Sunday 21 June 2026

The Brief – 21/06/2026

The top things worth knowing about in AI today.

  1. Nobel laureate leaves DeepMind for Anthropic

    John Jumper, who shared the 2024 Nobel Prize in Chemistry for the protein-folding system AlphaFold, is leaving Google DeepMind for Anthropic after nearly nine years. His exit follows the move of Gemini co-lead Noam Shazeer to OpenAI and the departure of AlphaGo researcher David Silver to start his own company. The losses land just as Google prepares to release Gemini 3.5 Pro, and show how hard the major labs are now competing for a small pool of senior researchers.

    Read more
  2. ChatGPT health answers reach free users

    OpenAI has updated GPT-5.5 Instant, the free default model in ChatGPT, so it now handles health and wellness questions about as well as the paid frontier models. The company says roughly 230 million people ask ChatGPT health questions each week, and in its own tests the updated model scored higher than doctor-written answers on accuracy and clarity. For the many people who reach for a chatbot before booking a GP, the free version now gives more careful replies, though it is still a starting point rather than a diagnosis.

    Read more
  3. DeepSeek raises $7.4 billion from Chinese backers

    Chinese developer DeepSeek closed its first outside funding round on 16 June, raising about 51 billion yuan (US$7.4 billion) at a valuation near US$52 to 59 billion. Tencent and battery maker CATL were among the investors, while a state-backed fund took the only voting rights and most commercial backers accepted non-voting shares locked up for five years. The round shows how much money is flowing into DeepSeek as a domestic answer to US labs, and how closely the Chinese state is now tied to it.

    Read more
  4. AI adoption is outpacing institutions

    The 2026 AI Index from Stanford finds generative AI reached 53% adoption within three years, faster than the personal computer or the internet at the same stage, with consumer value put at US$172 billion a year. Reported productivity gains range from 14% in customer support to 26% in software development, but employment among software developers aged 22 to 25 has fallen nearly 20% since 2024, and a third of organisations expect AI-related job cuts. The wider message is that laws, schools and workplaces are adapting far more slowly than the tools themselves.

    Read more
  5. Yann LeCun calls xAI a failure

    Yann LeCun, the former Meta AI chief now running his own company AMI Labs, called the xAI venture run by Elon Musk "kind of a failure" in a CNBC interview, pointing to the loss of its founding team and its retreat from frontier model building. Since merging with SpaceX in February, xAI has shifted to renting out GPU capacity, with Google alone paying about US$920 million a month for compute, while its AI arm posted a US$2.5 billion operating loss in the first quarter. The comments are a reminder that heavy spending has not guaranteed a place at the front of the field.

    Read more
  6. AI agents now most web traffic

    Cloudflare data shows automated traffic has passed half of all web requests for the first time, hitting 57.5% against 42.5% from humans, a point its chief executive had expected only 18 months later. As agents browse, compare and buy on behalf of people, they often skip the ad impressions and tracking that online advertising relies on. Businesses that measure marketing through cookies and impression counts will need to rethink how they reach customers who increasingly send software to shop for them.

    Read more
  7. Most Americans doubt AI helps society

    A Pew survey of 5,119 US adults found only 16% think AI will have a positive effect on society, even as chatbot use has climbed from 33% to 50% of adults since mid-2024. ChatGPT leads at 44% use and Gemini follows at 24%, yet 79% of people worry about misuse of their data and 76% distrust the accuracy of what these tools tell them. Adoption and trust are moving in opposite directions, which is worth remembering for any business putting AI in front of customers.

    Read more
  8. Nvidia Blackwell tops new training benchmarks

    The Blackwell chips from Nvidia posted the fastest result on every test in the latest MLPerf Training 6.0 round, including new workloads built on DeepSeek-V3 and an open GPT model. One submission trained the Llama 3.1 405B model from Meta in about seven minutes using 8,192 GPUs, and the newer GB300 systems ran up to 1.6 times faster than the previous generation. The results show how quickly training times, and the cost of building frontier models, keep falling for those who can afford the hardware.

    Read more
The Weekly AI Brief

Practical AI, distilled.

A short read every week — the few things worth your time, and nothing that isn’t.