OpenAI's Terminal X Matches Senior Investor Judgment at 88.6%
OpenAI's Terminal X evaluation agent matches senior investor judgment 88.6% of the time and catches 25% more fund mandate deviations using a 54-criterion rubric.
AI & ML
Frontier models, research results, agents, and developer tooling.
OpenAI's Patch the Planet Coordinates Open-Source Vulnerability Response
The initiative links researchers, maintainers, and incident responders so critical open-source flaws found by agents can be patched quickly.
Inside OpenAI's LLM-Based Ad Stack for ChatGPT
OpenAI details how it uses language models to serve relevant ads while keeping them distinct from ChatGPT's answers.
Claude Mods Turn the Agent Harness Toward Human Attention
An early 'Attention Interface,' Claude Mods let users reprogram how Claude Code surfaces information and requests human input.
OpenAI aligns safety policies across model access services
OpenAI is standardizing safeguards across direct and partner access to its models, with updated usage policies taking effect November 20, 2026.
Sim-to-Real Walking for Microduck with Isaac Lab 3.0
A new tutorial shows how to train a Microduck biped robot to walk in simulation and deploy the resulting policy on real hardware.
AI blind-sweep ultrasound matches expert gestational age accuracy in field study
A prospective study shows that AI interpretation of blind-sweep ultrasound by non-sonographers can deliver gestational age estimates non-inferior to expert sonographers in diverse clinical settings.
NVIDIA Kumo Relational Beats Supervised ML on Relational Data
NVIDIA's pre-trained relational foundation model makes predictions directly on databases, outperforming supervised pipelines without task-specific training.
AT&T uses OpenAI to give lawyers an AI-assisted intake system
AT&T is extending its LegalEdge program with OpenAI to connect in-house lawyers to company knowledge and streamline how new legal requests are handled.
ChatGPT Ads expands to 26 new countries, reaching 89 total
OpenAI says its ad platform now spans 89 countries and has attracted over 100,000 advertisers since launching eight months ago.
Vercel Open-Sources durable-harness for Crash-Proof Coding Agents
Vercel's new open-source package logs every model and tool call, letting coding agents resume seamlessly after crashes, timeouts, or redeploys.
Apple Music Rebrand Video Shows Why Human Craft Matters in AI Era
An Ars Technica columnist argues that Apple Music's behind-the-scenes brand video highlights the growing value of visible human effort as generative AI floods the visual landscape.
Judge rejects EPA bid to dismiss data center pollution suit against Musk's AI
A federal judge ruled that the EPA cannot unilaterally dismiss Clean Air Act citizen suits, allowing a pollution case against Musk's AI data center to proceed.
pgvector vs. Dedicated Vector DBs: Laptop Benchmark Shows Similar Latency
A new benchmark suggests that on typical laptop hardware, pgvector and dedicated vector databases deliver comparable latency, but performance is highly dependent on system resources.
OpenAI blueprint aims to empower cyber defenders with frontier AI
A new policy paper from OpenAI argues that frontier AI can shift the advantage toward cyber defenders if paired with safeguards and stronger public institutions.
OpenAI models land on Google Cloud via Model Garden
OpenAI's frontier models and Codex workflows are now available on all three major clouds, expanding enterprise options beyond Microsoft Azure.
Build a Lightweight RAG Pipeline with Scikit-learn and Open-Source LLMs
A new tutorial demonstrates assembling a retrieval-augmented generation system using scikit-learn for vector search and open-source models for embeddings and generation.
Python Descriptors for Controlled Attributes in ML
Python descriptors let developers enforce validation and computed behavior on class attributes, a useful pattern for machine learning pipelines.