A new AI grader may change the scores
Zhang and Yang tested three changes to an AI scoring tool using the same research proposals. After one change, 93% of scores had one value. The test did not use student work.
AI AgencyRead story
Lead story
News is being packaged for AI readers
WIRED reports that Axios has launched a deal-content feed for subscribers using AI tools. Every offers ways to ask questions of its reporting. Later Axios feeds are still plans, not launched products.
Source: WIRED · Oct 27, 2026Read original source
AI Agency
Read story
More top stories
Let teachers set the limits for AI help
Haq offers a design in which teachers approve what AI may do for each learning activity. The paper uses worked examples. It does not show better K-12 learning outcomes.
AI AgencyRead story
Featured
A big AI math run still needs proof
Import AI relays a researcher’s account of AI-assisted math work: about $135,000 in estimated token costs and roughly 600,000 lines of Lean code. Neither figure was independently checked here.
AI AgencyRead story
Agents & Automation
Make repeat work repeatable
Sabouri and colleagues made repeat steps in AI-assisted analytics runs reusable. Their set tests found more consistent tool calls. They did not test the method in a live office.
Beyond PromptingRead story
Agents & Automation
Ask what finished, not how busy the AI looks
Mittal and colleagues map problems people face when they delegate work to AI tools. A tool may keep running without moving the task forward. Their research does not give a workplace failure rate.
Beyond PromptingRead story
Beyond Prompting
More from this edition
-
Research & ScienceRead story
A visible reason is a place to check, not proof
Anbalagan and colleagues tested a display that ranks evidence for an AI decision. In a 50-person early evaluation, 92% found the evidence helpful for understanding and trust. That does not prove the decision was accurate.