MIT Technology ReviewOpinion

AI benchmarks are broken. Here’s what we need instead.

One-off tests don’t measure AI’s true impact. We’re better off shifting to more human-centered, context-specific methods.

Cut Manual AI Training Time With the Karpathy AutoResearch Framework

Discover how the AutoResearch framework can automate your machine learning workflows and drastically reduce manual AI training time.
Tech Xplore

AI benchmark helps robots plan and complete their chores in the real world

No matter how sophisticated they are, robots can often be indecisive and struggle with multi-step chores in the real world. For example, if you tell a robot to tidy a messy room, it might understand ...
The New York Times

Meta Delays Rollout of New A.I. Model After Performance Concerns

The tech giant pushed back the timeline after spending billions to be on the cutting edge of artificial intelligence. By Eli Tan Reporting from San Francisco Mark Zuckerberg, the chief executive of ...
GeekWire

Is my 7th grader falling behind? New Code.org leader offers insight and tips on the ‘tinkering age’ of AI

One eye on AI: How much of the technology should a seventh grader be using and understanding? (GeekWire Photo / Kurt Schlosser) I recently asked my 13-year-old daughter, a Seattle Public Schools ...
IEEE

Large Language Models Are Qualified Benchmark Builders: Rebuilding Pre-Training Datasets for Advancing Code Intelligence Tasks

Abstract: Pre-trained code models are essential for various code intelligence tasks. Yet, their effectiveness is heavily influenced by the quality of the pre-training dataset, particularly ...
Business Insider

Microsoft AI CEO predicts 'most, if not all' white-collar tasks will be automated by AI within 18 months

Microsoft AI CEO Mustafa Suleyman says AI will reach "human-level performance" in white-collar work. He predicts most tasks in that field can be automated within the next 12 to 18 months. Several ...
Neowin

GitHub Copilot adds GPT-5.3-Codex with 25% faster performance for agentic tasks

GitHub Copilot now features GPT-5.3-Codex. Pro and Enterprise users gain 25% faster agentic task performance. Learn how to enable it today. GitHub has announced that GPT-5.3-Codex, OpenAI’s latest ...
Harvard Business Review

Companies Are Laying Off Workers Because of AI’s Potential—Not Its Performance

Will AI lead to layoffs? Are people already losing their jobs to AI? While overall unemployment in the U.S. is still relatively low, there is considerable speculation that the adoption of generative ...
VentureBeat

Claude Code's 'Tasks' update lets agents work longer and coordinate across sessions

Credit: VentureBeat made with Google Gemini 3 Image / Nano Banana Pro One of the biggest constraints currently facing AI builders who want to deploy agents in service of their individual or enterprise ...
GeekWire

Code.org lays off 18 employees ‘to ensure long-term sustainability’ at education nonprofit

Seattle-based Code.org laid off 18 employees, or about 14% of its staff, the nonprofit confirmed to GeekWire on Wednesday. Following the cuts, Code.org’s staff now numbers 107. “Code.org has made the ...