AI Trends Weekly Issue 294: Agents must do more than look at images; Google's new benchmark directly tests real-world professional work
AI趨勢周報第294期:AI Agent不只要會看圖,Google新Benchmark直接考專業工作實戰
AI summary
Google DeepMind and Artificial Analysis launched the MKA benchmark, using 200 tasks to test agents on visual knowledge work. Claude Fable 5.2 ranked first with a 57% Climb WSR, and Gemini 4 Carbon ranked second at 54%, but at about one-sixth the average cost. The Cybersecurity Administration said overseas hackers have been using agent tools for cyberattacks since July this year, affecting more than one million personal data records in Taiwan.