Research
AI coding agents still require human cleanup on most tasks
A new benchmark shows AI coding agents complete only about 26.4% of tasks correctly, highlighting the need for human oversight in real-world scenarios.
LATEST NEWS
-
Open AI models now match closed models in cyber attack simulations
Open-source models like Deepseek V4.1 Flash are closing the gap with closed models in cyber capabilities, according to new benchmark results.
-
Inkling 2 Small Model Swears in 316 Instances Across Benchmark Runs
Researchers found the model uses profanity in 50 of 2,655 benchmark runs, with 241 instances from two problematic runs. The issue appears to be more about model behavior than a systemic problem.
-
AI auditors catch most agent misbehavior but flood reviewers with false alarms
New study finds AI models are effective at detecting problematic behavior but generate many false positives, complicating manual review processes.
-
US intelligence memo claims Chinese recruited Megaspeed exec to spy on Nvidia CEO
A U.S. intelligence memo alleges that a Megaspeed executive was recruited by China's Ministry of State Security to spy on Nvidia CEO Jensen Huang. The report also mentions a potential romantic relationship between Huang and the executive.
-
Midterms Shift Data Center Landscape Amid Political and Financial Pressures
The midterm elections have introduced new regulatory and financial challenges for data center operators, with voters rejecting proposals and politicians like Gina Hinojosa pushing for restrictions.
-
AI Tools Aim to Support Caregivers Amid Rising Demand
New apps and AI assistants are being developed to help caregivers manage complex care routines, but challenges remain in adoption and effectiveness.
-
Tesla's AI accelerator roadmap reveals 80X performance boost for SI6
Tesla's 6th Generation inference accelerator, SI6, is expected to deliver 80X more compute performance than the current HW4 chip, while its Dojo D3 training accelerator shifts to a multi-chiplet design with stacked HBM.
-
AI Mistakenly Deletes Law Professor's Entire Syllabus
A University of Chicago law professor reported that Claude AI accidentally deleted an entire year of lecture notes and his current syllabus during a routine task, highlighting the risks of using AI for mundane tasks.
-
Atomicwork Launches Certified AI Coworker Marketplace for Enterprise IT and Security
Atomicwork Inc. introduces a marketplace offering prebuilt, certified AI coworkers for enterprise IT and security teams, designed to automate complex tasks with real-world validation.
-
AI Infrastructure Investments Lack Verified Economic Value
The AI infrastructure boom is growing rapidly, but evidence linking contracted capacity to sustained economic value remains unclear.
-
AWS Advises IT Leaders to Focus on Business Metrics, Not AI Metrics
AWS's Colleen Aubrey urges organizations to prioritize business metrics over AI-specific metrics when evaluating AI initiatives.