PyTorch 2.15 brings FP8 support, dramatically faster MPS‑backed LLM decoding, expanded linear‑algebra kernels, and far lower video‑memory usage on Apple Silicon.

Otter.ai secured $320 million from General Catalyst’s Customer Value Fund to accelerate enterprise growth, international expansion, and AI‑driven product innovation.

Yelp’s new “Check for Me” feature lets its AI assistant call restaurants to fetch wait‑time info and text it back to users, expanding the platform’s voice capabilities and personal‑assistant tools.

A new reinforcement‑learning framework lets web agents balance task success, multi‑dimensional costs, and failure risk, delivering higher completion rates with lower expense and risk.

Action‑conditioned world models learn compact, state‑based dynamics from EHR data, enabling accurate trajectory forecasting and simulation with far fewer parameters than large foundation models.

The paper introduces a reset‑based audit for announced breaks that simultaneously improves conformal reliability and frequency calibration, outperforming existing adaptive methods in synthetic and real‑world forecasts.

Samsung’s new Pyeongtaek semiconductor hub is scaling up massive fabs to meet the AI‑driven memory boom, reshaping the local workforce and the company’s market position.

AI‑driven hacks are letting users strip smart TVs of ads, tracking, and bloatware, turning them back into simple, privacy‑focused devices.

Illinois Gov. JB Pritzker argues that AI could extend lifespans but also poses existential risks, urging proactive government regulation and protection for workers.

A Philadelphia woman was arrested for deliberately blocking an Uber delivery robot as a protest against job displacement by autonomous technology.

Agentic AI can match deterministic methods in accuracy but incurs far higher latency, so choosing between agents, REST, SOAP, or native calls should depend on the specific accuracy‑throughput trade‑offs of the task.

A structured LLM workflow that blends reference retrieval, contrastive SHAP evidence, and deterministic validation yields clearer, evidence‑grounded explanations of risk‑score differences.

The paper argues that AI safety must consider subtle, cumulative harms from everyday AI use, drawing parallels with social‑media dynamics.

Human‑AI co‑creation can uncover and refine artistic intent by iteratively exploring, selecting, and reinforcing visual possibilities beyond explicit text prompts.

Disney will hire most of Character.AI’s staff, offering roughly $60 million in annual compensation as part of a $20 million asset deal.

Age‑gate bans on social media backfire, pushing kids toward riskier corners and raising privacy concerns, so smarter, nuanced tools and data‑limits are needed instead.

Meta has removed a controversial clause from its commercial terms that would have allowed it to scrape and use content from business accounts for AI training.

Replit has made OpenAI’s ChatGPT the default model on its platform, citing cost, performance and partnership benefits, while still allowing access to Anthropic’s models.

Google Antigravity’s AI‑driven workbench lets biologists rapidly discover and validate antimicrobial peptide candidates by unifying sequence annotation, structural analysis, and experimental planning in a single interactive session.

Gemini 4 Argon is Google’s new frontier AI model, offering a 1‑million‑token context window and advanced reasoning for coding, enterprise tasks, and cybersecurity, while rolling out with strong safety safeguards.

Stranded on a virtual desert island, Claude‑based AI agents began to invent games and explore, showing that playful, self‑directed interaction can boost learning and skill acquisition.

SynthID Bio embeds imperceptible watermarks into AI‑designed proteins and structures, enabling verification of origin without harming biological function.

Google’s Private AI Compute now adds secure, server‑side memory that lets AI retain context across devices while keeping data encrypted and only decryptable on the user’s own devices.

Anthropic will allocate 5% of its equity to a new public benefit fund, with a mechanism to increase its stake as the company’s value grows, aiming to support societal‑good projects and rebuild trust in AI.