Skip to main content

AI & ML

Latest coverage
FrontierSWE Gave Coding Agents 20 Hours. Its Harness Changed How Much of the Clock They Used.
AI & ML WORLD
FrontierSWE Gave Coding Agents 20 Hours. Its Harness Changed How Much of the Clock They Used.
Across 34 hard tasks, the model gap was large; across a smaller six-task swap, time controls and checkpointing lifted...
OpenAI Now Has Three Misalignment Tracks. The Deadlines Are Still Private.
AI & ML WORLD
OpenAI Now Has Three Misalignment Tracks. The Deadlines Are Still Private.
The new framework clarifies lifecycle coverage and third-party notice, but it leaves the response clocks and objectiv...
Cohere Counted 696,291 MCP Tools. Its 2.6% Result Is Not a Runtime Test
AI & ML WORLD
Cohere Counted 696,291 MCP Tools. Its 2.6% Result Is Not a Runtime Test
The dataset is a useful audit of public agent-tool supply, but its own files expose the gap between an integration de...
Korea's AI-Exposed Industries Lost 268,000 Young Workers While Older Employment Rose
AI & ML WORLD
Korea's AI-Exposed Industries Lost 268,000 Young Workers While Older Employment Rose
Bank of Korea evidence maps a missing career ladder: youth employment fell in AI-exposed industries, employment among...
92% of Brazilian Capital-Market Institutions Use AI. Only 3% Report a Clear Advantage.
AI & ML WORLD
92% of Brazilian Capital-Market Institutions Use AI. Only 3% Report a Clear Advantage.
Near-universal reported use has rarely produced strategy-linked measurement, while BSM's narrower surveillance deploy...
AI Cuts Headcount When Firms Rebuild Functions, Not When Workers Use Chatbots
AI & ML WORLD
AI Cuts Headcount When Firms Rebuild Functions, Not When Workers Use Chatbots
New Census Bureau microdata splits business AI use into three layers, and only functional deployment and capital spen...
Daycoval's Twelvefold AI Sales Claim Does Not Measure Credit Quality
AI & ML WORLD
Daycoval's Twelvefold AI Sales Claim Does Not Measure Credit Quality
Daycoval says an agent can score, negotiate, and close payroll loans through WhatsApp and Pix, but sales growth answe...
The EU Pushed Its AI Credit and Insurance Deadline to December 2027. The Live Obligations Stayed Put.
AI & ML WORLD
The EU Pushed Its AI Credit and Insurance Deadline to December 2027. The Live Obligations Stayed Put.
The Digital Omnibus bought creditworthiness and insurance-pricing models sixteen more months, but the rules already b...
China Wins the AI Workload, the US Keeps the Wallet
AI & ML WORLD
China Wins the AI Workload, the US Keeps the Wallet
Chinese open-weight models now route more tokens than American ones, yet earn a rounding error of the revenue. The ga...
The EU Answered a Severe AI Warning With the Rulebook It Already Had
AI & ML WORLD
The EU Answered a Severe AI Warning With the Rulebook It Already Had
Three weeks after the ESRB called frontier-AI cyber risk "severe," the EU's three financial supervisors answered with...
The EU AI Act's First Binding Deadline Lands Sunday. It Is Not The One Brussels Deferred.
AI & ML WORLD
The EU AI Act's First Binding Deadline Lands Sunday. It Is Not The One Brussels Deferred.
Article 50's transparency rules bind every chatbot, deepfake and generative-AI provider selling into the EU from Augu...
Claude Opus 5 Costs Half What Anthropic Charges for Fable 5. Anthropic Ships Both.
AI & ML WORLD
Claude Opus 5 Costs Half What Anthropic Charges for Fable 5. Anthropic Ships Both.
Anthropic launched near-frontier Claude Opus 5 at 5 and 25 dollars per million tokens, half the price of the flagship...