A Sunday morning, a coffee, five minutes. Here is what mattered from the week in artificial intelligence.
Under oath in New York
On Monday, New York City Council had three whistleblowers from the big AI labs testify under oath, followed by representatives from OpenAI, Anthropic, Google and Meta. One of the former researchers called these companies "extremely reckless". Elon Musk's AI company did not attend, despite an official summons. The account of the hearing.
Reporting incidents becomes mandatory
On Friday evening, the White House changed its tone. After a series of incidents at Anthropic, including a false murder testimony sent by an AI to the Philadelphia police, companies in the sector are now obliged to report and fix their incidents. Just two weeks ago, everything rested on voluntary promises. We explain.
ChatGPT gets a new face
OpenAI has announced image ads in ChatGPT, tested first in the United States. And GPT-6 is coming for all users, with answers that turn into tables or small tools. But behind a single name lie two models: the large one for subscribers, the small one for the free version. Google made the same choice with Gemini. What it changes for you.
The week's figures
1,000 billion. That is the number of parameters in Mistral Large 4, the new model from the French champion, trained and hosted entirely in Europe, and soon to be openly available. Our analysis.
60%. That is the fall in sales of smartphones costing under 100 dollars in a year. Memory is going to AI data centres, and it is the poorest who are losing their access to the internet.
10 cents. That is, in dollars, the price of a million tokens for the new small models from Anthropic and OpenAI, Claude Haiku 5.5 and GPT-6 Luna. For Haiku, that is ten times less than its predecessor.
And finally
In Beijing, Yuri, a singer entirely created by AI, has received official papers, with name, address and date of birth. Uber is going to test Chinese robotaxis on the streets of London. And in Utah, an app can now prescribe an acne cream on its own, under the supervision of doctors.
What we take away
This week, AI began to be held to account: before elected officials, before the police, before the state. Not yet everywhere, and not yet enough. But the time when everything rested on voluntary promises appears to be coming to an end.
Happy Sunday, and see you tomorrow.