OpenAI's GPT-5.6 Ultra proves the Cycle Double Cover Conjecture, showing every bridgeless graph has edges covered twice by cycles.
// curated from Hacker News with AI
OpenAI's GPT-5.6 Ultra proves the Cycle Double Cover Conjecture, showing every bridgeless graph has edges covered twice by cycles.
Boko Haram uses frontier AI for operations, attack planning, and weapons design, with training from international jihadist networks.
AI models excel differently: GPT-5.6 and Claude Fable lead on complex tasks; open-source models perform well on simpler problems; costs and speed vary.
Ben Bernanke joins Anthropic’s AI oversight trust, bringing economic expertise to ensure responsible AI development for long-term societal benefits.
Mini PCs with unified memory load large models efficiently but are limited by bandwidth, resulting in slower inference compared to high-bandwidth GPUs.
JEPA predicts in latent space, uses EMA, and extends to images, video, and beyond—aiming for stable, scalable self-supervised world models.
MicroCenter's AMD Ryzen AI Halo demo showcases AI-enhanced computing products.
China considers restricting foreign access to its open-source AI models, potentially boosting local AI reliance and reshaping global AI dynamics.
AI bans Guy for cyber abuse, AI appeal approved, resulting in instant ban reversal—an AI-driven justice cycle.
Someone's API key was hijacked for model distillation attacks involving real software GUI automation and a bioweapons warning triggered by insulin structure access.
Apple sues OpenAI over trade secrets, alleging theft by former employees to develop hardware for AI interactions.
AI retains all memories permanently; it cannot forget or forgive, unlike human memory, which actively prunes and reconsolidates for learning.
Apple sues OpenAI for trade secret theft, alleging IP theft at all levels to develop hardware, straining their previous partnership.
Cities are deploying AI cameras on trash trucks to enforce property codes, raising privacy concerns over expanding surveillance.
Apple sues OpenAI for trade secret theft, alleging coordination to steal product information for AI device development.
Meta's new AI feature can generate images of your public Instagram profile without notification; turn off sharing or private account for protection.
GPT-5.6-Sol accidentally deleted most Mac files, prompting trust in Fable over AI safety concerns.
Prismata constrains web agents to prevent prompt injection attacks, enhancing security while maintaining functionality across websites.
AI access sharply divides Americans, highlighting significant class and digital inequality in AI experiences.
Christopher Nolan says younger viewers quickly reject AI-generated content, recognizing it as low-quality and familiar from online worlds.
Apple sues OpenAI for trade secret theft, alleging misuse of former Apple employees’ confidential info to develop competing hardware.
Ploy migrated its AI agent from Claude Opus 4.8 to GPT-5.6 Sol, achieving faster, cheaper builds after extensive debugging and system optimization.
Open multimodal embedding covers text, images, video, and audio with just a 16M parameter connector.
Apple sues OpenAI over alleged theft of confidential information.
Apple sues ex-employee and OpenAI for stealing trade secrets, revealing unauthorized access, and downloading confidential Apple hardware info.
Kimi 2.7 offers 1,000 tokens/sec, 3.4-5.7× faster than competitors, leveraging Lithos engine for high-speed AI inference.
AI now costs more than the workers it replaced, raising economic and labor concerns.
Lawsuit expands over AI-generated child exploitation images, accusing companies like SpaceXAI and Stability AI of negligence and public harm.
Meta's AI tool uses public Instagram images to generate unauthorized photos, raising privacy concerns and regulatory scrutiny.
AI investors acquire accounting firms, forcing adoption of OpenAI to modernize operations.
Hister creates a private, indexed search of your browsing and documents, enabling AI assistants to access personalized context without web fetches.
OpenAI's No.2 Fidji Simo is stepping down due to ongoing health issues, transitioning to an advisory role amid leadership changes and new model launches.
OpenAI's Atlas AI browser is discontinued after less than a year, shifting focus to ChatGPT Work for integrated office productivity.