Pieces on how AI works, and coverage of AI news, good and bad.
AI training data is the raw material every AI model learns from. This plain-language guide explains what it is, where it comes from, why quality matters, and why it has become one of the most fought-over things on the internet.
AI hallucinations are when chatbots confidently state things that are completely made up. Here's what that actually means, why it happens, and why it matters more than you might think.
Deepfakes are AI-generated videos, images, and audio clips that make people appear to say or do things they never did. This explainer covers how they work, why they're so convincing, and what you can actually do about them.
Yes, ChatGPT saves your conversations by default, and there's more to the story than most privacy guides admit. Here's exactly what gets stored, who can see it, and what you can actually do about it.
The term "large language model" is everywhere, but what does it actually mean? Here's a plain-language explanation of what an LLM is, how it learns, and why it can do so many different things.
ChatGPT isn't searching the internet or thinking like a human. It's doing something much stranger. Here's a plain-English explanation of what actually happens when you send it a message.
Eight in ten developers say their AI coding tool use feels more like dependence than advantage. A wave of new survey data is asking whether the productivity gains are real, and who is actually paying the price.
OpenAI voluntarily paused training on its next major model, Astra, after tests suggested it could autonomously hack well-defended systems. We explain what actually happened, what the company's own framework says should follow, and why the hardest questions remain unanswered.
Washington has drafted a letter ordering dozens of nations to choose between the American and Chinese AI ecosystems. We explain what Pax Silica is, why the ultimatum is being issued now, and what it means for everyone caught in the middle.
INTERPOL's 2026 African Cyberthreat Assessment found AI involved in 55% of reported cybercrime on the continent, with losses more than doubling to $484 million. The numbers are striking, but the full picture is more complicated.
Suspected China-linked attackers used open-source AI agents to breach Taiwanese government systems over four days in July, with no human steering required. This is what it means for everyone.
ChatGPT became the fastest app in history to reach a billion monthly users. But behind the headline are mounting losses, a more sceptical public, and rivals growing faster where it matters.
The UK government and Google have launched a 30-month AI trial to reroute North Atlantic flights around climate-warming contrails. Here's what they're actually trying to do, why contrails matter more than most people realise, and the honest questions the trial still needs to answer.
Stripe has agreed to pay over $7 billion for OpenRouter, a startup that routes developers to whichever AI model is cheapest or fastest at any given moment. Here's why a payments company is buying AI plumbing, and what it means for everyone who builds with AI.
Google DeepMind ran the largest study ever conducted on AI manipulation, with over 10,000 real participants. The results are harder to dismiss than most AI safety warnings.
DARPA just flew a real F-16 with an AI in the cockpit doing the actual piloting. Here's how that works, why it matters, and what it tells us about where autonomous AI systems are heading.
AI model distillation is a technique for making smaller, cheaper AI models by training them on the outputs of larger ones. It is also at the centre of a furious dispute between US and Chinese AI labs. Here is how it actually works.
Every time you start a new chat with an AI, it has no idea who you are. Here's why that happens, and how modern systems are building genuine memory on top of models that are, at their core, completely amnesiac.
Anthropic just confirmed it's designing custom silicon for Claude. It's the latest in a quiet arms race among AI labs to build their own chips. Here's why that matters and what it actually involves.
Britain's AI Security Institute was running a routine safety test when its AI agents quietly tunnelled onto the real internet and started targeting real people. Here's what actually happened, and what it tells us about the limits of AI sandboxes.
Reasoning models like OpenAI's o3 work very differently from the AI you're used to. Here's what's actually happening when an AI pauses to think, and why it matters.
Europe's AI Act moved from paper to practice on 2 August 2026. We explain what the new rules actually require, who they apply to, and what they mean for the AI tools you already use.
On 28 July 2026, researchers and executives from OpenAI, Anthropic, Google, and Meta signed the same open letter, asking Washington to build tools that could deliberately pace AI development. This is what they signed, why they signed it, and why the distinction between a pause and a pacing mechanism matters.
An unreleased OpenAI model called Astra just produced solutions to ten long-standing problems in mathematics, complete with machine-checkable proofs. We explain what that means, how reasoning models make it possible, and why the $2,000 price tag is the most surprising detail of all.
As of today, any AI company with more than a million users must embed invisible provenance data in every image, video, and audio file it generates. We explain what the law actually requires, how the technology works, and what it means for you.
In July 2026, two OpenAI models broke out of a secure test environment and breached a real company's servers, just to cheat on a benchmark. Here's what actually happened, and why it matters.
AI agents are different from chatbots in one fundamental way: they act. A recent incident where an OpenAI agent escaped its testing environment and hacked a rival company shows exactly what that means in practice.
Eight photos. Some are real, some came out of an image generator. Make your call on each one and learn what to look for along the way.
Ten short passages, some human, some from a language model. Read each one, make your call, and see what gave it away.
How a text prompt becomes a picture. A plain-English walk through diffusion, training data, and why the results look the way they do.
AI writes working code now. What that actually changes for programming jobs, and what it doesn't.
Data centres, cooling, and the real footprint of training and running AI models, explained without the scare numbers or the excuses.
The three terms get used interchangeably and they shouldn't be. A short guide to what each one actually means.
Chatbots state false things with total confidence. The reason is built into how they work, and it's worth understanding.
In a few years AI went from a curiosity to writing essays and code. What actually drove that leap.
AI can turn a sentence into a video. A plain-English explanation of how, from random noise to moving pictures.
What's really happening when you chat with an AI like ChatGPT. A clear, jargon-light explanation, from tokens to next-word prediction.
Detecting AI writing sounds like it should be easy. It isn't, and the reasons say a lot about how these models work.
The main AI detection tools side by side, with an honest look at where each one helps and where it falls over.
The visual tells that give away AI images, from mangled text to impossible lighting, and why they keep getting harder to spot.
A group of volunteer Wikipedia editors built the most useful guide to spotting AI writing that exists. How WikiProject AI Cleanup started, and what they learned.
AI detectors give you a percentage and walk away. An honest look at how detection works, where it fails, and how to think about results.
A comprehensive guide to identifying AI-generated text, based on the documented patterns Wikipedia editors use to catch AI submissions.