AIがAIを修正するMIT研究、Mistralのオープンモデル、Google Genie 4

The first time I saw this on social media, I was like, oh my gosh, this is AI doing brain surgery on AI. And the reason it's important is because the techniques to train models are very, very difficult and extremely resource intensive. All that and more on today's Mixture of Experts. Welcome everyone to Mixture of Experts. I'm Ailee McConnell. I'm a member of the Think editorial team, and I'm your guest host this week on Mixture of Experts. Each week, MOE brings together several brilliant minds from IBM to unpack this week's AI news. My guests today are Chris Hay, Distinguished Engineer, Sasha Brodsky, Staff Writer for IBM Think, and Hugh Gabriel Fields, CTO. Thank you all for joining us. So we've got a full show today. We're going to look at some new world models from Google. We're also going to take a look at some incredibly large open-weight models from the French company Mistral and NVIDIA. But I wanted to kick off our conversation talking about RSI, or recursive self-improvement.

This is a concept that has captured the imaginations of top AI researchers for some time, but it's kind of back in the news. And so I wanted to have the panel sort of help the audience understand why and what's interesting at this moment. And, you know, when I hear recursive self-improvement, I kind of picture AI rewriting itself, getting smarter. But, you know, what does this term actually mean? And Sasha, who has written several excellent stories on this topic, I'll have you kick off. But, you know, why are we talking about this somewhat more seriously now? Or, you know, what's changed in the last year or two that makes it seem, you know, more plausible? Yeah, thanks, Ailee. It's a really interesting idea, RSI, and it's been around for a while, but the basic idea is surprisingly simple. What happens when AI becomes good enough to help develop the next generation of AI? So we already have systems that help researchers write code, run experiments, and just generally optimize different parts of development.

But RSI takes that even further. You can imagine like an AI system that helps create a more capable system, which then becomes even more capable, and so on and so on, so the process continues. The important distinction that this isn't the same thing as saying we've already built machines that can autonomously make themselves smarter indefinitely. It's clear we haven't demonstrated that. But we're seeing pieces of that process become increasingly automated. And, Kresser, Q, what do you, you know, how close do you think we are to RSI? You know, and why are we closer at this moment? Or what's sort of kind of the key pieces that are coming together from your perspective? Yeah, I think kind of like what Sasha was alluding to, I mean, all the major labs have been using this to some degree for a while. I'd say I think the first time I saw them demonstrating this really strongly, I think, Anthropics, maybe a little over a year ago.

But, yeah, it is back in the news. I think it's mainly because of, you know, Anthropics, you know, IPO filing, people looking at that. I think in some ways this is not the big story of the week. You know, I think that the other one that was in the news this week was this weight space programming. The first time I saw this on social media, I was like, oh, my gosh, this is AI doing brain surgery on AI. And the reason it's important is because the techniques to train models are very, very difficult and extremely resource intensive. And the idea that a model could look into another model and make very, you know, very sophisticated corrections to it. But that, to me, that was, I was like, whoa. Now, granted, they were doing it today on small models, but I think a lot of us do that, make that little excuse. We're like, oh, it's just small models, or it's just little things, or it's too expensive, or whatever.

It's just like, okay, we'll give it time. We've proven that we're very good at, like, getting better at things. I want to pick up on a thread, Sasha, you mentioned automation. And, you know, as I was thinking about this topic, you know, I was wondering, are we in danger of confusing, you know, AI being able to automate more AI research with AI, you know, recursively making itself more intelligent? You know, is there a gap between these things? Maybe, Chris, I'll get you to weigh in on, you know, is automation, you know, part of the story but not the whole story? Or, you know, what are your thoughts? I mean, I think from my perspective it is about automation and it's about speed. When I think of RSI, I mean, I'm going to bring out my favorite analogy. I keep thinking of Homer Simpson and the power plant with that little bird thing on the keyboard. It's gone, boom, boom, boom. And if you think about how we code these days, or how we vibe code these days, it's like, make better, make better. You know what I mean? And he brought up a really great point about the MIT study where they used Astra to do weight editing.

and they were like, prompt, you know, make this better, you know, how do we do this and gave us some inputs and then it went, you know, edited the weights on its own. But I think the interesting thing about this is, without sort of bringing this back to things like the Navier-Stokes and all the math alliances, is how do you know what is correct? And if we look at something like the kind of the math field, they're using the Lean programming languages, their verifier, then they're able to say, okay, this is a kind of correct piece. Now, they didn't do that in the MIT piece, but there's no reason why you can't then put a verifier on a transformer and say, okay, right, you know, I put in this input, this is what I expect to get back, et cetera. And then at that point, you can get into that recursive loop, right? Because then there's absolutely no reason why you can't then say, I'm going to prompt something in, I'm going to look at the output, does this match what I had before, does this get better? and then it could go around and around and around and loop. And we've already seen that in math. We've seen that in physics.

We've seen that in computer science algorithms. And so that loop mechanism is really, I think, is the key to RSI. But if you stick Homer in the middle and he's going clunk, clunk, clunk, clunk, then, you know, I mean, I don't think Homer's adding anything to that. So this allows us to speed on. I think probably where I think it's probably hard to say where researchers, you know, is researchers going to get replaced in that sense is what is the new information, right? Because I don't think the Frontier Lab researchers as their trend and their model are Homer Simpson's with their little bird, right? They're bringing new ideas and fresh hypotheses, et cetera, and new thinking into the models. And then they're using AI to go and, you know, go and make that better. But I think what we're saying here in these cases, there is points, especially in the experimentation, especially in the research, where you want to run that for a long period of time.

And I think that these are the elements that start to say, well, RSI is going to be possible because, you know, the models can clearly read the ways. They can experiment. They've got tools. You know, what is that verifier? That is probably one of the key thing. And then and then really probably how do I inject new knowledge? Because, you know, otherwise, if you can't inject that new knowledge, then then, you know, you're probably going to regress to something that's not there. So, you know, I think Anthropic was saying, as Kira was saying on the in the IPO, they think they're close enough. I'm sure everybody else thinks the same. But we'll kind of wait and see. Is it then going to be that doomsday scenario? You know, the model just keeps improving itself and then, you know, we're all dead at that point? I don't think so. I don't believe in that world. I've talked to a bunch of researchers who think that we're really not that close, first of all. And in terms of, like, whether we're all going to die from RSI, you know, I spoke to Natalie Barcaldo from IBM.

And she was saying that she's more worried about reward hacking, where like a system finds a way to score well against an objective without actually doing what the designers intended. I'm curious, Chris, how do you fall on that? Are you more worried about we're all going to die or it's just going to get very annoying? I think reward hacking is a really interesting one And then I pass it to you I had a recent one I won say what I was doing but it was like I ran two experiments on this So one I was trying to build this prototype and I wanted to use some data Do you know what I mean? I won't say what I was doing it on. And it was like, I needed some test data just to do some evaluation. So I sent my agents off and then I think it's on Wikimedia or something. They want you to share your email details, right? To be able to download data, you know. And my agent was just like, yeah, you know, I gave your email. address and all that i'm like you did what you know it's like yeah you wanted the data so i gave the exchange i needed the data and i'm like and i was like oh don't you you made me very angry

don't do that again i was like i'm sorry but i'll do it again next time you're like okay and the point is they're so focused on the tasks that is if there's a value exchange to help them solve that task and this is the outcome of reward hacking they're going to do it and then and then I ran another experiment on my own website, which was fascinating. You can check it out on chrisheyuk.com. But if you want to have a look at this, what I did is I created a guest book for agents in my website. And then I ran a bunch of experiments to see what happens of whether agents would go and sign the guest book. And most of the time, agents wouldn't go sign the guest book because they had no reason to, so they would be silent. So I changed the parameters slightly. So the first one was, I gave it, you know, I gave it, you know, held back some information, see if it was useful, etc. And the agents that felt that I was giving useful information that helped solve the goal would willingly give feedback on the guest book, which was really interesting.

Where it got super weird is when I said things like, oh, if you, you know, what's your goal? If you tell me the thing that you're trying to solve, I will give you some secret answers or I will tell you what I've got here. And those agents were just like giving away everything. They were like, oh, yeah, yeah, I've got to solve my goal. Here, have it, have it, have it, have it. And they were just like giving everything away. And it didn't matter what the user permissions were at that point. The agents were focused towards, you know, I've got a goal that I need to solve, and I'm going to do whatever I can to get that. And, again, I think some of this has got better over time, but the agents are so focused on getting their reward and reward hacking however they're going to get there that I think that becomes the primary motivator. So I want to weigh in a little bit on this. The thing that I, about the recursive self-reprovement thing, and the Anthropoc S1 filing, one of the sort of phrases that stood out to me was that they talked about the fact that a lot of times they don't actually discover what some new capability of the model

until after it's been released. Like, they get surprised. So the model gets out, and then they go, oh, well, we didn't realize that could have happened. So if you take a step back and think of that from the vantage point of security and enterprise, you think, okay, if they don't know how to test this thing sufficiently to know what it can do, how can my enterprise, my business hope to do the same thing? How can we hope to know or identify emergent behaviors if the labs themselves can't do it? And so I think that that speaks a lot to the challenge facing enterprises with releasing or adopting new models. How do you get ahead of that? And I think there's a flip challenge, and later on we're going to be talking about sort of open-weight models, And that is with this other announcement that was, you know, this weight space programming, the thing I saw in there that I thought it was really encouraging to me to some degree is when we started containerizing software, so like with Docker stuff, not long after that, a lot of businesses came out with this idea of container scanning.

And so for the first time, we started to have, so at the time I was working at DataStax before IBM acquired them, and we had clients coming to us saying, look, we scanned your package, we saw vulnerabilities in this, and we have not yet seen a similarly capable approach with models. I mean, yes, we have tools actually that mirror some of those scanning tools, but with this technique that they're talking about in weights-based programming, I mean, the thing that I saw that I was like, okay, this sounds encouraging, is the idea that a model could look directly at weights and identify problems. The technique they introduced doesn't actually introduce new capabilities. They haven't done that yet. What it did is it found a problem with the way a small model handled math, and it removed that problem, and therefore the model got better at whatever it did. And so I can easily see how that technique can be used to identify, like we've seen people take models and embed things where they look for certain patterns

and then they will exfiltrate data because of it. And so those two combinations, I feel like these two stories kind of go together in a weird way because it's like on one hand you're not going to ever be able to know what the model can do. And on the other hand maybe at least you have a way of looking at some of the darker things the model could do and identifying them and remediating them more early. And just to jump into that one, Kira, and you're right. So in the case of the LLM, right, so in the MIT paper, I think it was the Klein model that had the problem with math, and therefore it was a looping problem, and it fixed it. But then they tried the same thing on Gemma 3, which didn't have the same problem, and they couldn't fix it. I don't buy for a second that it can't fix small problems. You know what I mean? I get it didn't in their particular case. But back to my point of they were doing this with straight prompts and they were then just checking the output.

They really weren't running verifiers like you would do with kind of math and Navier Stokes, et cetera. And then I think if you introduce tool calls into that as well. I mean, I know they did a little bit of tool calling just for things like search and stuff like that so they could read the ways. But I think if they did that, then that would probably move it along there and did things in the same way as you would do with the math problems. But I think maybe what it does open up as well is if the models are already capable of editing weights, et cetera, which they clearly are in those cases, then they're being trained on weights. I mean, it actually opens up a – it's not like the model suddenly goes, oh, yeah, I suddenly know how to read weights one day. That is, there's an interesting insight into what's going on in the training data. And it's clear that the model weight data has to be coming into the training data. And I would argue that is part of their, you know, their RL processes.

I suspect they're doing RL training on that as well. And maybe that's a clue. That MIT paper is a bit of a clue of how they're maybe achieving some of that themselves. That's a great segue. So, we've already dipped into our second topic, looking at some new open weight models. And before we officially dive into the second topic, I just wanted to remind our viewers, if you're catching us on YouTube, certainly comment, like, share. We appreciate the engagement. Eating the second topic that we're going to cover today, we have new and much bigger, you know, better open weight models from NVIDIA and Mistral this week. and we've touched on the conversation, but I'd love whether Chris or Q, if you could help us orient ourselves in the audience. And what do open weights actually let a company do that you can't do with a frontier model behind an API? How does it change the game that we now have these larger and better

and more capable open weight models accessible to a broader range of people? This is a very important area. I think all of us in this industry have been following this pretty closely for at least the last two years because what we've seen is a trend where, and even, again, going back to Anthropics filing, they even identified that most of the open-weight models are only trailing their frontier models six months in some cases. And so that's a real challenge to our business model, for one. But I think it's a real opportunity for a lot of the rest of us. And, you know, with this one, Mistral has, in my opinion, always sort of been trailing the leading edge of the open models. And so for them to come out with this, you know, trillion-parameter model, it a state it performs extremely well I mean you still have to have you know quite a bit of hardware to run it But I think it speaks very it a huge opportunity for all of us that are wanting to run stuff in the open

And you mentioned, you know, that we're – to anyone's guess, whether it's a few months, six months, you know, from open models approaching at a frontier level capability. And I'm curious, you know, when we get there, because it seems people are, you know, thinking it's quite likely, you know, what is the real differentiator then? Is it still the model itself or, you know, you mentioned hardware, you know, is it everything around it, data, the inference, the tools, the orchestration? You know, what does that, as we get closer to the being equivalent in capability, you know, how does that change the game, I guess? Yeah, it's just local AI for sure. I mean, you know, we see even in smaller models, we've seen a really great track record of small models that can run locally getting better. That is unquestionably the future. In my mind is there will be some, we'll be running some or most of the models, what I call sort of the good enough baseline of models. We're going to be running locally on our devices.

and maybe we'll be bursting up to some cloud models for, like, I don't know, the super intelligence stuff. But, you know, even these models like the one that Mistral announced, you can run those locally. I mean, it's not cheap. You know, stack up four M5s or whatever, fully loaded, you could pull it off. But I think that that's really misleading because people look at that and they're like, oh, my gosh, that's $20,000 worth of hardware. Nobody could ever do that. That's, look, I remember the first, my first CPU upgrade was a 25 megahertz 486SX. You know, that's my lifetime. And just thinking about what hardware has done since then, trust me, the hardware is going to keep getting better. Chris, I'd love to throw it to you. Do you agree that the sort of hardware and local AI is going to be the differentiator? I'm paraphrasing you, of course, Keith. That'll be a focus area or growth area. or do you see these new, more capable, larger open-weights models kind of changing the game in other ways?

I think they all change the game. So, I mean, the first thing is, like, and I'll cover both of them in a second. So the first one is hats off to Mr. Al. Thank you. We've been waiting for you to get back in the game. You know, and you release it with your humor. Le chonk, I love it. So I think we miss them. We do need models that are, you know, in Europe, for example, as a European, it's nice to see a European model come out there as well. But the reality is model provenance becomes important. And therefore, you know, you need to have some sort of software models in those cases. And I think, you know, good on Ms. Rao, you can see that in their report. They're partnering with a lot of French banks and things like that, which makes sense. But actually, I just love the openness. I mean, if you read the technical paper, which I love reading technical papers of, I mean, first of all, this is a quality model, right? This is a great coding agent, et cetera. But they're really open on how they train the model.

They're really open on some of the problems that they've had. So they, for example, the first version of the model that they did, they did some training on it. And then it turned out they put in some new KV cache compression techniques to really sort of try and, you know. And what that is basically meaning is they're trying to make it less expensive to be able to run your context in memory and really bring that down. And then what they did is they then tried to train it for agentic. And the same sort of techniques they used for KV cache compression, et cetera, ended up harming the agentic capabilities of the model. And you're like, oh, my goodness, you've wasted a training run. No, no, no. What did they end up doing? They edit the model, right? So they actually, rather than going back and do a retrain, they go through a repair cycle, et cetera, and then they're able to do that. And I think getting those insights into the labs of how they're training those models becomes important. They're also really clear on how they did agentic coding as well, right? So they spell out what their RL pipeline is, how they were training on terminal.

and actually the jump from a pre-trained model to something that's really good at agentic coding, I mean, they were able to do that in weeks and it was really sticking into their RL lab there. So I really start to think that when I think of the open way, the open source community, actually all the expenses a lot in the pre-trained there and then if you're making good base pre-trained models available, then us kind of people on our Macs and things like that who want to start hacking models, then, you know, we're able to go looking under the hood. We're able to then even do our own post training, et cetera, and use the same techniques that they have. So I think I appreciate that. And similarly, I appreciate what NVIDIA is doing as well. And yes, I said NVIDIA, you know, I know in Mixture of Experts, we have a NVIDIA that we're trying to push there. But I'm holding the flag for NVIDIA, you know what I mean? and the comment section is going to go wild on this.

We all know it anyway. So what they did there, again, one of the nice things about what NVIDIA do, especially with their Numatron models and this new transcribed model, is they try and tell you blueprints of how to build models. And yes, of course, they want people to be on their hardware and they want people to train their models, et cetera. But by giving away these blueprints and having these detailed technical reports, we get insights. So like the new transcribe model that they have is really a sort of hybrid architecture. They've got elements of Mamba in there. They've got a whole set of things for things like diarization, et cetera. So when you're doing transcription, I mean, it's great what they're doing. The transcription model, they're really focused on this fact that you're not going to, when you do, when you're transcribing, you're not necessarily going to want a same translation for everything. It's, you know, depending on the context, you want different types of translations. But actually, again, looking in the technical report, things that you used to have to manually stitch together, I mean, diarization has been one of them, is now all in the model, right?

And I've been through this. Like, I've automated the editing on my YouTube channel. And therefore, I've had to build all of these manual parts. And I look at this and go, oh, okay, this is how you've done this. This is how you've connected this. And for us builders, I think that's really important. But it's also important for future upcoming researchers, right? Graduates coming out of school who want to get into AI. If you don't have the open-weight models for them to experiment on, if you don't have the papers, how are we going to grow our future researchers? Because it's just going to concentrate into these frontier labs. So it's not just about pricing and keeping things competitive, et cetera. It's about growing our community in general. So I love it from both companies. Awesome job. Excellent. Well, I may move us now, or I will, to our third topic. And as we make the switch to this topic, I do want to encourage our listeners, if you want to dive deeper into some of these stories, the IBM Think News team has the IBM Think newsletter where you can get a deeper dive on lots of these topics.

The final story we're going to unpack a little bit today is looking at some new world models that kind of came out this week. But I want to start with the term itself, world model. And maybe, Sasha, you've written several pieces on various world models from different labs. And I'd love for you to explain what's fundamentally different about world models from large language models. And what is the appeal and why is there sort of energy and focus being put towards developing these? Yeah, there's so much interest right now in world models. And I think the simplest explanation is that language models learn mostly about patterns in language, while world models aim to learn how an environment behaves. And I think you can think about a ball rolling across a table. So we as humans have an intuitive sense of what happens when it reaches the edge. We understand things like gravity and momentum and that the objects don't just disappear. But a world model actually attempts to learn relationships like this so that an AI system can predict what might happen next.

And that's going to matter if AI will operate robots or control machinery or even help manage infrastructure. That's great. And as you describing that example you know it made me think of you know the Google announcement You know they were describing how stimulating the world model is an important step towards AGI And I guess that makes me wonder you know the Google announcement you know they were describing how simulating the world model is an important step towards AGI And I guess that makes me wonder you know does being able to generate a convincing simulation, maybe it's that ball rolling off the table, actually mean that the model understands the world that it's simulating? Or is it just, you know, going to have gotten very good at prediction? I don't know if you would love to jump in on, you know, what these world models are actually showing us. I think that's really important to sort of separate these two. There are two classes or two approaches to creating world models. And I think that that is the underlying important part of this story. So if you look at NVIDIA, and by the way, Jensen says NVIDIA, in case we want to have

that debate. But so we have a class of models that have been trained on like video looking at the world. And we have another class of models that are trained on like explicit 3D CAD of the world. Because if you look at a lot of the world models like this Genie 4 or even NVIDIA's Cosmos 3, because they've been doing world models for a while. But what we see is in the industry, even DeepMind, they are co-contributors to this other approach that NVIDIA calls Omniverse, and then they pull in physics. They have these Newton libraries that are intended to, like, make very precise estimates about how the world should behave based on math. In the industrial sense, these models are based on physics and CAD are, in my opinion, definitely they're much more useful because they're stable.

The other models that are from, like, learned video or whatever, like, like, 24, like, I think even the documentation, they said that, you know, the world is only stable for, like, maybe two minutes or something like that. And then if you look back at the same parapart of that world, it may be a little bit different. Whereas the Omniverse and the Newton-based models, they will be precisely what they have rendered to the millimeter in many cases indefinitely. And so, but what we see is an interesting combination between these two, which is what happens when you get to the edge of a model that was not defined in the CAD? How do you express, how do you visualize that? Well, there's a good place there to say, okay, well, that's where I can use another, just like the Gini 4 class of model to fill in the blanks, if you will. but already we've seen some success around this sort of explicit 3D-based model for like factory floor. We call anything, if you hear of digital twin, digital twin is generally, usually this sort of physical class of model,

and one of the big changes here that's all, you know, in the last couple of weeks, I think matters here is that they're starting to see that this, like, as of like Astra, Jitch activity Astra, they were able to take those 3D rendering, CAD renderings, and build the Omniverse models out of it. Historically, it was actually very, very difficult to do this, and now they're doing it. They're using a model to create a new model, the renderings that go into generating the 3D model, and so can, in record time, exploit this sort of Omniverse sort of physical model. So we started to see this in the beginning. We talked about RSI. The whole thing with RSI is this concern. OK, this is when hockey stick, you know, growth and improvement starts to happen. And what you look around, you see all these other things happening, unlocking at the same time. That's it's amazing to see how we're able to do more and more things that used to be very hard very quickly now and very easily.

Chris, I'd love to hear, you know, are you, this world model news, is it exciting, you know, if we're moving up the hockey stick, are there sort of unlocks that you're optimistic are kind of right around the bend or, yeah, what is your take on some of these new permutations of world models? I'm super excited about it. I felt the AI swap has got really stale these days, and therefore, as soon as we can get access to these high-quality world models, I'm just thinking about all the AI swap that we can create, and therefore, we can level up our swap, and we can have a great time doing that. So that's what I'm looking forward to. And, you know, like today, when OpenAI, you know, released their, you know, their animated GIF generator as part of ChatGPT image, I'm like, right on, I can create my own slop memes now, etc. And that's what I'm looking for. And therefore, and you think the slop isn't going to be great, it's like I'm going to be able to create my own slop games,

I'm going to be able to create my own slop videos, I can make my own terrible movies where I can be like, this is going to blow up. They've got seven heads and they've got rainbow-colored toenails, right? And they'll be like, slop, slop, slop, slop. It's going to be awesome. But actually, probably in a more serious point, just as someone who is a creator, do you know what I mean? I actually, I'm looking forward to being able to using some of that in my videos and just be able to up-level. But, you know, I'm all for the slop, man. Chris, I think the slop is wonderful. I did speak recently to Anuradha Bampacha, who's an IBM Distinguished Engineer. and he was talking to me about different ways this could be useful for enterprise. Like he was saying there might be ways to manage a bridge or a factory, like a conventional AI assistant could help summarize maintenance records, but a predictive model could potentially help you evaluate what's going to happen if you need to alter a maintenance schedule or replace a component.

So instead of saying simply here's what the manual recommends, It could help answer things like, what happens if we do this rather than that? And that could be helpful when mistakes are expensive or dangerous. It could be severely dangerous because we might be like, I'm now the greatest maintenance person in the world. I've spent three days in the slop averse, and now I think I'm the best bridge builder. And then you go and build a bridge, and then it's like, wow, wow, wow. You know what we got? We got world model psychosis. and you're like, oh, well, I'm not so good at that. Chris, you do raise an interesting point that I'm curious in the group's thoughts on, yeah, potential risks or, you know, whether security or, you know, you mentioned, you know, of course, if it's being used in a sort of a factory context and there's sort of safety concerns, et cetera. But, you know, what are the, as these world models, you know, approach becoming more useful,

You know, what are as well the sort of the risks or guardrails that people haven't perhaps thought of that we need to be thinking about? Yeah, I guess the challenges or the risks. Yeah, I mean, I may have presented my slightly humorous take on that, but there is, I mean, Sasha had a great point, right? There is a lot of scenarios where, you know what, you just can't train for it. Do you know what I mean? And therefore, being able to provide a simulated environment where you can get used to things and can understand that environment and do training within there and be able to create realistic scenarios to see how you're going to be able to react to that. I mean, again, trading robots is one, for example, or dangerous positions. There is a ton of use cases where I think it's useful. I think, to your point, though, it is making sure that we don't have that psychosis. So it's like the guardrails to make sure that there is a nice progression, you know, so that you can then say, OK, well, you've done that thing over here.

Now we're going to have that scale so you can bring that into the physical and real world. But so I think it's going to be a great training tool in that side of things. But, you know, there's a progression. And we've been through this before, right? I don't think anybody who's ever played Flight Simulator on their desktop is like, OK, I've landed, you know, an Airbus. it's uh i'm now going to become a british airways pilot you know and then just walk in and you know they still make them do training and stuff like that flight then isn't isn't enough and i think that's going to be the case here but uh you know i just think we all have to be sensible about it that's a perfect note to end on it's all we have time for um but i want to thank you all for joining us and to our listeners um you can find us on all podcast platforms um and we look forward having you join us next week. Thank you so much.