AI-Writing Detection Is a Moving Target: What a 23-Model Study Found
A new 23-model study shows why AI-writing detector accuracy can collapse after a model release—and how institutions should respond fairly.
Tag: artificial intelligence
AI Across 15 TV Boxes: What Anthropic’s Self-Replication Study Actually Showed
Anthropic’s contained IoT experiment shows why AI self-replication deserves attention—and why it is not evidence of a wild autonomous AI worm.
OpenAI Released 719 AI Maths Manuscripts: What Counts as a Verified Result?
OpenAI’s mathematics release separates formal proof checking from human understanding, scholarly significance and community verification.
Can an AI Model Really Forget? Machine Unlearning Is Not a Delete Button
Machine unlearning promises to remove unwanted influence from AI models, but recent research shows why refusal, deletion and verified forgetting are not the same.
AI Is Making Science Faster—But Verification Is Becoming the Bottleneck
New evidence suggests AI can save researchers time and generate more hypotheses, while shifting the hard work downstream into validation, experiments and judgement.
GPT-6.1 Sol: How to Compare AI Models by Cost per Successful Task
OpenAI’s GPT-6.1 Sol puts near-frontier capability at a lower published price. This guide shows how teams can test total cost, task success and deployment risk before switching models.
AI Labels Should Explain the Human Contribution
A binary AI-generated label can hide the person who supplied the idea, judgement and final responsibility. Use a contribution record that shows how human and machine work combined.
A Scenario Is Not a Prediction: How to Use ‘What If?’ Thinking Without Fooling Yourself
A practical guide to using scenarios and thought experiments for better decisions—without turning a plausible future into a false prediction.
Let Students Explain the Edit: A Classroom Method for Learning With AI
Make AI-assisted learning visible with an original draft, a controlled edit and a student explanation of what changed and why.
An AI System Card Is Evidence, Not a Safety Guarantee: How to Read One Properly
Learn how to read an AI system card critically: identify the tested model, evaluation scope, deployment context, limitations, gaps and monitoring needs.