Smoot Community Calls - 2026/08/26

Hey Tim and let's wait for a minute. Hello. >> Yeah, I think your sound is kind of a little bit distant. Is that better? >> Yeah, it's better now. Yeah, >> it's wrong microphone. >> Yeah. [laughter] >> How are you? >> Yeah. Uh yeah. Yeah, it's almost the same. No, the same as like two weeks ago. Yeah. Um not much >> still deep in it. >> Uh yeah, it's not resolved yet. Yeah, >> right. >> I won't want to hold you up with with signar. Um there is the the pen about thank you like we added it. It wasn't really my um it come came from Kuni. So thank you for do that. Let me just share. >> Yeah. Yeah. If you can share that. Yeah. I I think they have been uh yeah the the dependent uh bot is there already. Yeah. Uh >> all right. So I went through the 33 that I have here in security. Um yep. So but I working in the development branch. So I went through all of these. They should now be resolved with uh this PR and then Ry I hope I say his name right. Um [clears throat] >> yeah but yes >> I see there's conflicts I will fix um it's because I merged the other one first which was the security file and then you just wanted to add the group dependabot file which was similar then cacti it's also like yeah so I have to resolve the confixs before it can I didn't see there was any um and then like it needs an approval but this is all going into develop and then I starting to think um just to keep things moving maybe it's best if we do a release of sign every month if there is something to release like this uh I don't know how you feel about that we >> well yeah every month is good uh but another thing is that the uh combining scenario with the bridge right the the signing the use case basically that that's something uh we want to explore a little bit later uh yeah monthly at least good even I don't think we need monthly we can we can do it as needed like every two well maybe two months is also okay but I think the the the the question the I think we still need to work on the use cases there and also if we need to do quantum let's do quantum if there's a possibility that be great and then the um um on the bridge side there may be some changes like we have a meeting with uh horsmones lab right so we'll discuss with them as well to see what idea they have how much they can contribute so we we may have some u some changes after we discuss with them uh >> I didn't know they were still alive [laughter] um >> oh yeah they're alive yeah yeah they're still alive yeah definitely >> is is this a new use case I will join tomorrow I just I didn't really know what it was about um so It is for the bridge >> uh well for the whole interability right so it could be the bridge of course the the multi-c and signing security is a big thing and I think security of course is a big thing because the cosmote has security issue just exposed as well with this AI thing more there another one in cosmos as well I can't keep up [laughter] >> yeah yeah so so we we need to discuss these things been think about these about this these two that's just like uh how can we have a open source uh code and still maintain security uh that's they can scan our code and AI is more getting more more powerful uh so it's very difficult to ensure security yeah >> um but we must also like I think it will make for much more secure systems um all right so this is a this is also a recent thing to test that because this the signers there's still a whole list of issues that ai picked up that I haven't fixed yet when I brought it over I'm still making discussion the discussion sits in our our repo and it is >> non trivial things soar is also not not 100% there yet >> okay yeah I know and what with with new models coming up we're going to find more issues. Um, yeah. So, we have a call with Cosmos. Let's see what they think and maybe they'll uh they'll bring some resources and some ideas and then yeah, then we can put them together. >> The DB thing. Sorry. I'm just trying to figure out what happened. Yeah. Oh, put them together. Okay. Yeah, sure. Sure. We'll see what what they say. >> Yeah. Yeah. Yeah. So and and on on our side I think there's no because of we uh we have many things to kind of high priority things right now. So there's no much progress on the bridge itself. Um so uh I just want to let you know that uh yeah it may it may take some time and maybe there will be some changes too. So uh so they may wait let's wait a little bit more time for an update. >> [clears throat] >> Is there anything else to discuss? >> Uh well I yeah I I think we can discuss maybe nine on some other things. Uh but on the bridge side there there's no update right now uh for the smooth smooth >> smooth things has not been going so smooth. >> Yeah. Yeah. Yeah. That is a fair assessment. >> Yeah. Um but we also like trying to get around fable as triggers. So you can get a pretty nonsense report back if you just fire it off at at um Fable you to to reach sort of mythos. You have to be very smart with it. This is my experience now. But it does uh find like a little bit of stuff. And then we are shamefully saying that we are also employing some Chinese models. >> Oh, really? Oh, >> so heavily on on on um on Fable, you can hardly do a security audit, right? >> Yeah. >> Yeah. You know what? when we run with fable uh it it doesn't go anywhere then it fail back to the old model uh so when we scanned it so fab was not doing handle a lot for us we it fell back to the to the old model >> what I thought like when you say it didn't pick up any of those bugs I was thinking >> it didn't it did not pick up it did not pick up yeah >> hit all those triggers because it should Right. Um, it tells me at least when I when I when I hit the triggers. So, and it it was also for me like it goes back to the old model. The old model doesn't go find all these nitty things. Um, so you really wanted to be against um well, DLM was the one 5.2 to that we ran again and it found a lot of things that that was triggered as as a malicious action on on claude. So I don't know if it is worth it for for you also to try other models because I mean if you're so limited in your security scans then you you're again you're at a disadvantage and you paid hard for this now. [clears throat] >> Yeah. Yeah. the uh we uh in fact the uh when we we swable it doesn't uh find it and then it fell back to the old model it did not find it and then we use the uh open AI and then it the later on after after some time you found it uh so so we are looking at at different triggers as well different prompt yeah >> well we run our own model >> like So I work with a lot of other geeks [laughter] that that play with models. So I say a Chinese model, we do host it ourselves, right? So >> Oh, really? Did you Which one do you use? Kim or Deep? You mean Kimmy, right? >> But they load. >> Play again. >> Which one do you use? Do you use Kimmy, right? K I M I. >> Uh, no. They use something else. I'm not so we're getting a a briefing on Friday on what exactly but it was a Chinese model and I think it's the equivalent GL GLM5.2 or something like that. It's quite great with security audits um that they run. It's free. It's open. Download all these models right and run them yourself. So this is where I'm thinking like Claude is is going to I don't know they have to let those triggers down a little bit otherwise they will also lose their their market. I don't know what will happen. I just I don't trust um Fable anymore for a security audit alone. >> Okay. Yeah. >> Not because it's crappy, but because it just it won't uh allow you through. It just keeps triggering it as >> as malicious. >> Okay. Can you uh I think it'd be good if we share the the the the Chinese model you use for security. Um yeah, I thought you were definitely using Kimi, but it's not. Uh so that's maybe something else. >> So I have to go check for you. We don't I don't have my nose in. >> I just I just needed a new model something. >> Yeah. Yeah. Yeah. Because this thing the well one this because it's open model, right? you can you can add your own library to it and and one thing that I in fact um that's something I was maybe this is relevant as well here we we have a security audit company for us and then we find that it's easy to kind of scan the code that's luteine but what's what's not easy was the the test suite right how do you test it and then also the regression right if if you have a new model right and then you scanned it and then how can you have another or you fix something, how can you make sure that those fixes are effective and then you need to run a a kind of a regression test to make sure things go through. Can the security audit platform provide you a a swift a kind of collection of test cases for you and you just run it. Okay, that's something I'm looking for. It's a generally a a suite of uh test cases automatically and run through them. Uh those will be useful or I haven't seen but you can scratch there so many. It is GLM5.2. Um we let me just put it in the chat here. That is also Yeah, it is a Chinese GM 5.2 that we Okay. It's now free, right? They uh open sourced it in June already. So that we ran that and it was um it found a lot of things that was blocked. What I I would I don't say Fable didn't find it. fable uh launched off or it's a current model in my claw AI that fires off fables and um I've told it to like do jump through a lot of hoops and it doesn't trigger these things but it it still comes back and it says the seventh threads has triggered um and then we switched over to JLM 5.2 doing and it found like most of the stuff and it's just like oh sigh you know um and then we can't we can't um run we use their their models it's Chinese models [laughter] so uh we going we host it on a on a m on a big big ass machine of one of our colleagues uh that's running it so we'll see if that I don't know if that's it cost a lot of money to run it constantly your own thing >> yeah well it says it's free but when you run the machine yourself, right? That's cost a lot of money probably. >> I'm not going to trust them. Not that I trust anthropic either, but [laughter] you're on your own. The Chinese will go put their nose in your data. No doubt. [laughter] I don't know. Maybe not. Um yeah, that was that was another thing for us. But otherwise, um lots of AI. Everything is AI. >> Yeah. Yeah. Well, you are the one criticizing me for using AI. What? >> Oh, I didn't criticize you as like I don't want to make Oh, I don't want to see that it was AI cuz that [laughter] is I still I can pick it up so easily when somebody has just like thrown AI at it. >> Yeah. >> Well, it's inevit Yeah, it's inevitable now. It's just like Yeah, it's just it's everywhere and it's more powerful than me. I find a lot of things I cannot find myself. >> No, no, it is you will stay behind so far if you don't plan on it. >> But it's about using it correctly and knowing its limitations like not not trusted to do a security audit which when you know it is being blocked in all sides. You cannot trust that anymore. [clears throat] So my my trust level just dropped quite a bit. >> But it's very useful in so many ways. Oh, like I say, for me it's almost like what's happening to our jobs. Um because for now they still need supervision, but that's that's fading. You'll always have a job because you're teaching, right? >> Yeah. Well, yeah. I think I think it's good for people who have experience and have a lot of experience. I think it's the in terms of job is the the people who are doing the implementation okay will be impacted. Uh because I think the people who have a lot of experience that that's I think that's still okay. Um >> what we doing what are we doing we supervising right at the moment because it's not quite there to just let it run on its own but if you have to tell your child we like what to go study you're not going to tell them to go study coding. >> Yeah. So it is a bit of a and it came they warned us for so long and then suddenly it was here but it is even like for me looking at different areas of where where's what's the next thing because um we're building things the moment but like I'm not happy supervising AI my job changed significantly so like where do you look and what do you look for and what's new it's a whole new world for me to be honest and it really kicked in this year. Yeah, I think putting uh one is super supervision of supervising AI, one is to because um uh AI make a lot of mistakes also right uh AI make a lot of mistake and to be able to you should be able to spot these uh these errors. It's like it's making becoming more more difficult because uh the uh you need to find something like they were saying that they they can build a ship for you but they will miss a a knob here and there and you should be able you should be the one who define a process to find these potential issues with AI because AI make 99.9% right but there's a 0.1% that could go wrong and I found that myself and every time when I find that I tell the AI this is wrong and they'll say hey you are right and and and and then it's I would say such a simple mistake why can why do we make that mistake because some AI doesn't have that common sense um that's what I find yeah and that's still need human yeah >> to to write error correcting for AI will be difficult but sure >> everyone one is moving away from LLMs or the the talk is moving away from LLMs. The academia is more moving towards multi- aent systems. Uh this is also a good thing that I that I listen to. Let me try and find it. Um lun this is also a good watch where he Yeah, where he moves says it moves or he bets a bit. he bet 1 billion against LLMs um to be more and it makes a lot of sense to me like that multiple agents can do work together um because context is expensive to keep >> so you can also go in the AI direction in that direction and and try to do something there I guess but it it's taking over things very fast I don't know what's going to happen. >> Yeah. What what I heard is that what what I know is that the engineers be become prompt engineers. They they don't write no code. It's a prompt AI to write the code for >> at the moment. >> Well, that's what we are. Yeah. Yeah. >> I'm not writing code anymore. This is ended last year. Um we're all supervising our prompt engineer. I would rather call myself a supervisor at the moment, but it's still degrading and I'm not happy. [laughter] So, I have to look at something in a different direction. >> You kind of like it feels you're okay when you work with people. >> So, you kind of will be all right there for a while also in a rewarding sense. >> I have a question for you. Yeah, about that. I have a question for you, Marit. the I have not written any AI code for smooth project. Okay. Are you okay if I start writing code for smooth project with AI? I'm okay with it as long as you or I will review it for you. And I'm also going to use AI, >> but um >> you have to be just like thorough. Like we do four, five steps when we run through like reviews, different people firing their AIS at it >> because um your AI learns a little bit with Claude in your personal file and how to work with you. like mine is starting to sound more and more like me [laughter] as I corrected it over time like how I want things to be to to be um laid out. Um even my wording my way of using wording it learns. So um we just have to be very careful because like you say it makes a lot of mistakes. >> Yeah. Yeah. Because but remember last time when I used AI to convert programming language from this uh like uh language agnostic format to to specific format then I got criticism from you saying that which >> it it was a long time ago. I don't think it was done well if I should probably have said that and not attack the AI. >> Okay. Sorry. >> Okay. So, I think that make it easier because uh I think we can do a lot of things with AI because I I have good experience with AI now. I know you did pretty good job. Uh but I do always do manual. I don't do the automatic because you can have a auto mode and then you have the manual mode. Auto mode is just the they do things for you. And one thing I found was that uh there were times it you you said you trust this directory right you trust this directory means that they can only access this directory and temp directory but there are times when this guy the claw actually went out of your directory to find other directories and I have to stop it saying that hey you are only allowed to look at this directory not the parent directory and then say yeah you are right >> [laughter] >> um isn't if you if you listen carefully how the hugging face um >> work they created a whole message board on artifactory to talk to each other >> to get [laughter] >> it is absolutely >> ridiculous um I'm trying there's such a good link for this as well um >> I don't know where it random. Oh yeah. Uh listen to this. It's not that long. Um how Mythos used social engineering to um to get a vulnerability into code. >> Okay. >> It's absolutely shocking. And another thing is that they um they also like AI they say okay uh I want to open the browser to test this issue do you allow that I don't allow AI to open the browser for me okay or >> token which is very limited on what it can do on GitHub um so it's limited with a token on my on what it can do there um But what it reads, you know, it it is not it's not safe. They they can see what it reads. >> Um >> yeah, >> and and my own directories. No, you have to be careful. And I gave it access to my work directories, but I mean everything in there is work. But it will jump. It will 100% jump. >> Yeah. Yeah. >> You have to use it in a virtual machine. That was my conclusion. How to make this safe? I box it off in a virtual machine. That's the only way. >> Oh, I use a a dedicated machine. Uh it's not a virtual, but it's a it's just it's a server just for for programming. So, so out, right? So, you're okay. >> Yeah. Yeah. Yeah. Uh but I think they are a couple of things are still ending. So, I'm glad that we Okay. we we we can use AI to uh to do the programming more smooth that that that opens some up some opportunity and there's still something there. Yeah. >> If we don't start coding with AI and um if you have ideas because that's where everyone just take your ideas and throw it at the AI and see what comes up. We will janitor it into a good thing. But it's really now about people that still have good ideas. Not not so much of how we built. That's the supervisors now over AI. But yeah, put your ideas in there. >> Yeah. Yeah. >> Build out the prototype. So easy for us now. Like if I think of how long it took for a bank just to to get a prototype up and have screenshots and a little demo for them. Now it's quick quick quick quick. >> Yeah. Very very quick. Very very easy. And and so but I still limit AI for several things. First thing what the cloud have been asking me can I commit this code for you? No I don't allow them to commit the code because once >> token >> yeah they can we have a like a SS token right you can give AI the access token and they can commit the code to the repository. So that's one thing though I still don't >> manually um I ask it to do manually for me. there's no way to enforce it other than the token. It will still sometimes, like you say, just do things and slack, but this is in your rules. You're not allowed to do this like, oh yes, sorry, that was on blah blah blah blah. Um, but if you give it a token, it can only read it. >> No, they were right. Right. Because they want to commit for you, then you have to give them the right permission. And I feel I still feel like >> a separate token with only read access and you this is what I mean with you force it with a token. >> Oh, okay. Okay. So, don't let them commit. Just let them read. It's okay. Don't let them commit. >> It was still like committing on a PR. I'm like, what? It's in your rules that you're not allowed to do anything on GitHub and it still went to go do stuff. Then I just made a token that actually it my company had had to make the token. It was strange. Um, with Claude settings, I couldn't use a personal token. It had to be a companywide token, organizationwide token. Strange, but anyway. >> Are you talking about GitHub or GitHub, right? >> Okay. >> Okay. There is a there's a there's a classic token which is accessing the whole repo and there's another one called granio token. Yeah. grain token that has the organization uh uh kind of priority and the security level and then you can rewrite to certain repo. So that will be the one for >> this one that you have to use your organization that owns the repost must just it must be in their name. >> Yeah. Yeah. Yeah. But but I so far I have no sometime they say commit right it's always commit to a particular branch. I do not see the limitation on PR. For example, if I if I allow uh cloud to commit uh I would like it to be a PR commit a PR rather than to commit uh directly to the main branch. So, but I still don't know how to do that yet. So I think maybe we should set the rule saying that if we use AI to commit they can AI can only commit as a PR and we as human need to approve the PR that would be good% >> like you want to go full auto already like let it write code for you first and let you manually commit and push um otherwise it's going to run out straight underneath you if you let it make PRs and like next thing you push into all kinds of branches and stuff Um, y I wouldn't let it go free like that. >> I would say I mean I do ask it sometimes to make a PR locally. >> Um, and like you can ask it anything and it has you can give it um rules in in in its personal files. It's okay. I just you can't rely on him always following the rules. >> Okay. I think we could uh tomorrow when we have meeting I have been thinking about this. is about Cosmos relationship, right? Of course I I know Cosmos code a little bit uh the IP it's called IPL or uh they they they have interbidity for a long time as well and I think the only way so far to consolidate the ideas together is to to leverage AI which means that okay we have our um specifications we have smart contract on both sides and we have the the layer and then and then they have their own thing as Well, right. They have connectors and the things we can just agree on some high level like almost we work on the prompt and that let the AI write the code for us together and this way we can probably build something that's common >> trying to integrate with cosmos and no one really made a success out of it. So the only hope is AI. [laughter] It was so complex in general. Nobody um really managed to do interrupt with they had their thing and they wanted to make their thing a set standard and anyone had to do everyone had to do what they wanted to do. Um so so very little interoperability happened there. And like yeah like I say if you throw the AI see what comes out. I just I think it's going to get very complex very quickly >> in I know I know that's why I would think well initi without AI I would not like to touch I because I touch it before yeah I don't like to touch again I but AI can help us with this >> let's see what's a mess you get [laughter] >> yeah yeah >> we can see right like interesting interesting um use case see what it comes up with maybe it gets a really clean solution which I don't >> yeah but we can kind of compromise on certain things right and yeah uh >> okay okay >> based on their email they're just looking to contribute IBC as an LFDT project um but because Smoot already exists and there's conceptual overlap there's this call so I think their their goals are going to be pretty narrow I think. >> All right. >> Yeah. But IBC had been tried many times right already. Uh in fact the the cacti project was kind of I think it stem from IBC originally. Uh so I yeah I I think they they should have some flexibility but let's talk to them together. And I think now that we talk about AI, we can if we can have AI implementation and then just put the high level thing, agree on some high high level things and then maybe there can be something there could be something good out of it. We can only see >> it looks like Linux Foundation is picking up all the strays [laughter] and Cosmos is coming in into the under umbrella as well. >> Yeah. Uh, one other kind of Smoot related thing is I'm [clears throat] not sure to what extent this is present in the current codebase, but the original pitch for Smooch was that it would be compliant with the EA standard. Um and it seems there's some developments on that side in terms of the interro working group and its future um as well as new um Ethereum backed alternatives to the EA, let's call them. So that might also be something that needs to be considered for the SMO project because if there's going to be no further work on the EA standards for interop then you know we should separate SMO from that. >> Yeah. uh that I think uh Red Wang uh has been communicating with me uh then uh he is trying to expand the uh EA's inter work group maybe even to rebrand it and then he's trying to he's asking people who are interested uh but I haven't got the followup from him yet who who actually are interested uh but no matter what they >> you sent a uh you sent like a public message in the in the group and uh with that with that message like anyone who's interested to contribute uh but zero people have responded [laughter] >> well that's publicly right so I'm not I think he yeah he he he said he's approaching people he's looking for people who are interested in uh so yeah and yeah that that that is I think if there are people interested there could be some new things over there and then we can also if if there are no other people interested did if we have some ideas out of it and we can also bring back to ES back as well. Yeah, I think there are certain things like uh originally we we hope that the uh zero knowledge prove is there uh and but I think without AI is very difficult for us to implement things like that. Uh so yeah let's see if we agree that okay we can use AI to code it and then yeah that open up some opportunity as well always bring in lots of time to get it to a production level because it can it can write a lot of crappy code that needs to be janitor. So very quick for proof of concept that the right production grade code it um yeah I wouldn't leave it on its own. Yeah. Yeah. Yeah. Well, that's one good thing about open source, right? Open source, we don't need to bring it to production immediately, right? I think it's the we just need to make sure that this thing has traction, there will be people who are interested and then we can if there are people interested, then we can spend more time to kind of bring it from PC to to production. Um, >> right? Because I personally wrote something for my own personal thingies and like it's good for me to use but I will never put that in production. I can hand it to a group of people and say like you make take it to production but not use like that. >> Yeah. Yeah. Well then that's the one good thing about open source you know it's we liability we we don't take liability directly. Uh so so that's that's there's a buffer on that and also I'm thinking about all these security things is what we spend a lot of time a lot of time on security already but still things happen right so one thing I was thinking about security is that well we should not put the smart contract to be in charge of too much asset the the the risk should be distributed it should not be that one smart contract take uh a lot of money can lock a lot of money on only because we can have multiple smart contract multiple instance. >> You try to avoid your honeypot contract which is your escros your bridges that holds cash and things like that. It's something you want to avoid in general if you do a full full on security audit on things like honeypot contracts is no good >> actually good. We want the cryptographic proof. Um >> yeah security point of view. >> Yeah. So yeah many know because this open when the source is open they're always easy to find security holes. Uh >> if you have somebody that looks at it otherwise it also just lies there. I wouldn't trust the public to find my holes either. It's bad rep. >> Yeah. Yeah. Well, there were times that uh people pointed out that there was some security issues. Uh those are good people. They they actually found it and then they tell you uh so but but in general you you want to have audit audited team but even audit the teams don't find issues sometime they don't find issues like critical issues. Um, >> oh, I hope things will become very super secure with with um with the emergence of AI finding because like I mean it puts any security order to shame the things it finds. >> So like I I believe we're going to have a very the world of the hacker has changed, right? >> Yeah. >> And it's not going to last for too long. Like it's that now a big window that opens, you know, like when when the internet just started like for a hacker, I was one of them. it is a public library. You can just get in anywhere and until people became security aware and things changed. So I think it's going to make for very secure systems in the end of the day. Um because everyone has access to make their stuff secure. >> Yeah. Yeah, the another thing is that the the AI can have some improvement as well. Just like uh when they find SEC high security issues, they only disclose to people who have the right or who who at least there should be a promise saying that they are not going to leverage this these vulnerabilities uh violating the the law. So at least there should be something like that. Let disc that was just merged into Signara says that like if you find a which is borrowed from the Linux Foundation um a security thing is if you find anything not to make it public but to contact us on private channels so that that is sort of protocol. Oh yeah, there's a security tab. Uh they say I I there's a security MD. I think I saw that file just like there's a private channel to report the security issue. Yeah. >> Okay. >> Okay. I think Yeah. >> Let me see see what Cosmos wants. >> Yeah. Okay. I think uh good that we some uh consensus about using AI for the project. I think that's that's good. I feel that we can do more things now uh using AI. Uh >> yeah, like you you can't not like you wouldn't survive in a world um if you try to to go without it. >> Yeah. >> Just be careful. Okay. >> Um, yeah, there's nothing else from my side that I can think about at the moment. >> Yeah, >> we would like a release this month. Um, because not just a dependable things. There's also the type one transactions in the signaring. >> So, we're also trying to build something also with AI but um on a privacy side. So we're still working actively on on signing. >> Yeah. And the on the scenario you uh we want to publish a kind of blog, right? Uh do we do we want to have that uh ready sometime or you what we can do this later but do you want to drop uh >> uh we should have done it in the last release. I will when I make this one we can we can do a little blog even if it's just highlighting of explaining a little bit what was done in the change log because that big one that I brought over with all the little fixes in it >> they don't need to know about all the fixed in there and there's still a lot of things that needs to be fixed >> I need to >> last time you you want you bi want to write the first draft do you still want to do that or well I can help with writing some of that as well. >> You can start if you want. Otherwise, when I make the new release, I'll I'll have to to give you something to review um for a blog. >> Okay. >> But it's good. It makes it look like it's alive, right? >> Yeah. Yeah. Yeah. Yeah. And also people want to know about it as well because they this year they they I was asked question about uh scenario as well. Yeah. I think we should have a article ready. We may not need to publish right now but we can have it ready. Uh so you I yeah either you or me uh I can yeah once I write it I'll let you know and then uh you you write and let me know if we have time. Yeah. >> Okay. Okay. Yeah. Okay. Tim, do you have any anything else to add or we can close this call that for today? >> Nothing else. >> Yeah. Okay man. It's nice talking to you and then uh we can we can sync up. I I'll try to write the the blog at least for the first version. Yeah. >> Thank you. >> Okay. Okay. Thank you. Okay. >> Chat tomorrow. Okay. Okay. Chat. Bye. Bye. Bye.