Nish Kumar

It seems nonsensical.

Yeah.

When Jimmy Wales, the Wikipedia founder, was on the show, we talked a lot about AI hallucinations.

And he's somebody who's really, really interested in large language models and uses them a lot.

But he did talk about the limitations of this.

He talked about asking a chatbot who his wife was and was told that she was, in fact, married to Tony Blair.

And her wedding to Tony Blair was described in great detail.

Why are these hallucinations happening and how can we spot them?

One of the things that Jimmy said was he said that he was surprised with how much something like ChatGPT had spent more of its time on the interface, on that tone of flattery that it adopts with you, rather than actually fixing the hallucination problems within the actual system.

But then is that fixable or is this...

Listen, I read something the other day about AI and I thought it was really interesting and I thought it would be good to ask you about to explain this to me because I didn't understand it.

This idea that the actual more effective AI, and I think it goes back to what you're saying about the potential for therapy bots, right?

There's this idea that the version of AI that we have now is big data, small task, massive data sets, but then you actually ask it

what should I have for dinner?

And actually the more effective is if you invert that and you put small amounts of data, highly specialized data, and then you can use it to yield a big task.

Is that the shift that's happening or the shift that you think should happen?

Pod Save the UK is brought to you by Vanta.

That's why Vanta is a game changer.

So whether you're prepping for a SOC 2 or running an enterprise GRC program, Vanta keeps you secure and keeps your deals moving.

Get started at vanta.com forward slash PSTUK.