Topic
Explore the latest artificial intelligence news and research.
ChatGPT advertising has scaled to a $1 billion annualized run rate. This piece examines the self-service purchasing, click-based pricing, and performance measurement that built the sales infrastructure—and the limits the numbers don't reveal.
A theoretical paper in Trends Open classifies biological naturalism, the view that denies machine consciousness, into an unfalsifiable Type-A and a Type-B that can coexist with computational functionalism. It lays out requirements for scientific testability grounded in falsifiability.
Drawing on reports of bulk Mac purchases by OpenAI and others, this piece examines how, for computer-use agents, the bottleneck in AI training is shifting beyond raw compute to the supply of isolated desktop environments.
Chinese DRAM giant CXMT is reportedly producing small quantities of HBM3E, with Alibaba's T-Head and Cambricon testing it for adoption. Before any 2027 deployment, customer qualification, yield, and supply volume still need to be confirmed.
OpenClaw 2.0 keeps conversations and credentials on the Gateway while letting processing move to local devices or the cloud. It supports shared work, but strict user isolation still requires a separate Gateway.
NVIDIA's NVL72 market is shifting toward a structure in which rising per-rack prices from the Vera Rubin generation transition, combined with long-term guarantees and third-party capital formation, underpin the rollout of costly AI computing infrastructure.
Researchers from Meta AI and UIUC propose EvoHarness-RL, which trains AI agents to decide when to access external memory. On ALFWorld's in-distribution 140 tasks, Qwen3-8B achieved a 96.9% average success rate.
Stolen, already-authenticated Claude sessions from infected devices were exploited to consume users' usage limits without their action. Here's how this bypasses multi-factor authentication and which credentials need to be revoked during recovery.
This analysis breaks the 35 Sony- and Warner-affiliated publishers' complaint into four layers—pirated acquisition, model training, lyrics output, and copyright management information—to clarify what the Bartz ruling covers and what remains unresolved.
Claude Code's permanent weekly limit is 25% higher than the old standard, but because the temporary 50% boost ends, users will see about 17% less than current levels. Here's how the scope, the undisclosed absolute limits, and the 5-hour cap differ.
