Briefing Chat: Baby T. rex were killers from birth, suggest new fossils

Hello and welcome to the Briefing Chat podcast, the Friday show where we talk about a couple of articles that have been highlighted in the Nature Briefing, Nature's daily email newsletter of the latest science news. I'm your host Nick Petrichow and joining me today is Lizzie Gibney. Lizzie, how's it going?

Hi Nick, thanks for having me. Well thank you for coming on and you know I've got a very interesting story I think this week to tell everyone about but I know you've got maybe even more than one to talk to me about tell me what you are most interested to talk to me about this week. Well I'll start with the story

that I wrote a little bit of bias in there it's a story on how London is becoming an AI capital of the world but a very specific one kind of AI safety is something that London is like carving out a niche in so you might remember the mythos model for Anthropic from a month or two ago that It was like the most powerful model that was released.

And there were all these fears about hacking happening because it was so good at breaking cybersecurity. But it actually took like a London organisation, the AI Security Institute based in London, checking it out, evaluating it before the whole world's researchers said, actually, yeah, this is quite worrying, but maybe not as worrying as some people had said.

So this institute has kind of gathered steam and is the best resource of this type in the world. and it's got loads of really good researchers working there as well, which is quite hard to do as a government body. And so, yeah, the story is about how London is kind of carving out this little niche for itself.

So is this institute, I guess, sort of like a centre of mass that is attracting other things with its gravity? That's exactly right. So, I mean, London, when we think about London and AI,

already like deep mind is probably what springs to mind. So that was founded here and crucially stayed here, even when it got taken over by Google. But now, like if you go walking around King's Cross,

which is where nature is also based, There's going to be an anthropic office, OpenAI. There's all these other kind of smaller organizations, and they're all based here in King's Cross. They're calling it the Knowledge Quarter.

And some of them are quite specific to AI. So there's also Apollo Research, which does a lot of work into scheming. Have you heard that word? It's like when models effectively try and, I don't like using the anthropomorphic language, but to deceive, you know, the people running it.

They do things that they are not supposed to. And Apollo Research tests for that. And there's GovAI, which was a spin out from the University of Oxford. So there's kind of like a long history of responsible AI research, ethical AI research happening in London.

And it's now booming. Does this present any issues, though? Because, you know, when you think about the big tech companies, you think of California. Obviously, it's a long way away.

So does it work having a AI safety capital in London? It seems to. I think there's actually an advantage in that being outside of that Silicon Valley bubble, there is just a maybe a bit less hype over here in the UK like there's a different lens on how

people view AI that was the feedback that I got when I wrote this story but what I would also say is this kind of long history that the UK has of AI safety research has kind of now got a Silicon Valley gloss on it though so to go back ancient terms in in AI 2023 was when Rishi Sunak who was then prime minister about 12 prime ministers ago so you might not remember that either um he had

this AI safety summit. It was the first one that had happened. And again, that was kind of putting forward the UK's pedigree on AI safety research. But since then, obviously, Trump has come in, and that's put a very different slant on the kinds of things that people are thinking about when they think about AI safety. And it's now very much about security. It's about cybersecurity.

It's about not developing bioweapons. It's about national security, like how can we stop these frontier models from doing really dangerous things? And that's not really what are the only impacts of these models as we as everyday people who use them know you know there are psychological potential issues there have been many many cases of AI psychosis there's how it's affecting

education the justice system exacerbating inequality and all of those things kind of come under this idea of ethical AI or safety in AI and that's not what's being tackled so it's kind of great that we have somewhere that is taking AI safety seriously and is kind of flourishing in that way. But I would say it's not doing the whole job. We've got this kind

of Silicon Valley idea of what AI safety should be about. And do you think that will be something that they look into in the future? Or how do you see this progressing? It seems that very much because of the geopolitical situation that we're in,

it seems like that is the direction we're going to stay in. So for instance, the AI Security Institute used to be called the AI Safety Institute, but it changed its name. So I think this direction And the we're in for now is what's going to continue. But there is room for companies, you know, like, as I say, everyday people are using this.

And we want assurance that if you're using an AI agent, it's not going to leak all your data. It's not going to crash your computer. Like there are lots of ways where using AI can go awry. And that's maybe an opportunity for companies to spring up that try and help other companies and the public to make themselves safe while using AI.

So that is also something that is growing up a little bit here. So maybe that's a way forward. But is this sort of centre of gravity being in London enough? Is there more that needs to be done to ensure AI safety?

There are a lot of conversations happening about regulation of AI. So the European Union has an AI Act.