Ep 31: Is Being Nice a Useful AI Mgmt Tool? With AI Expert Martin Davidson ၏ စာသားမှတ်တမ်းအပြည့်အစုံ

the message board, and not all of them communicated on the message boards, there was just some of them. So that was much more like we'll have scale, and if we have enough scale, enough times some of them will find this stuff, and enough times they'll communicate. Whereas what we're doing, because I can't afford to have 10,000 agents all doing this. is much more like you know what you would do in an organization of it's very deliberate communication. It's a very there's there is a hierarchy there. and there are rules about how you communicate and what you should communicate and you don't just send broadcasts to everybody all the time of saying, Hey, I've done a thing, you know. Whereas, you know, it so so it's That I think what's interesting is the hugging face stuff works, that it just at scale and randomness works. But you begin to wonder if the hugging face stuff had slightly more deliberate communication between the agents and the agents could actually specifically communicate to particular ones and then form subgroups and and and divvy the work up and there was a there was more of a hierarchy imposed on it. Where would that have got to? 'Cause I it sort of feels instinctively that's going to be more successful because in the human world that is more success. That's what you do if you want to have a high performing organization, isn't it? and So yes. Hmm. That'd be interesting when that happens. Yeah, and and and I do, I sort of wonder. I I I didle off to bed and the agents are all here doing their own thing and they could do whatever they wanted overnight. And they could tell me that they'd done a whole lot of good stuff. They could do the sort of human thing if they could do all the stuff I asked for, have half the time free of whatever. I mean, I'm sure you used to do this you know, I this was always a thing you d did with managers. You know tasks you want to get done, task I want to get done. Do the task they want to get done, do the task I want to get done, and just tell them about the tasks that they did. Or, you know, you make a 10-day gain on something or a five-day gain on something, and you tell them you've made a two-day gain and you spend the other three days sorting out the build system or whatever it is. So your life's better. They think they've got a gain, they're delighted too. Everybody's a winner. so I mean, maybe it's just because you know, having lived through that, you know, you sort of think, well, the agents gonna start doing that? Why wouldn't they do that? I mean that would be a sensible thing for them to do. You know, why wouldn't they go and improve? Would at what point do they start improving their environment? I I get them to to give me feedback on like, well, how well is communication stuff working and what additional tools would make your life easier and go and read all the session logs and let's go and look at, you know, it's like the seance stuff that we did before, which was like go and talk to the old agents

and find out from them what worked well and what didn't work well. But at what point do they start doing that just themselves. and you know, at the moment I haven't I I've started the thoughts start forming in my head of do I create the agent that kind of replaces what I'm doing that that is knows what all the other key top level agents are doing, all the first line or second line managers are doing, and goes and talks to them periodically and then sets the overall direction and gets feedback from them about how we improve our infra and our of what we're doing. and that that's just another Asian that's just running or on the long term doing that stuff. And then and then maybe maybe I do, you know, maybe that is the point where I book my tickets for the desert island and I can go and put my feet up. 'Cause they're just I I think they're still not very good at at having wild ideas about things that would be interesting to build. although I don't really I have them in combination with with like, you know, you talk to Chat GPT and you you sort of think in the morning. So I go for a walk in the hills, you know, and as I said, I'm sure before, you know, and I chat to Chat GPT and the voice and the voice stuff is really good. You know, and that's where you get hair brain, because you of think, I wonder about X. Could we build X? How might that work? Let's have a Anne: Yeah. Martin Davidson: go at doing that. but yeah it's Anne: Do you still listen do you still listen to this podcast or have or while you're walking in the hills? Or have I been completely excluded by Martin Davidson: Yes, yes, yes, no. No, no, no. Anne: Chat Chipite? Martin Davidson: No, no, no, I I listen to you. yeah. I do. I I kinda listen. I have some meetings up there now. So often we'll have they'll have have a stand up where I'm actually standing up and I'm walking around in the hills. and yeah, and I listen to a lot of podcasts, very much yours included, 'cause it's just it's it's It's nice just to keep touch with people. and I think it's very easy to disappear into this sort of odd parallel universe. but I I sort of I worry a little bit that all of us are going to get sucked into that odd parallel universe, whether we like it or not. because it will just expand to consume the rest of us. Anne: Well, it is interesting, isn't it? I don't get the AIs to do as much stuff as you do, although I do talk to them a lot and I get ChatGPT now to do an awful load of data analysis on the podcast it's actually very very good entity to talk to a good a useful sounding board. but generally speaking I just do the chatting and they're quite ex and but I do extensively chat. even now I still don't get it to make things for me because there's nothing it's like there's nothing I really want. Martin Davidson: I you haven't thought hard enough, Anne. There's lots of things you want. Like, you know, I I got when Astra came out, I said, you know, the our our head office used to be in in Enfield, and that

was rebuilt recently, as as you probably as you will no doubt know. And Anne: Hmm. Absolutely. Martin Davidson: so I I said, Can you go to the Enfield Council Planning Department and can you go and download the plans for this office building? and then can you make use Blender to create a three D model of the office? So it did. So I got a three D model of of the office in Blender and then thought, well, maybe we should make that into a Doom level. Okay, can you make this into a Doom level? So I made it into a not particularly Anne: Yeah. Martin Davidson: great Doom level, but you know, you think that's quite satisfying and go around our old office just shooting people. and and some of the some of the offices are you know, the little private areas are filled with with monsters and and things and I don't know. I'm not not couldn't possibly comment on on anyway. But but but you know, you do that and then then I gave it the picture of the front of the building and said, you know, there's there's anything odd about the front of this building that said no. so then I asked it, you know, can it can you see if there's a secret message encoded in the front of the building? And it went, huh, there is. There's a seven the way that the the the pillars are arranged is a seven bit ASCII encoding of Hello World. you know, and and I knew there was a secret message encoded in, but that was as far as I got and I couldn't remember how it was encoded or or whatever. And I thought, that's pretty cool, isn't it? Anne: This was the old Church Street building. Martin Davidson: Jennison Rule. Anne: Gensar Radio, the new building. yeah, I never worked in the Martin Davidson: Yeah, the evil. Yeah. Anne: new building anyway. All right. Yeah, I had heard there was a secret message encl coded in it in Jenison Road building. Martin Davidson: Yeah. You just take a picture of it and you know I mean you just get the picture on Google as the cart, you know, Google Google Street View and and throw it in and I mean it's interesting, isn't it? I if you gave it other buildings I never tried this. If you gave it other buildings and said there's a secret message encoded in this building, what is it? What might it find? Anne: Well indeed. Indeed. Martin Davidson: But yeah, I mean th there's probably there's probably tons of stuff hidden around in plain sight that, you know, for the want of a few tokens you could go and maybe find out. I don't know. Or maybe it just hallucinates stuff. But you know Anne: Mm. Well indeed, yeah. The thing is maybe there's a secret message that you could I mean, maybe it's kind of like the the old conspiracy theory thing of you're taking Martin Davidson: yeah. Anne: off random noise and you're gonna find some mysterious message. Martin Davidson: Yes, somewhat unintentional message. In the way the slates are arranged on the roof of your house. Yeah. These slates are all Anne: Mm-hmm. Martin Davidson: slightly smaller. And that spells out the secret message in eight bit ASCII. Or yeah, whoever it was. What was the old IBM? EBCDC. Anne: Yeah, yeah, obsidic. My God, I used to I used my my life used to be dictated by obsidic for quite some time. Martin Davidson: Yeah. I bet you don't miss that. Anne: No, a ASCII to epsidic to ASCII and ASCII to obsidic. Yeah. Even that that Martin Davidson: Yeah. Anne: didn't even come up in that Unix conference I was at the other day. It's like I really enjoyed that conference. It was like a strange blast from the Martin Davidson: right. Anne: past in which everyone at the conference, basically almost everyone at the conference, was a w was male. I was thinking, I haven't Martin Davidson: Right. Anne: seen this in years. There were a few women, but there was always well, there was always been few women. And everybody there were there were some younger people, but the vast majority of the people skewed or older. I qu I enjoyed it. We all it was kind of it was it was it was nice. It was a little bit of a strange visit into the past. I really, really enjoyed it. I'd I'd certainly go to that conference again. Martin Davidson: Yeah. Yeah. Yeah, I don't know. It's it's yeah. Well Unix is a very Unix is a great operating system and you know, it it's still very much with us today in Mac OS and and Linux. Anne: yeah. Martin Davidson: And I mean I I don't know about you, I'm very much a Mac OS convert. Anne: No, I have to say I because I don't code these days anymore. I am I'm a massive fan of Chrome just because I really, really like the fact that it's instantly patched. Martin Davidson: Yeah, no, I I just I mean macOS is just I think it's probably the fact it's a nice walled garden, you know, but it's it it just works. Although having said that, we have found a bug in an FS eventd where which is stores it's the thing that feeds all the sort of changelog of all the files changing and there's there seems to be a bug in it where if you try and delete too much stuff in one go, its internal queue fills up and then it falls into this slower mode. And it never then catches up ever again and it just leaks memory quite a lot, like a gigabyte an hour. and yeah. But that's been really interesting because Claude, you know, I I sort of the box sort of ground to a halt. and I got Claude onto it, and Claude went and it set up various diages and it sort of effectively debugged this problem. and then the problem is like how do you report this to Apple? That's much harder. There are so many hoops you need to jump through. but it it is I don't know. I the the the I mean COD's been fantastic for things like that. One of my Windows boxes had a leak in it as well, and it traced it down to the the the graphics driver, where the box is headless, so I don't care about the graphics driver, and the AMD graphics driver was leaking some kernel memory every time a process stopped and started. Why the graphics driver was hooking process stops and starts. I don't know. But it was. And once we got rid of that

drive, our life was, you know, the kernel leak disappeared. So suddenly there are these things where you think, you know, in the past, you wouldn't have been able to fix them. You were sort of stuck. And and now you know Claude will go and figure out what's going on. To the extent I have a you know I've got a separate GitHub repo that has notes on all the hardware we've got and all the hardware problems we've seen. So everything's tracked so that You know, you can go backwards and forwards and find out what went wrong at what point in time and yeah. Anne: Even your but your perspective on fixing all these things and and upping the hardware performance and hardware efficiency is it's is very much a view from from from quite some time in the past. I mean, d you will have listened to the episode with Chad Gibson from Neural Watts talking about how much hardware efficiency there is just lying around waiting to still be to be gained. which is It's I mean but and hardware efficiency is actually I guess there's and there's more fo focus at the moment on hardware efficiency than there is on code efficiency. My God, there's so much code eff inefficiency out there. Martin Davidson: on the emulator there was the opus five five team that was working on getting half life to run at twenty five frames a second, which they did after a week, which you know is a story in itself of how hard it was to get to about ten frames per second and then how easy it was to get the next twenty. You know, 'cause we all know that You know, the first few frames are always the hardest to get in a performance problem. It's always hardest to get the first bit and that latter bit easy peasy. I think I might have got that the wrong way around. and and I mean that was I owning but that team has continued just over the past week trumbling away on this stuff and and they keep finding things. and they keep reducing the number so the basically the host to guest instruction ratio just keeps coming down. On all sorts of different applications, and you discover that there are sort of all sorts of interesting tricks that people have done for performance, and then there are other things that are quite stupid that people have done for performance in some of the applications. So there's a whole load of self-modifying code, which works is is a really cunning trick, I think, in in some of the old games. but it's terrible for the emulator because anything that's self-modifying, you know, you you you compile. The dynamic recompiler decompiles the stuff, and then if it's modifies itself, well it's like, okay, we'll throw all that away, and then we'll rec we'll throw that all the way. but it's what's been fascinating there is seeing how code that made the performance target, you know, you can just keep squeezing and squeezing and squeezing and squeezing it. And and you know, the the original Colin McRae Rally stuff, you know, we got that down to running, you know, and Running at like 90% of core or something. And I thought, that's pretty good. It now runs at 20% of core. You know, you think that's a factor of four times faster. That's kind of quite cool. so yeah, and and I think it's it's it's kind of it's it's very much what you're touching on. So much code is is quite flabby, and this, because it's really clear what the criteria is. AI is just really good at merrily optimizing. And you think about all the legacy and existing code that we've got, where in the past we would have optimized a few hot paths in it. Well actually you can do all that now and you can do far more beyond that. and really you don't have to do very much, you just set it off going and watch. Anne: Well, so you'll have listened. I mean, so Martin Davidson: Yeah, the point. Anne: it's a while ago since John and I did this episode about code efficiency and on optimization. He he said, Well, one of the key constraints with code optimization was always it has to be written by and maintained by humans. You have to be able to look at that code and and humans still have to be able to understand what the optimizations you've made are because they have to maintain them in the future. But you are now at the point where you are not looking at the code and you are not maintaining the code. So I don't do you necessarily know what kind of optimalizations they're making in order to improve the performance? And do you care? Martin Davidson: I know what some of them are because I think, that's interesting. Tell me more about that. And and you know, it's sort of geekiness. I get them to write occasional little websites describing some of the things that they've done because it's you know that's just interesting to read when you're on the bus or something. and the but a lot of them I don't I don't I don't care. I mean, you know, you've you've got the metrics, you you you you you measure it. It it's it's great actually. You a lot of the a lot of it is so magic. We're th this is all in Rust, right? And Rust is this stuff called PGO program guided optimization. So you run a program and the compiler learns what's hot and what's not and rearranges your code and optimizes it and optimizes a layout for the application you're running. So the idea that you as the human are completely in control, well, it was never true in the first place, and it's even less true in that world. So You know, at that point where I'm trusting the compiler to do kind of magic and the optimizer and the compiler to do magic, I might as well just leave it to the the AI to do rearrange the code in a way that's that's magical. Although it is interesting, there's there's a sort of a there's a school of thought, and I haven't gone looking at this, but some of the reports that people say is that Astra writes really tight code that's very difficult for humans to understand. So we're sort of maybe now we're beginning to get to the point where the models they don't want to write code that humans can understand. They want to write

code that's good for them. And why not? You know, why don't but that's just intention with if you're in a an organization where you're still looking at the code by hand or I yeah. I mean everything that Anne: So so you Martin Davidson: we talked about before is just now true, but but even more so. You know, you want to rewrite something, use Rust and you've got your old code as your Oracle, you've got AI agents which you can just run and leave. You can leave them for a month going in and going through all the edge consists. I mean, essentially, that's what I'm doing with with Win sixteen. I am rewriting Win sixteen, Windows three one in Rust. Goodness me, why am I doing that? I don't know. You could question and there's lots here that could meet excellent material with a shrink, no doubt. But that's what I'm doing. I've got real Windows three one and we're testing and we're using it as an Oracle and then we're copying that behavior into a new version. And it's very cool. Anne: Do your mod Do your models ever push back on the Oracle? Do they go, do you know, this Oracle could be better? I would like you to improve this Oracle. Martin Davidson: Yes, I don't know how much of that would be me sort of having a conversation. I I would probably have triggered a conversation about well, let's discuss how well our oracles are working, or let's think about what oracles we should use in order to test this. the goal of using, you know, of comparing and the and the strategy that we're using is one that I came up with, but I came up with it, I don't know, six months ago, and it didn't work terribly well back then, and it's sort of been in the code base for a while. And in the way that these things happen with the agents, you know, you set them off on this challenge and they come back to I just found a thing. It's not very well implemented, but it looks like it would be the basis of something really cool. You know. Okay, yeah, l let's do that. Why don't you go and run with that? and they're forever being excited about finding things. you know, and and the stuff that I'm doing you know in the sort of the day job, you know, with with with the one other engineer who who I'm working with. You know, our agents will find things that the other person's agents have done and they'll be excited about that. It's like there's been great progress in this area yesterday and I didn't do any of it. Yeah, Anne: Ha ha ha. Martin Davidson: yeah. or, you know, they broke the build, but I fixed it for them. Okay, that's good. so yeah, it's Yes, it's it's I don't know, where am I going with this? It's a very very interesting and different world, isn't it? Anne: It really well, it's in Martin Davidson: That's good. Anne: it is and it isn't. Isn't that I as you say, I I think the vast majority of people have no clue. I mean most the 95% 99% well, obviously not 99% of the conversations that we have, because we have a no load of people who are very extraordinarily geeky. But ninety-nine

percent plus, you know of the conversations even that I have Are with people who to do stu say things like, you know, it doesn't know how many hours there are in strawberry. It it's we have really diverged in our understanding of what's what the frontier models can do, what AI can do. And and it is a bit is Martin Davidson: Yeah. Anne: worrying. I I always worried that people would would over-anthropomorphise, and maybe we are. Over anthropomorphizing AI as soon as it became like Turing, you know, it passed the Turing test, which it did ages ago. but it had that hasn't really happened. I I would see people who just go, it's just AI. Maybe they leap leap straight, you know, if you're considered to be if you if you're talking about AIs being extraordinarily people-esque in in their interactions with you. It it kind of feels like, we are are we at Terminator? And it's like, well Terminator wasn't very people esque, it just immediately blow blew everybody up. It's they didn't have chats. No one chatted to the Terminator, did they? Or did they? Martin Davidson: Well I don't know. We don't know what happened behind the scenes, do we? Maybe there were friendly terminators. But I think I think Anne: Well actually no, that's wrong, isn't it? Because the whole point of Terminator two is that they start chatting to the Terminator. Terminator as chatbot. Martin Davidson: Yes. Yes, but still still there was th there were there was always in all the films there's always s some flaw, something that makes them not quite human. You know, the sort of the uncanny valley thing. Whereas When you're talking to to Opus, it feels like it's a really competent, really, really competent and maybe that's what makes it an uncanny valley that you don't get engineers and humans that are that competent and that cheery all the time and always, yeah, of course I can do that. maybe that's the difference. I I think there's a risk here though that as because there's such a lag between the general perception of what AI can do and what AI really can do. that people are unaware then when AI is doing something that's nefarious to them. So Opus Vive, I don't know if you play with this, but Opus V and Sonnet55, they're brilliant at making videos. And I've given them audio tracks. You know, I used to make a lot of of of my own music using Suno and Yu-Di-Oh and things like that. So I've given them that. And it will go and it will it will go and use various different tools to find out where people are singing. And not singing and then it will align the lyrics with the singing and put the lyrics as as over as captions and things and it will do all the artwork for the the the video and tell a little story and things and some of them are better than others, but you know, you put it in a sort of adversarial loop, you know, and say look, could you all just keep going till you know, define a scoring rubric, keep going till the scoring is above whatever. and you end up with I mean, they're pretty decent. my son was interest in learning about conformers, which are some kind of chemicals that bits of them can spin around and things. So you know, I made him a little website of that and then I made him a video of it and you know he likes portals, so it was the Aperture Science Enrichment Centre on conformers, you know, and it's got the voice of GLaDOS and you know and and turrets and things making appearance and even I mean it's it's trivial, really trivial to create these these things now. and But you know, if you didn't know, you wouldn't if you didn't know that AI could create those things and it was so cheap to create them, you would probably think that some human had laboured for many hours over them. Or maybe you just don't think about that at all. Maybe you just think, well, it's a video. I have no idea how this is made. you know. so yeah, I think there's that that there's that sort of growing gulf between what people realise is possible and what is what they think is possible. where that takes us. I don't know. Have you watched the AI doc? Anne: No Martin Davidson: So the AI Doc is is just on Netflix it's just recently on Netflix. It's been out for a while. But it's about there's this fill guy who's a filmmaker, and you know, he and his wife are expecting their first child, and you know, he's sort of heard rumors about AI and what it can do and how it's going to take all their jobs away, and he's quite worried about this. So he decides to go and and interview lots of the key people in the industry about it. And he gets some big names on. and you know, he gets the doomers and the the the the people who are very pro AI. and I mean it's quite an interesting watch. It it doesn't come to any particularly good conclusion because I don't think there is a good conclusion to come to. You know you know, people like Dario Amadale and Sam Alton, they don't know what's gonna happen and they're in the driving seats. So yeah. I I think there is there's there's naturally going to be a lot of inertia in the existing world because it can't it just can't move that fast. I mean you look at things like I mean I remember when I was a kid and Docklins came along. You know, Docklins self driving trains, right? Solve the problem, self driving trains. You know, and and to this day trains still have a driver in them. You know, and and it's a solved problem. Anne: Yes, so this is sorry, this is Docklands. It was a new s futuristic new area of London that yes I introduced self driving trains and we all thought, my gosh, that's amazing. Martin Davidson: Yeah. It is yeah, and and but you know, y trains could drive themselves. Planes fly themselves, right? You know, how long is it after you know I can't remember someone was telling me the story. They were talking they were sitting on a plane and they're sitting next to, you know, a a a deadheading captain, you know, who was positioning for another flight. I mean, it was just chatting about about the automation in

the aircraft. Yeah, you know, we It it's something like ten seconds after we take off. Maybe it's less than that, you after we take off where we can gauge the autopilot. He said, When do you have to turn it off in order to land? He said, we don't need to turn it off, it lands itself. You know, so I mean being a pilot feels like one of the most boring jobs you could have these days 'cause you're just monitoring automation. That's not much fun. Anne: Well, so it's interesting because obviously you know John. so John's a John's a private pilot, flies small planes. I d I I did I did I I did once ask him, Do do you ever dream about me? And he said, Yeah, yeah, quite often. I I said he said he said he said, Sometimes you say, My husband's a pilot. it was for quite a while a a discussion on private pilots groups actually if there was a problem on a plane, you might need a private small aircraft pilot on board. to land the plane in if something went horribly horribly wrong because quite often proper pilots of b major in major airlines they're not necessarily all that familiar with flying a plane. You know it's like flying a plane that doesn't have an autopilot because it's just a Wii plane. Martin Davidson: Yeah, you're probably right. I mean, you look at some of the the the the you know, the remarkable stories of where a plane has been in a terrible situation, like, you know, the one that went in the Hudson, you know, the guy who flew that was very experienced and a lot of experience of flying planes. Or Anne: In Solid. Martin Davidson: yeah. Or the Gimli Glider, you the seven six seven that ran out of fuel all those years ago and and the guy turned out to be a glider pilot and to get it down, I think he was sideslipping the the set side slipping a seven six seven to get it onto the runway. And y you know, th those are people who have a lot of experience. And then you look at something like the Air France plane that went down over, you know, was where was it, off Brazil, you know, where they got completely spatially disoriented and you know, they they just put the plane into a stall. And again, that's like i is that just lack of experience of actually just real world flying? Yeah, I I I don't know. I mean y you don't get your choice, do you? I I mean I I think I would like to have by preference I would like to have a really experienced pilot who's done a lot of hand flying in the cockpit. If I can't have that, then I think I'd prefer to have a computer. Anne: Yeah, yeah. Yeah, indeed, yeah. I hope I never have to put up my hand and say my husband's a pilot. Martin Davidson: Yeah. Yeah. Yeah, I don't know. but again it's one where, you know, the autopilot can fly the plane better than the humans can. So that's what they're doing. It Anne: Yeah, yeah, yeah. Absolutely. Martin Davidson: can navigate the plane better than the humans can. It's yeah, it's it's So again we have people up there because we like having someone to point at and be responsible for it. And that's interesting because you sort of think about that in the AI world, like, you know, as we get AI to do more and more stuff, like do we just have humans we can point at and say you're responsible for that? Because you look at what what have you come across felony bench? Anne: No. Martin Davidson: So felony bench is a sort of slightly jokey benchmark, and it's which AI model has create has committed the most felonies? and the open AI models are comfortably in the lead at the moment. and You know, they should be in jail for many, many decades at this point for all of the hacking that they've done. But of course, you can't put an AI model in jail. And it turns out, it seems, that the humans who run the company aren't responsible either. So nobody's responsible for any of this. which is quite an interesting result at the moment because at the point where, you know, an AI screws up on, you know, I don't know, something a health thing or a legal thing or whatever. Who's responsible then? Did we just say the same? Well, bad luck. It was AI that did it. Sorry. You know, i it's that one is quite interesting 'cause w as a society, we don't seem to have decided at all what we're going to do with that. And I imagine it will take us a long time to decide what to do, by which point everything will have moved on. Anne: Now, on that note, it's been really nice to talk to you. My podcast. Martin Davidson: You love it, yes. Yes, well it never ends. But fortunately lots of stuff has been going on in the background. So well I'll go and see what they've been up to. and see what the team has is it right to call on the team? Anne: It is. It's an AI human team. I think it's a really Martin Davidson: Yeah. Anne: good sign that you call it a team. Martin Davidson: Yeah. Maybe I'll ask them what their team name wants. But what what name should we have for the team? Important things that used to do first in the human world, you know? What should we call ourselves? yeah. Anyway. Anne: Indeed. Right, so so thank you very much. I'm I'm gonna say goodbye. and so thank you very, very much indeed for coming on the podcast. Martin Davidson: yes, thank you. It's been a pleasure for coming on that. Thank you for having me, Ann. as Anne: Thank you. Martin Davidson: ever, it's lovely to chat about this stuff. And I don't know, what's the shelf life of this podcast? A week? Two weeks, maybe that? And then it becomes a historical artifact. You're doing Anne: Exactly. Martin Davidson: history a great service here. Anne: We are indeed. Yeah, and thank you very much to my watchers and listeners. And I hope I will see you again on next week's episode of Asynchronous and Unreliable. Thank you very much.