A week in Generative AI: Bernie, Mistral & Runway

News for the week ending 7th December 2025

Dec 07, 2025

This week wasn’t one for big headline, frontier AI releases, but its been an interesting one with a great video from Bernie Sanders about the threats from AI, a huge set of open-weight models released from Mistral and Runway launched their latest text-to-video model.

There was some interesting Web 4.0 news this week as well with reports that AI shopping grew significantly year on year over the Black Friday period with both Amazon’s Rufus and ChatGPT driving sales. There was also a report that ChatGPT’s growth has stalled and lots more rumours about their incoming advertising product.

There’s also a good Long Read from Stratechery on Google, NVIDIA, and OpenAI that’s worth checking out.

The threats from AI are real

Sometimes Bernie says it best, and it’s great to see a US senator taking a proper look at AI technology and the issues we’re all facing around it. My only bugbear in these debates is how we talk about risk: people frame the ‘threat of AI’ as if the technology itself is dangerous, when in reality the danger lies in the choices humans make when building and using it. Technology isn’t a threat on its own.

Below are some of the questions that Bernie raises and says must be answered as AI brings about an unprecedented transformation of society:

Are we comfortable with the Billionaire class controlling AI without any democratic input or oversight?
What impact will AI and robotics have on our economy and the lives of working people?
What impact will AI have on democracy?
Could AI redefine what it means to be a human being?
What impact is AI having on our environment?
How will AI and robotics impact foreign policy and warfare?
Is AI an existential threat to human control over the planet?

The video is 15 minutes long and well worth a watch - it gives a good overview on the big important questions we’re facing around AI as well as some good examples of the challenges it’s presenting.

Source

Mistral closes in on Big AI rivals with new open-weight frontier and small models

This is a huge release from Mistal, 10 open-weight models in total that includes a large frontier model as well as smaller models that can be customised and used on device (i.e. offline).

A lot of Mistral’s business customers are looking for models that are faster and cheaper to use, as well as easier to deploy. They’re also less worried about standard benchmark comparisons as they think that the value for businesses is when they customise their models and rate them against their own internal benchmarks.

Their new large frontier model, Mistral Large 3, has both multimodal and multilingual capabilities and has a context window of 256,000 tokens. It’s suitable for powering document analysis, coding, content creation, AI assistants, and workflow automation.

Mistral are leading the way for AI research in the EU and its great to see this release from them.

Source

Runway says its new text-to-video AI generator has ‘unprecedented’ accuracy

There’s been some really impressive progress made on text-to-video models this year and Runway themselves only released their Gen-4 model back in April. Since then we’ve also had Veo 3 (May) and Veo 3.1 (Oct) from Google, and Sora 2 (Oct) from OpenAI.

I think we’re now in the territory we were in with text-to-image models around 18 months ago where most of them are now ‘good enough’ and are able to generate mostly realistic and convincing content. They’re incredibly powerful tools that should be used responsibly, but I think we’ll see a lot of creative uses of them next year start to hit the mainstream.

Source

Web 4.0

AI Ethics News

Long Reads

Stratechery - Google, NVIDIA, and OpenAI

“The future is already here, it’s just not evenly distributed.“
William Gibson

Discussion about this post

Ready for more?