A quiet Sunday news day with one story that deserves your full attention: a new report says Anthropic's models are now improving about twice as fast as a year ago. Meanwhile the market is selling off AI stocks and some bosses are discovering that ordering staff to use AI is not the same as getting value from it.
📊 By the numbers: AI mood index -19.4 · most-mentioned this week: OpenAI · 127,123 articles tracked — AI Pulse dashboard
1. Anthropic's models are improving about twice as fast as a year ago New
The Verge reports on a new report, from the evaluation group METR going by the article's own link, finding that Anthropic is using AI to help build its next AI and that its models are now improving roughly twice as quickly as a year ago, with little sign of slowing. That is why researchers are alarmed, and it backs up Anthropic's own warning about AI that improves itself. For anyone using these tools, the practical upshot is simple: whatever Claude could not do for you six months ago is worth testing again now, and any workflow you built on today's limits has a short shelf life.
2. AI stocks slide as the midterm backlash reaches investors ▲ Day 4
The new piece this week is the money. Seeking Alpha says chip stocks have sunk deeper into a bear market and dragged the Nasdaq down with them, and The Times argues the midterms were the biggest needle yet aimed at the AI bubble, while TechCrunch says the results look rough for Elon Musk's data centre plans. A falling stock price does not change the tools on your desk tomorrow, but cheaper capital is what has kept AI subscriptions underpriced, so expect price and usage-limit changes to keep coming.
3. Nvidia's China chip sales look poorly policed New
The Verge, citing The New York Times, reports that one of the Chinese firms cleared by the US to handle Nvidia's H200 chips is little more than an empty office. TechCrunch separately reports on a US intelligence memo claiming a Chinese tech executive was recruited to spy on Nvidia's CEO. Neither touches your software directly, but chip export rules are now a political football, and that is one more pressure on the supply and price of the hardware every AI service runs on.
4. Bosses who forced staff onto AI are now drowning in workslop New
Futurism reports that executives who pushed employees to use AI are now buried in what people are calling workslop, polished-looking output that says nothing, and are changing their tune. That is the predictable result of a mandate with no training behind it. AI is not the problem here; telling people to use it without showing them where it helps and where it does not is.
5. Fireworks says paying for one AI model for everything wastes money New
Fireworks AI, a company that runs models for other businesses, told The Register that companies are wasting money unless they use a mix of models. That is a self-interested pitch from a firm that sells access to many models, but the core point holds. Sending simple jobs like sorting email or pulling data from forms to a small, cheap model and saving the top model for hard work is one of the easiest ways to cut an AI bill.
Models that improve twice as fast, stocks that fall, and staff told to use AI with no instructions. Only one of those is fixable by Monday.
How this brief is built: an autonomous pipeline collects 350+ AI headlines a day into an 80,000-article archive, clusters the last 26 hours into candidate stories, and Claude picks the five that matter and writes this in my voice. It publishes itself every morning — no human touches it before you do.