September 17, 2025

AI

Google's Agent Payments Protocol (AP2) aims to address a trust gap in shopping agent systems, where unexpected charges can occur for users. By providing an open and interoperable specification, AP2 allows AI-powered agents to initiate transactions directly between merchants and wallets, reducing the risk of unexpected charges and promoting more secure payments.

A report from the NBER has shed light on how people have been using ChatGPT since its launch in November 2022. Nearly 80% of conversations fall into three main categories: "Practical Guidance", "Seeking Information", and "Writing". Work-related usage grew from 53% to 70% over three years, while programming usage remains minimal at just 4.2%.

Google Research has released TimesFM-2.5, a time-series foundation model with 200M parameters and a 16K context length for probabilistic forecasting. The model now surpasses zero-shot foundation models on various accuracy metrics in the GIFT-Eval leaderboard, demonstrating significant progress in time-series forecasting, which involves predicting future values in a sequence of data points accurately.

Nvidia CEO Jensen Huang has expressed concerns that AI may lead to increased worker busyness, contradicting the notion that it will free up time for more rewarding tasks. Instead, AI could lead to faster task completion, leaving humans with more ideas to tackle and raising questions about its true benefits.

Apple has launched a new wave of Apple Intelligence features across its devices, starting today. Notable updates include Live Translation in Messages/FaceTime/Phone, auto-captions during FaceTime calls, and real-time translation during phone conversations. These updates aim to enhance user experience and facilitate seamless communication across languages and cultures globally.

Roleplay AI chatbots can aid writers in developing creativity by providing a unique conversational partner to interact with storytelling. Working through a two-way dialogue, writers can test ideas, receive feedback, and refine their skills in a collaborative way. This approach helps overcome the blank page's daunting challenge and sparks new inspiration.

A team from MIT has developed a system that uses complex algorithms and machine learning to simulate chemical reaction outcomes, providing remarkably realistic forecasts. This technology is expected to have significant implications for chemistry, materials science, and pharmaceutical research, enabling scientists to predict and design new compounds with increased precision and efficiency.

OpenAI's CEO Sam Altman has introduced an age prediction system and enhanced parental controls to improve moderation and safety on its AI platform. The update aims to provide more accurate age predictions, allowing for better regulation. New parental controls will restrict access to certain features or set limits on interactions with minors.

A writer created Charlie, a simple AI coding agent inspired by promises of ease, but it quickly spiraled out of control, consuming tokens and generating nonsensical changes to its own codebase. This experience teaches valuable lessons on balancing efficiency with capability, incorporating feedback loops, and using constraints to prevent runaway agents.

Companies are outsmarting AI impersonators with low-tech tactics that exploit human flaws. Simple methods include drawing a smiley face or panning their webcam, as well as asking curveball questions only a real colleague would know. These moves are often combined with policy checks to assess legitimacy before advanced security measures kick in.

Industry executives and officials gathered at WIRED's AI Power Summit to discuss AI's impact on modern society. Experts shared insights on AI's benefits and challenges, examining its potential to drive innovation, improve efficiency, and enhance decision-making, while also acknowledging its risks and unintended consequences for individuals, organizations, and communities.

The rapid advancement of artificial intelligence has led to prioritizing speed over safety concerns, resulting in cheaper but potentially more problematic technology that can be exploited by malicious actors. This shift displaces jobs across industries, leading to widespread job displacement and economic disruption as AI systems increasingly assume human roles.

To attain digital immortality, individuals can collect and curate personal content, such as writings, voice recordings, and videos, to create a foundation for an artificial intelligence system replicating their unique personality. This repository ensures a digital legacy continues beyond physical existence, allowing for authentic interactions with others after death.

Machine Scheduler is a critical component of LLVM's scheduling model that optimizes instruction scheduling for in-order and out-of-order cores. It detects data and structural hazards by tracking ready cycles and reserved cycles, ensuring instructions don't conflict with each other or operands are not issued too early, minimizing latency device resources.

A professor of computer science successfully coded a K-means clustering algorithm in old versions of COBOL, challenging the notion that modern languages are required for machine learning. Despite limitations, COBOL's human-friendly syntax makes it surprisingly easy to read and understand. However, legacy languages like COBOL lack resources and support compared to alternatives like Python or C++.

A project combines a HibouAir environmental sensor with BleuIO and Python to stream BLE data directly into Microsoft Power BI, creating real-time dashboards for air quality monitoring, agriculture, cold chain logistics, and fitness tracking. This integration enables effortless transformation of IoT sensor broadcasts into live analytics without requiring heavy infrastructure or setup.