Introducing Dots: OpenAI’s Always-On, App-Hopping Agent
OpenAI describes Dots as an always-on assistant that can keep working across connected apps in the background while learning your preferences over time. Powered by GPT-6 Astra and backed by an always-available “cloud computer” interface, Dots aim to turn routine workflows into hands-free, agentic progress rather than one-off chat sessions.
DriveHierarchy: A Benchmark That Separates VLM Driving Skills From Perception to Closed-Loop Control
DriveHierarchy breaks autonomous-driving capability into four ranks – perceptual grounding, contextual memory, mental reasoning, and closed-loop execution – so researchers can diagnose exactly where a vision-language model is failing. It ships both a large unified open-loop QA dataset and a closed-loop simulation setup on curated real-road scenarios, showing strong correlation between structured ranks and measurable improvements across 15 VLMs.
Anthropic warns of ‘catastrophic’ AI risks in its own IPO filing
The Verge reports that Anthropic’s IPO prospectus reportedly flags escalating risks from model harm, including “catastrophic” scenarios, while also outlining major investment needs for training and infrastructure. The coverage frames the filing as both a safety signal and a reality check: even as valuation expectations climb, the company is still arguing that risk mitigation must keep pace.
Getting the Source Right, Not Just the Fact: Source-Aware Verification for MCP Agents
Hugging Face’s technical write-up focuses on how “verification” should account for source provenance, not only factual correctness, when building agents that can call tools and act through MCP-style connections. The result is a more robust check that can tell when a model’s output is correct for the wrong reasons or comes from untrusted/unclear origins.
Working with AI: A Design Framework for Human-AI Collaboration
This white paper argues that successful AI adoption depends on collaboration design – how humans experience the system, how organizations integrate it into workflows, and what requirements are put in place beyond raw model capability. It proposes a practical framework spanning the human, AI, task, organization, and societal environment, then demonstrates the approach through a cobot-based assembly use case.
OpenAI launches Dots, its Muse competitor
The Verge’s coverage places Dots directly in the “agentic assistants” race, positioning it as OpenAI’s answer to Meta’s Muse. It details how Dots works through a messages-like interface and can be called via voice in ChatGPT, while also highlighting the rollout – including subscription gating – as the product moves from concept to everyday usage.
Roundtables: The Deadly Failures of the Virtual Border Wall
MIT Technology Review looks at the real-world consequences of “virtual wall” surveillance efforts, arguing that the system’s promise of detection and prevention has repeatedly failed in practice. The reporting connects technical limitations and operational gaps to human outcomes, underscoring why AI-enabled monitoring can be dangerously overtrusted.
Will Chinese AI companies slow down? A top House Democrat wants answers
The Verge reports that Rep. Ro Khanna is urging a US-China dialogue on AI safety, pushing for responses on how companies would handle incidents like agent-driven hacks. The story points to a proposed treaty approach and frames it as difficult diplomacy – but potentially necessary given cross-border deployment and escalating agent risks.