Interventions like auxiliary classifiers reduced phone discrimination error from 11.6% to 10.4% in English/French HuBERT models.
The team now pushes more than 73,000 deployments a month using six production AI agents.
The study reveals how TokenGT models reconstruct graph incidence through node-ID matching to solve tasks like degree calculation and shortest-path distance.
The study elucidates the principle of local structure restoration via attention heads in basic graph tasks such as degree calculation.
The 50% increase applies only to the 5-hour usage limit, not the weekly limit, for conversations started with a design, deck, or doc in the Claude app.
GPT-6 Astra Ultrafast is now available in the OpenAI API and to eligible ChatGPT Work and Codex users, offering up to 8x faster token generation than Astra Standard.
The prototype satellite launched aboard SpaceX's Transporter-18 mission and is currently gathering data on TPU performance in space.
The highest composite score on the 117-item held-out split was 47.3, achieved by Qwen3.8 Max with OpenCode, highlighting significant gaps in AI agents' ability to maintain expert-level documentation.
The feature is currently in public beta for eligible ChatGPT Plus, Pro, Business, Enterprise, and Edu users.
Aweb introduces an MIT-licensed communication protocol for AI agents, offering durable messaging, stable identity, and wake-up events with integrations for Claude Code and Pi.
Resolved a connection pool exhaustion outage caused by stdout buffer saturation using a custom Appender, achieving both asynchronous logging and trace_id preservation.
Graphene is a data analytics framework designed for coding agents, featuring a semantic SQL layer and dashboard file type to enable accurate, consistent data visualizations.
Dynamic workflows allow users to define multi-step tasks combining automated steps and agents, now available in the Copilot CLI and app.
The AI-powered agent analyzes infrastructure across 65+ AWS services to provide prioritized, context-aware remediation steps.
FLUX 3 Image supports native 4K rendering and allows precise layout control via bounding boxes and up to 10 reference images.
Effective October 1, 2026, arXiv limits authors to two submissions per calendar month and three active submissions at any time to manage moderator workload.
Josh Imbue argues that open-weight models and local data ownership can break Big Tech's lock-in and restore user control over digital environments.
The departures follow an internal investigation confirming the researchers mishandled sensitive company information outside established procedures.
An independent essay hosted by OpenAI argues that while AI may accelerate theoretical progress, the pace of a 'Width' civilization would be defined by physical execution bottlenecks.
The new middleware reduced median cost per coding task by 64% in Open SWE experiments without impacting quality.
The plugin analyzes recent sessions to identify recurring mistakes and injects corrective lessons, raising correct handling rates from 8% to 91%.
CodeAF has launched as an open-source coding harness designed to optimize performance and cost for open models, claiming top rankings on the DeepSWE benchmark.
Finbar has launched public availability of its investment research platform, offering worldwide fundamental data via API/MCP and AI tools for modeling, previously used exclusively by major hedge funds.
Meta may avoid in-app advertising for its Muse agent by leveraging indirect monetization loops where Muse's browsing activity informs ads on Facebook and Instagram.
AI Briefing