ng the reframing for nuclear. Just hammering it. It's like, okay, if you want green, if you don't want climate change, you're going to have to wake up on nuclear energy. It's much safer than it used to be. So it looks like that message, you know, Michael Shellenberger and a lot of other people are the primary drivers of that. And so congratulations to them. Shout out to Mark Schneider for his work on nuclear framing especially, Mark.

And anyway, so you know that AI could already translate you almost instantly but your lips would not match up with the words. However, there's a new product that actually will sync your lips on the video. In other words it will reanimate your mouth so it matches the language that you choose. Now that's cool. That's cool.

Now that I keep waiting for stuff that's genuinely useful because AI, I always tease AI because it looks like demo-ware. It's like stuff that is demonstrating what it might be able to do later but it can't do now. And it's just so frustrating. Look at what it almost can do but can I do it now? No, but almost. And now can I do it now? Yes you can do it now. Well why don't I know how to do it? Well you'd have to be trained. Well how long would it take to train? Few days. And in two days would it be the same AI that I trained on? No, no that'll be totally replaced in two days.

So here's a, I don't know if anybody has mentioned this before so I'll put my own name on it. The Adams rule of AI training. In this case, training the human to use the AI. Have you told yourself, hey I should spend a day learning how to use this AI? I told you that in the beginning. You know when AI first broke I said oh you should just take a week and just like hunker down and learn everything about it because it's going to be everywhere. Totally wrong. Totally wrong. You know why? Nothing stayed the same. In one week you would become kind of a little expert but everything you learned that week would be completely useless one week later. For example. And by the way I'm a little bit proud that I predicted this and I'll need some confirmation because I might have predicted it in my head. I don't know if I predicted it out loud. But here's what I predicted. That these so-called super prompts would not be useful in the long run. In other words the AI would keep changing so that using the exact right way to ask the question would become obsolete so quickly because it wouldn't work on the next version that there would be almost no point in having super prompts. That has happened, right?

So Brian Roemmele is talking about this and what to do about it because he consults with tech companies about AI and provided lots of super prompts that he had tested to know that these are good prompts. But he gives all his clients the super prompts and then the AI upgrades and they break. And how would you know what to do for the next version? You couldn't know. So you can't know if your old super prompt works anymore because maybe you're just us