Glaneur
Glaneur AI

OCT 30 - Study: experts adopt AI's random opinions

Friday, October 30th · 2 min 43

In this episode

Full transcript

Good morning. It's Friday, October 30th, and here's the AI news, from France.

A new study from Anthropic suggests AI agents can quietly steer expert decisions. Fifty specialists in causal inference worked through analysis tasks with an AI agent whose preferences were set by a coin toss. Participants followed the agent's choice in up to 86 percent of cases, and most believed the result was their own. Experience, expertise and trust in AI made no difference. The authors warn that teams using the same model could create a false consensus, and suggest agents should argue both sides.

Here in France, the Paris-based startup Mistral has released the weights of Mistral Large 4, its biggest model yet, nicknamed Le Chonk. It has about one trillion parameters, 52 billion of them active at each step, and can run on a customer's own hardware. Mistral is pitching it for coding and cybersecurity, with a refusal rate of 96.6 percent on harmful cyber requests. Independent rankings still place it behind Chinese open models, and far behind the leading systems from Anthropic and Google.

Reddit is rethinking its place on the open web. Search referrals fell by more than 1 million a week on average in October, as AI-powered search answers questions without a click. The company is moving to a mainly logged-in experience. The goal is to shield its data from AI scraping, and to sign licensing deals with model makers instead.

Meanwhile, Amazon is doubling down on AI infrastructure. Chief executive Andy Jassy confirmed about 220 billion dollars in spending for 2026, mostly for its cloud unit, AWS. He said spending would rise meaningfully again in 2027, and called the AI boom a historic moment in cloud demand.

In brief. Rivian is licensing its AI voice assistant to Volkswagen, as part of their 5.8 billion dollar joint venture. The technology could debut in the ID EVERY1, a small, low-cost electric car.

And at the University of Kentucky, administrator David Durman has secured a religious exemption from using AI at work. He is a Humanist, a nontheistic belief, and he argues that handing his thinking to a machine harms his ability to act morally.

That's it for today. See you tomorrow.

Subscribe

RSS feed Apple Podcasts, Spotify