GPT-6 Astra performs best in national lab benchmark
A new benchmark shows AI agents producing plausible but false research papers, with most submissions failing integrity checks for fabricated experiments.
Read the story ↗The Binary Byline reading room
Agent systems
11 stories / Artificial intelligence
100-task OpenAI benchmark assesses AI agents' ability to handle routine wet-lab science workflows
A new 100-task benchmark from OpenAI evaluates how AI agents interpret raw experimental data and handle routine laboratory science workflows.
Read the story ↗
Axiom Math formally verifies Stacks Project using AI agents
The startup converted roughly 7,654 pages of mathematics into computer-checked code, aiming to prevent unchecked proofs from polluting the field.
Read the story ↗
Uber executives cite native app speed as hurdle for AI agents amid 500 food orders
Executives cite rapid native app speeds as a hurdle for personal AI agents during the busiest period.
Read the story ↗
Oracle Lakehouse adds semantic layer for real-time AI agent data access
Oracle’s new semantic layer lets AI agents query multiple data sources in place, prioritising real-time access over copied snapshots.
Read the story ↗
Oumi’s AI Engineer now retrains model weights for custom agents
Oumi extends its agent builder to retrain underlying models using production failures, moving beyond prompt engineering limits.
Read the story ↗
Google Antigravity agent proposes rigidified macrocycles in an hour
DeepMind adds cheminformatics skills to its agent, enabling compound search and synthetic route generation for drug discovery workflows.
Read the story ↗
YC chief Garry Tan spends up to $200,000 yearly on AI agent tokens
Y Combinator CEO Garry Tan disclosed his annual spending on AI agents and shared how he uses them for coding and personal assistance.
Read the story ↗
Musk cites unsustainable Grok Bot costs amid subscriber limit complaints
Elon Musk stated that the operational expense of running the Grok Bot agent exceeds its revenue generation, following user reports of rapid quota exhaustion.
Read the story ↗
Meta’s Muse agent handles bills and reservations, raising cognitive offloading concerns
Meta launches a personal AI agent for administrative tasks, while psychologists warn that outsourcing mental effort may reduce cognitive reserve over time.
Read the story ↗
MPA deploys no-code AI agent for Trusted Partner Network alert triage
Legal operations at the Motion Picture Association use a sales CRM and an AI agent to manage global anti-piracy investigations and security alerts.
Read the story ↗