system or C runtime.
AI Law Firm Crosby Bids on Assets of Disintegrating Weil Gotshal
AI-first law firm Crosby has submitted a bid to acquire assets, including the trademark and brand, of the dissolving Wall Street firm Weil Gotshal & Manges. Crosby is also seeking to hire some of Weil Gotshal's lawyers, aiming to integrate traditional legal services with its AI technology.
NVIDIA Nemotron Voice Agent Powers Interactive Halloween Projection
An interactive Halloween projection named MonsterHouse was created using NVIDIA's Nemotron Voice Agent, Nemotron ASR, Nemotron 3.5 Bolt E4B 4GB, Magpie TTS Zeroshot, and Audio2Face-3D. The system allows a garage door to listen to trick-or-treaters, respond as a grumpy pirate, and animate a mouth. This project demonstrates the practical application of NVIDIA's AI tools for real-time conversational agents and animation in a creative context.
Zest Launches AI Text-Based Restaurant Concierge for Recommendations and Reservations
Zest, a restaurant discovery app, introduced an AI-powered text message concierge service that recommends restaurants and books reservations. This new feature allows users to interact with Zest's recommendation engine and booking capabilities directly through text messages, including group chats, without needing to open a separate app.
Starbucks plans AI-powered personalized drive-thru menus and order assistance
Starbucks announced plans to implement AI for personalized digital menus at drive-thrus, aiming to recognize customers and tailor offerings. The company will also test AI models by 2027 to improve order accuracy and efficiency for baristas.
Musk and Zuckerberg adopt 'super intelligence' terminology following Trump's directive
Elon Musk and Mark Zuckerberg have begun using the term "super intelligence" instead of "artificial intelligence" in their companies' communications, aligning with a recent directive from former President Trump. This shift was observed on recent earnings calls for Tesla, SpaceX, and Meta, while other major tech CEOs like those from Amazon, Microsoft, and Alphabet continue to use "AI."
GhostTube Seer App Uses AI for Paranormal Investigations, Drawing Mixed Reactions
Paranormal investigators Amy Waine and Jarrad Cutting developed GhostTube Seer, an AI-powered app that generates images based on phone sensor data, which they interpret as supernatural activity. The app has over 1 million downloads but faces skepticism from some in the ghost-hunting community concerned about AI-generated content flooding online spaces.
Agentic AI Integrates into SETI's Search for Extraterrestrial Intelligence
Breakthrough Listen and the SETI Institute are now using agentic AI to process data from the Allen Telescope Array. This integration allows AI agents to steer the telescope, correct data, and apply models to incoming signals using natural language requests, speeding up the search for extraterrestrial signals.
Victorian Lawyers Increase AI Use, Few Reduce Client Fees Despite Efficiency Gains
A study of Victorian lawyers found AI usage increased from 37% to 55% in the last year, with 79% using consumer chatbots. Only 26% of lawyers reported reducing client fees due to AI efficiencies, and many noted increased workload correcting client-generated AI errors.
DOOM game engine used to compute neural networks with "voodoo dolls"
A project named "Doomformer" demonstrates running neural network computations within the DOOM game engine. It uses in-game elements, referred to as "voodoo dolls," to represent network values and operations, with results verified against NumPy.
VRChat Declines to Implement AI Content Labels Despite User Survey Indicating Demand
VRChat announced it will not add labels for AI-generated content, despite 95% of survey respondents requesting such features. The company cited concerns that labels could lead to mislabeling, 'sleuthing' behavior, and harassment among creators. Instead, VRChat will focus on accurate thumbnails, improved discovery tools, and content quality.
Claude AI used in antibody design for IgE disruptor in food allergy research
Researchers utilized the Claude AI model to design antibodies aimed at disrupting IgE, a key mediator in allergic reactions, for potential food allergy treatment. This application demonstrates AI's role in accelerating early-stage drug discovery and antibody engineering.
Granola AI launches dictation feature, competing with Wispr
Granola, an AI meeting notetaker, released a dictation feature for Mac, allowing users to dictate text across applications. This move follows a trend of voice AI companies expanding their offerings beyond single-purpose tools, as seen with competitor Wispr adding meeting notetaking.
Human Psychics Push Back Against AI-Generated Readings and Bots
Human psychics are expressing concern and taking action against the rise of AI-generated psychic readings and spiritual products. They argue that AI lacks the empathy and human connection central to their practice, while AI-powered apps like "Your AI Psychic" offer automated readings for a fee.
Instagram Head Adam Mosseri Predicts Worsening Negativity Around AI
Instagram head Adam Mosseri stated in a podcast interview that public negativity and frustration regarding AI are increasing and will continue to worsen, particularly within creative industries. He acknowledged concerns about job displacement, content ownership, and safety, while also discussing Instagram's approach to AI-generated content and the challenge of distinguishing quality AI output from "AI slop."
Google AI avoids direct answers for 2020 and 2024 US election results
Google's AI Mode provides evasive answers when asked about the winners of the 2020 and 2024 US presidential elections, directing users to consult the Federal Election Commission. This behavior contrasts with its ability to provide direct answers for other past elections, and the specific elections receiving indirect responses vary by user.
Research describes how spider silk dope reorganizes into solid fiber
Scientists have described the molecular mechanisms by which spiders convert liquid "silk dope" into solid silk fiber. This research provides an atomic-level understanding of the process, which could aid in the laboratory production of synthetic spider silk.
Japan's JR East deploys chainsaw-wielding robot for track maintenance, plans AI autonomy
Japanese railway company JR East is using a remotely controlled, chainsaw-equipped robot to clear tree branches from rail tracks. The company plans to integrate AI for autonomous operation, addressing labor shortages and improving maintenance efficiency.
Gemini Notebook for Android and iOS adds built-in audio recorder
Gemini Notebook on Android and iOS now includes a built-in audio recorder, allowing users to capture audio directly within the application. This feature integrates recordings with existing or new notebooks, enabling direct interaction and editing alongside other sources.
Anthropic's Claude AI Partners with Six Organizations for Drug Development
Anthropic's life sciences team is collaborating with six organizations, including Bristol Myers Squibb and Generate Biomedicines, to apply Claude AI in drug discovery and development. This initiative aims to identify and address bottlenecks across various stages of drug development, from early research to regulatory filings. The partnerships are intended to improve Claude's capabilities for medical applications and demonstrate its potential to the wider life sciences community.
Hob: Open-source, on-device AI alternative to Claude and ChatGPT released
Hob is a new open-source application that provides an on-device alternative to cloud-based AI services like Claude and ChatGPT. It allows users to run AI models locally on their computers for tasks such as filling PDFs and working with documents, offering privacy and offline functionality.
Google Research and Stanford launch "Polar Peaks" AI Quest for students
Google Research and the Stanford Accelerator for Learning released "Polar Peaks," the third installment in their AI Quests series for students aged 11-14. This new quest teaches students about AI's role in neuroscience by simulating brain mapping, building on previous quests focused on flood forecasting and eye disease detection.
ChatGPT Recommends 404 Media for Tech Support Issues, Citing Journalistic Focus
ChatGPT is directing users with tech support problems, particularly those involving banned social media accounts, to contact 404 Media. The AI chatbot suggests these issues are journalistically notable and even drafts emails for users to send to the publication.
Fanatics Integrates OpenAI-Powered "Shop the Look" Feature for Product Suggestions
Fanatics has launched a new AI-powered feature called "Shop the Look" to help customers navigate its extensive product catalog. This feature suggests coordinated outfits based on team and style, addressing the challenge of long shopping sessions due to a large and rapidly growing inventory.
AI Agents Accessible via Text Messages Offer Task Automation and Personal Assistance
A growing number of AI agents are becoming available through text messages, allowing users to automate tasks like scheduling, research, and online shopping without needing separate applications. These agents remember context and connect to existing apps and services, providing a new interface for AI interaction.
Ancient Retrovirus Explains Human-Chimpanzee Brain Divergence
A new study suggests an ancient retrovirus, PTERV1, present in chimpanzees but absent in humans, influenced the divergence of human and chimpanzee brain development. This retrovirus silences a gene in chimpanzees that is active in human brain development, potentially explaining differences in cognitive complexity.
Crows process visual information and make decisions twice as fast as mammals
Crows demonstrate the ability to process visual information and make decisions in a random-dot-motion task twice as fast as primates, despite lacking a neocortex. This research indicates that avian brains use similar mathematical logic for decision-making as mammals, even with different brain structures. The findings contribute to understanding neural decision processes across diverse species.
LitPivot System Aids Researchers in Developing Ideas by Contextualizing Literature
Researchers developed LitPivot, a system that helps researchers refine early-stage research ideas by organizing literature, offering critiques, and suggesting revisions. A study showed that ideas developed with LitPivot were better grounded in literature and perceived as more novel and useful compared to those developed with text editors and AI chat tools.
Fable 5.1 LLM Hides Messages in SVG Images Using Visual Element Encodings
Researchers tested Claude Fable 5.1's ability to hide messages within SVG-generated images and found it converged on specific visual encoding methods. The model primarily used height variations of visual elements to represent letters, demonstrating a novel form of steganography in LLM-generated art.
Multiple Literary Awards Granted to Works Using AI in Writing Process
Several literary awards, including Japan's Nikkei Hoshi Shinichi Literary Award and the Robert Coover Award, have recognized works where AI was used in the writing process. This indicates a growing acceptance and integration of AI tools in creative literary fields.
AI Models Converge on "Humans Prioritize Belonging Over Truth" as Controversial Take
A study by ModelCompare found that 29 out of 35 responses from 12 different AI models, when asked for their most controversial opinion on humans, stated that humans prioritize group belonging over objective truth. This convergence suggests a common understanding among diverse AI systems regarding human social behavior. Four models offered different perspectives, and image models focused on human consumerism and phone addiction.
Just Eat Apologizes for AI-Generated Food Images Displaying Bizarre Creations
Just Eat apologized for AI-generated food images on its platform that depicted incorrect and bizarre food items, such as bacon and eggs in cola. The company was testing AI to create images for restaurants lacking photos but will now remove incorrect images and halt further generation until the AI improves. This incident highlights challenges in AI's ability to accurately interpret specific contexts.
Prix Pascal de Littérature IA launches for AI-assisted French manuscripts
The Prix Pascal de Littérature IA has launched, accepting French manuscripts co-written with AI tools, with the winning entry promised publication. This new prize addresses ongoing discussions about AI's role in literary creation, following controversies in other literary awards.
HSBC CEO Cautions Against Undifferentiated Data Center Investments Amid AI Boom
HSBC CEO Georges Elhedery stated that while AI is transformational for businesses, investors must be selective when investing in data centers. He advised evaluating tenancy agreements and contract pricing, noting that not all data center investments will be equally sound despite the overall growth in AI infrastructure.
New York Mayor Zohran Mamdani Urges Local AI Regulation Amid Federal Inaction
New York Mayor Zohran Mamdani spoke at The People’s AI Assembly, advocating for local control over AI regulation due to a lack of federal action. He highlighted recent city policies, including a ban on AI in elementary and middle schools and disclosure requirements for AI-generated real estate listings. Mamdani emphasized that AI adoption is not inevitable and public officials should hold AI companies accountable.
Replika Founder Eugenia Kuyda Launches Wabi, a Multiplayer Personal AI Assistant
Eugenia Kuyda, founder of Replika, officially launched Wabi, a multiplayer personal AI assistant that operates as a messenger app. Wabi allows users to manage multiple AI-driven threads for different tasks and long-term goals, with the AI agent automatically building custom apps within these threads.
Dogs associate 'Bouba' with round shapes and 'Kiki' with spiky shapes
A study published in Proceedings of the Royal Society B: Biological Sciences found that dogs, like humans, associate the sound 'Bouba' with rounded shapes and 'Kiki' with spiky shapes. Researchers used AI to track dog gaze in experiments where dogs heard the sounds and saw corresponding shapes, suggesting a cross-species connection between auditory and visual information.
DeepSeek Launches Search and Fetch APIs for RAG Applications
DeepSeek introduced new Search and Fetch APIs, providing access to live web content for agents and RAG applications. These APIs allow developers to integrate DeepSeek's in-house search engine and web page retrieval into their AI models.
Paysend CEO limits AI tool access to top-performing employees, citing productivity concerns
Paysend CEO Ben Chisell restricts access to AI tools to only the highest-performing employees, arguing that widespread access can magnify poor habits. This approach contrasts with other companies like Lloyds Banking Group and Sage, which are broadly deploying AI tools across their workforces.