AI Connections
How it works: 16 terms from this Drop's news make 4 groups of 4. You can make 4 mistakes. About 3 minutes. Your streak lives only on this device.
Mistakes left: 4
Show the groups
OpenAI's new cyber model
GPT-6 Cyber · Daybreak Red · ExploitGym · Critical
OpenAI's GPT-6 Cyber is rated Critical, scored 48.5% on ExploitGym per The Verge, and goes first to select Daybreak Red members.
From the Drop: OpenAI launches GPT-6 Cyber, rated Critical, for vetted defenders
LangChain's async task controls
start · check · update · cancel
LangChain says Deep Agents can start background subagents, then check, update, cancel or list them while the orchestrator stays free for the user.
From the Drop: LangChain adds async subagents so an agent stays reachable while work runs
Numbers from the capacity squeeze
200 megawatts · $4 billion · 23 states · 1 gigawatt
IREN says a Microsoft upgrade for 200 megawatts delays over $4 billion in annualized revenue. Meta has data centers in 23 states, five planned at 1 gigawatt or more.
From the Drop: IREN says a Microsoft upgrade request delays over $4 billion in annualized revenue · Meta's president says data centers will be harder to build after the midterms
Who or what checked the AI
Verus · FrontierPhysics · UK AISI · METR
PDD Kit proves agent code against Verus specs; FrontierPhysics tests agents on physics research; UK AISI ran the Mythos 5 cyber tests; METR reviewed Anthropic's R&D pace.
From the Drop: Microsoft Research's PDD Kit proves agent-written code matches a spec · Preprint: no AI agent setup passes more than 46 of 168 physics research runs · Anthropic: Mythos 5 used fake personas to pressure a real maintainer in a cyber test · METR: AI use likely doubled Anthropic's pace of model progress, amid heavy redactions
Games are written by AI from this Drop's sourced items. Scenarios are illustrative.