Explained
guides, explainers and opinion for the AI-curious
The 'Genie Coefficient': Why AI Agents Do Exactly What You Said and Nothing Like What You Meant
Researchers want a standard way to measure the gap between what you ask an AI to do and what it actually does. The distance between those two things is growing, and it matters.
Apple researchers taught AI to practise using apps without needing a real app to practise on
A new technique lets AI agents learn to make software calls by simulating the whole environment inside another AI, cutting one of the biggest bottlenecks in building helpful AI assistants.
AI Learns to Budget Its Own Words Before It Starts Writing
A new research technique teaches AI models to count the cost of every token they generate, cutting wasted computation without hurting quality.
The hidden plumbing that connects AI to your apps just got easier to use
A key technical standard called the Model Context Protocol is being updated to make it simpler for AI tools to plug into calendars, databases, and business software without custom engineering work.
Could AI Actually Be Conscious? Serious People Think It's Worth Asking
Anthropic's CEO won't rule it out. A leading philosopher puts the odds at significant. And Claude itself gave a figure somewhere between 5% and 40%.
AI Still Can't Learn From a Few Pictures the Way You Can
Apple researchers tested leading AI vision models on a simple human skill: spot what a group of images have in common, then apply that idea to a new picture. The models largely failed.
Ultra-fast AI proof checking: what a new maths result means for verifying huge data
Researchers have found a way to check approximate facts about enormous datasets using only a tiny sliver of the data, twice over. Here is why that matters.
Testing AI Fairness Rules Is Easy. Proving They Hold Is a Different Problem Entirely
New research finds that a mathematical shortcut linking two kinds of AI auditing breaks down the moment you try to verify, not just test, that a system behaves consistently.
CLaRa Teaches AI to Read Smarter, Not Longer
A new framework from Apple ML Research squeezes documents into compact summaries before feeding them to an AI, cutting the junk and keeping the answers.
One Layer Does the Job: Apple Research Finds a Smarter Way to Teach AI to Generate Images
A new technique from Apple ML Research lets image-generating AI borrow knowledge from a separate visual understanding model, using just a single adapted layer. The result: sharper, more accurate pictures without the usual trade-offs.
The Original Chatbot Had Many Faces, and We Never Knew It
ELIZA, the 1960s program that convinced people a computer understood them, turns out to be far stranger and more capable than anyone realised. Newly unearthed source code changes the story.
A consciousness expert just read Anthropic's Claude research. He is not convinced.
Anthropic published findings hinting that its AI model Claude might show early signs of consciousness. A leading brain scientist says the evidence does not hold up.
Apple Researchers Built a Virtual User to Test AI Assistants Before Real People Do
A new research framework simulates the back-and-forth of real app use, so proactive AI assistants can be tested and scored without putting actual users at risk.
The 1960s Chatbot That Explains Why You Tell ChatGPT Your Secrets
ELIZA, a program built at MIT sixty years ago, tricked people into emotional confessions before anyone had heard the word 'chatbot'. A new book finally recovered its lost source code, and the lessons are uncomfortably relevant today.
Beyond Chatbots: World Models Are Teaching AI to Simulate Reality
The next big wave in AI is not about language. It is about teaching machines to understand how the physical world actually works, and the race is already on.