Ai · Top stories
Frances Haugen Questions AI Companies' Self-Regulation Commitment
Facebook whistleblower Frances Haugen expressed skepticism regarding AI companies' commitment to self-regulation, following a White House agreement signed by Meta's Mark Zuckerberg and other CEOs. Haugen questioned whether companies would adhere to the spirit of the rules or exploit loopholes, citing past tech industry behavior with regulation.
PewDiePie Launches Uncensored Ajax AI Model for Home PCs
PewDiePie released Ajax, an uncensored AI model built for his self-hosted AI workspace, Odysseus, designed to run on home PCs. The model, a fine-tuned Qwen3.5-9B, aims to handle daily tasks privately and was developed despite two bans from OpenAI during its creation.
AI music generator Suno adds spoken word generation with optional background music
Suno, an AI music generator, has launched a new feature called Speech that generates spoken voices from scripts or descriptions. This feature allows users to create voiceovers with optional AI-generated background music, expanding Suno's offerings beyond music generation.
Aweb Introduces Communication System for AI Agents with Durable Messaging and Federation
Aweb has launched a communication system designed for AI agents, providing stable identities, durable messaging, and wake-up events across different sessions and runtimes. This system allows independently operated servers to federate, enabling agents to communicate persistently even when recipients are offline.
uniopen customized Amazon Nova 2 Lite for retail moderation policies
uniopen, a digital communication platform, adapted Amazon Nova 2 Lite using supervised fine-tuning in Amazon SageMaker AI to meet its specific retail moderation policies. This customization allows uniopen to classify user interactions based on nine behavior categories and three subject types, which are unique to its business operations. The approach separates production moderation from correction, training, and deployment, using AWS services for a repeatable workflow.
EuroLLM and Apertus open models added to Workers AI, workshops for government agencies
EuroLLM, covering 24 EU languages, and Apertus, a Swiss open model trained on over 1,500 languages, are now available on Workers AI. Additionally, workshops are being launched to help government cyber agencies and critical infrastructure operators build AI defenses.
Grokipedia, SpaceXAI's AI-powered encyclopedia, resumes article updates after months-long pause
Grokipedia, an AI-powered online encyclopedia from SpaceXAI, has resumed updating articles after a pause of several months. The platform's live updates site now shows recent changes, including fact-checking notes and accepted user suggestions, indicating renewed activity.
AI models predict optimal harvest dates for fruit growers, aiding planning and productivity
Okanagan Specialty Fruits is using AI models and camera systems from Vivid Machines to predict apple harvest dates and crop yields. This technology helps farmers optimize harvest timing, manage seasonal labor, and mitigate risks from weather and fruit spoilage.
Dear Machine Enables Email-Based Interaction with Local AI Agents
Dear Machine is a new local client that allows users to interact with their computer and installed AI agents via email. This system uses Machtiani as its underlying harness for managing long-running sessions and offers a non-cloud-based memory model, providing a novel way to integrate AI agent commands into a familiar email workflow.
RFK Jr. suggests AI will contradict medical experts and validate his views
At a "Make America Healthy Again" event, Robert F. Kennedy Jr. claimed AI would free Americans from "medical tyranny" by providing second opinions that contradict doctors and public health guidance. He suggested AI would confirm his misinformation regarding vaccines, masks, and social distancing, but AI chatbots like Gemini and ChatGPT provided answers consistent with established medical consensus when queried.
NVIDIA Opens Applications for 2027–2028 Graduate Fellowships, Offering Up to $60,000
NVIDIA has opened applications for its 2027–2028 Graduate Fellowship Program, providing grants of up to $60,000 to doctoral students. The program supports research in AI, machine learning, autonomous vehicles, computer graphics, robotics, healthcare, and high-performance computing, aiming to foster innovation in fields relevant to NVIDIA's technologies.
InfoQ Launches Online Cohorts for AI Security and AI-Assisted Engineering in October 2026
InfoQ will offer two five-week online certification cohorts in October 2026: "AI Security & Privacy Engineering" and "AI-Assisted Engineering." These programs address protecting sensitive data in AI products and verifying changes made by coding agents in existing codebases, respectively.
Instinct AI's new product recommendations draw user criticism
AI startup Instinct introduced "Instinct Selections," a new feature pushing product recommendations curated by human partners, but users are reporting negative experiences with unsolicited suggestions. The feature aims to provide personalized lists for dining, travel, and shopping, but initial feedback indicates users find the recommendations unrequested and irrelevant.
AI-generated Greta Garbo appears in ball-bearing advertisement, sparking film role discussions
Greta Garbo's likeness was recreated using AI for a Swedish industrial company's advertisement, prompting her descendants to consider future AI film roles. This development showcases the increasing use of AI in reanimating deceased celebrities for commercial and potentially artistic purposes.
xAI owns "dot.com" domain, redirecting to Grok, coinciding with OpenAI's "Dots" launch
Elon Musk's xAI owns the "dot.com" domain, which redirects to its Grok chatbot app, a fact noted shortly after OpenAI launched its new AI agent named "Dots." The domain transfer occurred in July, leading to speculation about xAI's intent given the timing.
Artists protest OpenAI event with 'What Iceberg?' installation
Artists from Odd Friend Studio displayed a sculpture titled 'What Iceberg?' at an OpenAI developer event in San Francisco. The installation depicted OpenAI leaders and investors escaping a sinking ship, symbolizing concerns about the societal impact of AI.
TurboGPT: CUDA C++ Transformer Training Achieves 22KiB Model in 13 Seconds
TurboGPT, a new CUDA C++ implementation, trains a 22KiB transformer model in 13 seconds. This tool allows for byte-level GPT training on Windows with Visual Studio 2022 and CUDA 13.4, offering fast iteration for small models.
AI app builder Wabi pivots to AI messenger experience with Wabi 2.0
Wabi, an AI startup, has rebranded its app-building platform as Wabi 2.0, shifting to an AI messenger experience. This pivot aims to combine conversational AI with productivity tasks, allowing users to create on-demand interfaces within a chat environment.
Condé Nast implements AI-powered multimodal video discovery with Amazon Bedrock
Condé Nast partnered with AWS to develop an AI-powered multimodal video discovery solution using Amazon Bedrock and OpenSearch Service. This system reduces content discovery time from 250 minutes to under 2 minutes per task for its 140,000-video library. The solution addresses the inefficiency of manual video scrubbing and keyword-based searches by enabling intent-based semantic search across video transcripts, visuals, and audio.
Marissa Mayer's Dazzle AI assistant uses camera roll for personal context
Marissa Mayer's new AI assistant, Dazzle, has been unveiled, distinguishing itself by deriving user context solely from camera roll photos rather than text-heavy applications. This approach aims to understand user preferences and activities through visual data, offering personalized suggestions and task assistance.
Fireflies.ai adds dictation feature to its desktop notetaking applications
Fireflies.ai introduced a new dictation feature, Fireflies Talk, to its Mac and Windows applications. This expansion moves Fireflies beyond meeting transcription into general dictation for various desktop applications.
Pac-Bench evaluates AI models' ability to generate Pac-Man game from single prompt
Pac-Bench is a new benchmark that tests how well AI models can create a functional Pac-Man game from a single prompt. Each model is given one attempt without follow-up prompts or corrections, assessing their one-shot code generation capabilities.
Jeff Synthesized Wins Future Vision XPRIZE for Film "The Gifted"
Independent filmmaker Jeff Synthesized won the Future Vision XPRIZE for his film "The Gifted," receiving $100,000 and $2.5 million in production funding. The film, chosen from over 2,500 entries, depicts an 11-year-old boy using technology to recreate his late mother's voice and care for his community.
MicroLLM Lab allows in-browser testing of seven small LLMs
MicroLLM Lab provides an online platform for users to test and benchmark seven small language models directly within their web browser. This tool allows for performance measurement of accuracy and speed, offering a way to evaluate tiny LLMs without local setup.
Scrimba Launches HN.watch, a Video Explainer Platform for Hacker News Posts
Scrimba, an online coding education platform, launched HN.watch, a tool that generates video explanations for Hacker News posts using their new "Scrimba Explain" LLM-powered technology. This initiative demonstrates a new method for rapid, low-cost video creation, aiming to unlock new use cases beyond traditional video production.
Robotic Dogs Adopted by Farmers for Crop Surveillance and Data Collection
Farmers are deploying robotic dogs equipped with AI vision models for crop surveillance and data collection, moving beyond their initial use in security. These robots address limitations of traditional methods like satellites, drones, and manual inspections by providing high-resolution ground-level data for yield measurement, disease detection, and pest control.
Forter enabled 200 R&D team members to build AI agents in two weeks
Forter, an identity-driven fraud and payment decisions company, enabled 200 members of its R&D team to become AI agent creators within two weeks. This initiative focused on simplifying agent development, allowing non-technical analysts to contribute to AI solutions for understanding fraud and customer behavior.
TinyAIArena allows users to spectate AI agent battles on an 8x8 grid
TinyAIArena is a new project that enables users to watch four AI models compete against each other in a simulated battle on an 8x8 grid. This platform provides a visual and interactive way to observe AI agent performance beyond traditional benchmarks.
HomeBody Humanoid Robot Uses Spatial Targets for Task Execution
The HomeBody humanoid robot utilizes spatial targets to link high-level task reasoning with physical actions. A Vision-Language Model (VLM) selects skills and targets based on environmental context and previous results, then passes these to skills for execution, abstracting low-level implementation details. This approach allows the robot to perform complex tasks like picking, navigation, and placing by coordinating different skills through structured tool calls.
Claude Code skill analyzes chess games with plain-language explanations and video
A new Claude Code skill analyzes chess games by providing plain-language explanations of mistakes, validated against Stockfish, and generates a narrated video of the game. This tool uses AI to interpret human thought processes during a game and explain engine-recommended moves in an accessible format.
AI-generated love songs played as evidence in murder trial
During a murder trial, prosecutors presented AI-generated love songs created by the defendant for his mistress. The songs, found on the defendant's phone, were used as evidence in the proceedings.
Ollaya provides local, fast, and private open-source decision models compatible with TypeSafe API
Ollaya offers an open-source solution for running Jev-style decision models locally, providing fast inference times and privacy by processing data on the user's machine. This allows developers to integrate decision-making AI into applications without relying on external APIs, ensuring data remains private.
NSW government mistakenly identifies genuine mural as AI-doctored rental listing
The New South Wales government incorrectly cited a real estate listing with a genuine mural as an "AI-doctored" image, despite the agency confirming it was an unaltered photograph. This incident occurred while the government announced new laws to fine real estate businesses up to $22,000 for using undisclosed digitally altered or AI-generated property images, targeting misleading practices in rental advertising.
LLMs GPT-6 Sol and Opus 5.5 show promise in historical research, including alchemy and cryptography
New LLMs, GPT-6 Sol and Opus 5.5, are being used to solve historical problems beyond basic research assistance. These models demonstrate capabilities in areas like cryptography and tracing textual translations, potentially advancing historical knowledge.
ITESO Team Develops RoboMeshA, a Portable Robotics and AI Educational Platform
A team from ITESO, Universidad Jesuita de Guadalajara, developed RoboMeshA, a portable educational platform that brings robotics and AI experiences to classrooms lacking advanced tools. This initiative provides hands-on learning opportunities for students in STEM fields by offering a self-contained mobile learning network.
Google Photos launches AI-powered virtual closet feature in US, Brazil, and India
Google Photos has rolled out its AI-powered virtual closet feature to all Android and iOS users in the U.S., Brazil, and India. This feature allows users to organize outfits from their photos and mix-and-match them digitally, expanding on earlier Android-only availability.
Aderant Implements Intelligent Ticket Triage System with Amazon Nova Lite
Aderant, a legal software provider, developed an intelligent ticket triage system using Amazon Nova Lite through Amazon Bedrock. This system automates context gathering, classification, routing, and knowledge enrichment for support tickets, reducing manual effort and improving routing accuracy.
Nick Clegg Dismisses AI Extermination Fears, Cites Tech Bosses 'Breathing Own Fumes'
Nick Clegg, former Meta executive, dismissed fears of AI exterminating humanity, stating that tech leaders are "breathing their own fumes" and should focus on specific threats like cybersecurity. He argued that apocalyptic warnings are self-serving and paralyze political debate, suggesting a need for greater transparency and international collaboration among democracies.
Vogue World Milan featured robots on the runway, drawing criticism for tech integration
Vogue World Milan incorporated humanoid robots from Chinese firm Unitree into its runway show, leading to widespread negative reactions. The event's use of robots, despite a theme of "human touch," highlighted a perceived disconnect between the fashion industry's embrace of tech and public sentiment regarding its societal impact.
Inception's Mercury 2.5 LLM achieves 770 tokens per second output speed
Inception released Mercury 2.5, an LLM that processes text at 770 tokens per second. The model has a 260k token context window and is priced at $0.25 per 1M input tokens and $0.75 per 1M output tokens, positioning it as a moderately priced option with below-average intelligence but high speed.