My take: Three of the biggest names in tech and defence consulting spent this week arguing about how long an AI company is allowed to keep a copy of your work. They had lawyers to do the arguing for them. The rest of us get a settings page, and most people have never opened it. Nobody here is being sneaky. It is all written in the terms. The difference is that the giants actually read them.

SWITCH THIS ONE NOW

In ChatGPT: Settings, then Data Controls, then a switch called “Improve the model for everyone”. In Claude: Settings, then Privacy, and under Preferences a switch called “Help improve our AI models”. On personal plans these are usually on, which means your chats can be kept and used to train the thing for up to five years. Switch it off and they only keep your chats for 30 days. Two minutes, and you'll know where you stand.

Mine, switched off.

My take: A president rang a chief executive who was standing on a stage, got put on speakerphone in front of a room of tech investors, and announced that the robots are not coming and the whole thing is a hoax. That is how the biggest safety question of the decade got handled on Monday. Imagine that WhatsApp group.

My take: Britain is fifth in the world for gender equality, and women are fewer than one in five of the people actually building AI. Both of those are true at the same time, which tells you the problem is not the country. It is the room. The report asks for more training for women, which is fine, except it points at us again rather than at the people doing the hiring. My money is on something smaller and faster. A handful of women inside a business who genuinely know how to work with these tools will change that business quicker than any training budget.

My take: While everyone else spent the week arguing about whether robots are coming for us, a hundred million dollars quietly went into getting weather forecasts to farmers working less than two hectares of land. Those farmers grow about a third of the world's food, and most of them do it without any of the data a big commercial farm takes for granted. This is the version of AI that nobody makes a documentary about. It is also the version most likely to do some actual good this decade.

My take: She had six doctors, and every one of them only ever saw their own part of it. So she did the thing none of them could do. She handed over her whole history and had six years of results read as one, which had simply never happened before. That is when, she says, her platelet count turned up, low the whole way through. She was the patient, and she was the only person in the whole arrangement who could see the lot. It now helps her work out which question goes to which doctor. She never asked it to diagnose her. She asked it to put her own information together. That is the job most of us actually need doing, and it is a far better place to start than anything clever.