Speaker 1: 0:04

Hi, this is Omid Ashtari.

Speaker 2: 0:17

And Natasha McElhone. Welcome to our new podcast. It's called when Shall we Meet. It's sort of undisciplined but interdisciplinary. It's perfect for any shallow polymaths out there.

Speaker 1: 0:32

Here's a medley of the upcoming episodes.

Speaker 3: 0:39

That's just defunct. It's finished as a kind of political model, and yet they're still addicted to it. Defunct, it's finished as a kind of political model, and yet they're still addicted to it. So I would say that that he's got a point about the politicians, less of a point about the civil service, and I think that if you're so, if you're a young person thinking about how am I going to make a difference in the world? Absolutely right, you can do it through business, you can do it through culture. There's lots of different ways you can do it, but ultimately, every single thing that you do is going to have, at some stage, contact with the political process somewhere in the world. So, therefore, we still have to have good people who are going to become our politicians. Yes, and my worry is that politics has become so unfashionable, so derided, so low, that we're ever narrowing the gene pool of the people who even think about doing it.

Speaker 1: 1:29

So tell me, what are we seeing in terms of female-male ratio on these dating apps right now?

Speaker 4: 1:35

It's a bloodshed. Dating apps are a bloodshed. Why? Because it's a very simple mathematical distribution problem 75% of dating app users, on average across the apps, are men. That goes to 76% for Bumble, it's about 75% for Tinder, 68% for Hinge. Not only there are significantly more men on dating apps, men also tend to give more likes. Men also tend to give more likes, so it's exponentially getting unequal between the two groups, which is leading to a place where the top 20% of men in a dating app cleans up, gets 80% of the likes. The remaining 80% of men struggle to get one date per month.

Speaker 5: 2:36

The way ChatGPT is trained is, roughly speaking, 90% unsupervised, 10% supervised. So the 90% is where it just reads the internet. It's just browsing the internet, learning everything from it. Where it just reads the internet, it's just browsing the internet, learning everything from it. The 10% is where OpenAI instructs that system to behave in a particular way. It says I know you know how to be racist, I know you know how to be sexist, but I want you to always speak in a polite tone and to mind these rules that we think are important.

Speaker 1: 3:10

This is human reinforcement learning, exactly Gotcha.

Speaker 5: 3:13

Now the reason it becomes a little bit more complicated than even that is that in that 90% unsupervised learning stage, what is the algorithm consuming? It's consuming text that has been written by humans, so it's still a human artifact that is consuming in an unsupervised way and that itself is a form of.

Speaker 1: 3:35

Supervision.

Speaker 5: 3:36

Exactly Right. It's still distinct from how a cat learns about its environment, because if you drop a cat on earth, it'll do it. If you drop it on the moon, it'll do it. If you drop it on a planet where it never sees a human, it'll still do it, but, chat, if you drop it on a planet where it never sees a human, it'll still do it, but chat, gpt will struggle on a new planet where there aren't any humans so he dropped his gun, grabbed his camera and I didn't have a film in the camera.

Speaker 6: 4:02

this was, you know when you need a film, put the film into the camera and and got me to stand between himself and the bear pulsing with a gun, like, come, come on, teddy, make my day, and all this in minus 50.

Speaker 1: 4:16

Yeah, exactly.

Speaker 6: 4:17

This was later in the trip, so maybe it's minus 20 or 30.

Speaker 1: 4:20

Oh yeah.

Speaker 6: 4:20

But all this happened super quick and then the bear turned towards us, started to dig his forefeet, lower his head, and we knew she was going to charge. And the bear can run up to 60 kilometers an hour, so maybe this ran in like 40 kilometers an hour. So it's from 20 meters. It's where you are. It's a very short time. But we still had to wait until it was really close up because we had only handguns and to save weight, we had short barrels, just two-inch barrels barrels, and if you're going to hit it in the chest which you have to to to not only kill the bear but also to stop the bear in its stopping power. So we had to wait and it was really close and the bibo fired so that was a close.

Speaker 6: 5:03

You know it was like you know who's going to who for dinner.

Speaker 1: 5:10

All the GPUs are going into big data centers owned by big corporations. So where are we going to bridge the gap between that vision of the world becoming a computer and that centralized infrastructure that we're seeing right now?

Speaker 7: 5:24

Yeah, and obviously data centers and the big supercomputers and others are the key points of infrastructure for this sort of machine earth that we're building. But what's been very interesting over the last few months to me has been researching this concept of dark compute. In fact, there might be two, three, four who knows how many times more compute available in the fans and smoke alarms and pregnancy tests and fridges and ATMs and everything that we've been consuming over the last 20, 30 years, which have way overpowered chips inside them. It's to say, the latent capacity sitting dark and idle in our homes, our cities, our doctor's offices, our airports everywhere is actually something that could be taken advantage of now that AI is the thing that's going to unlock that dark compute at the edge where the internet meets the real world anyway, um, I thought this was a really great discussion and we ended on a quite quite a um interesting philosophical point, but it was great to to go on this musical journey with you and appreciate it yeah, thank you very much.

Speaker 2: 6:26

Thank you so much, and you're not getting away with not playing any music, so it's happening, good to know.

Speaker 1: 6:35

I will play something. I will play something.

Speaker 2: 6:59

That was great.