ThursdAI Nov 5 - GPT-6 Cyber Finds 948 Linux Exploits, AI Edits AI's Weights, Kimi K3.1 & more

0:00 Will will the Chinese models slow down? >> I think the the skis and the rest of the boat are made in China except maybe the the engine. >> No gradients, no training run, no weights and biases. Astro changed 0.016% of the weights, about 131 numbers, and the math score 500 went from 7% to 48%. And that's always been the case, but it's just become so acute because we've entered a world where a 100,000 times more software is going to become disposable.

0:42 >> Welcome to Thursday, November 5th. This is Alex. I'm going to joined by Wol from and Peter Gosv. and other friends are going to join later in the show. welcome to Thursday. My name is Alex Bulov. I'm AI evangelist with Core Weeave. as you can hear a little bit getting over some stuff, but we're going to get through this because we have we had quite a week. I don't know if only AI is related this week. There's a few other news happening on the timelines. It's going to be very interesting to talk with my friend from Germany and my friend that lives in London about the the stuff that's happening in the US and whether or not it took over your timelines as much as it did mine. but also we have a bunch of AI stuff specifically in the cold open the Astra reading Quinn Weights directly. I think that that's something for us to definitely chat about that's been quite crazy. but meanwhile I'll just say hi Wolf from welcome Peter w I think that very exciting week but not a huge week in terms of releases. Let's just like open this up right like we don't have huge new models to talk about this week at least not yet. There's quite a lot of open source stuff that we should talk about. Kimmy finally dropped their weights. Kimmy 3.1.

1:58 we have a few others from a few labs. So definitely something to talk about there. another reflection, not to be confused with recursion. There's a few confusing labs, but still we will welcome any open source open weights labs to the arena to try and compete. So there's quite a few stuff from that perspective, but we also have two interviews today. One is a longtime fan of the show, friend of the show, John Davis Wolf. Thank you for bringing it up. John has cooked something super cool. He's going to be here to tell us about this in about an hour or so.

2:33 >> Breaking news on the show. >> Breaking Oh yeah, he he's bringing breaking news. That's great. and then I have been trying to get Max for a while. Max Kennet Alexander who is now going viral with his talk from AI engineer in New York is going to come here. Max is a principal engineer at Capital One, has been at Google for a long long time, been in charge of code quality and a bunch of other stuff.

2:56 Max is a fascinating dude to talk to and I'm very very excited for that conversation. Please stick around for that as we as we kind of like progress through the show. Let's start with highlights. I think here we go. So let's start with highlights of the week. What was the most interesting stuff? but also I will say we are not a political show. We try not to touch politics too much. I feel like this week it is inevitable because the politics touch us. The changes, the updates, the midterm elections in the United States. If you haven't if you've been living under a rock, they have just concluded with with a few things that are important for AI. So, we definitely should at least mention this. and I think I'll start this with a prompt to all of you and then see how we how we go from there. I see this is already ready for the hot takes. Is AI over? Is it all over? Are we done?

3:54 Let's start with that prompt. >> No, >> I mean AI won't be over. I have talked to protesters about this in front of AI engineer for instance there the pause AI guys. I told them what do you think will happen if they they they pause. AI is just the genie is out of the bottle. The technology is too important too powerful to just pause. So it's not about pausing AI or AI being over. It's about access who will have access to AI.

4:21 So if the pause movement wins, it just goes underground. Some secret labs will be doing it for the governments only and only the rich and powerful will profit from this. But if AI is widespread and available to everybody on the same terms, then it is much more that is the only safety we have safety in numbers in a way. If you have strong AI to protect you otherwise only the elites will have AI that they already have the super intelligence, they have the lawyers, the doctors, they have all of this because they have the money. AI makes it possible for a little sum. I wouldn't say the free models are on par, but AI makes it possible for you to have the same or at least empowerment in a way. And if this is just regulated out of the public view, it's not going away.

5:05 It's just going into the hands of only the few. And that would be the worst outcome, I'm sure. >> All right, more more takes. But Peter, what's your what's your take on this? How how are you viewing this latest update in the world of AI? Have your timelines updated? It was interesting to hear to the semi analysis team who were tracking the data centers and how the different moratoriums could be affecting things and so on. So and it is quite interesting. So what what they take is as far as I could make out is that for 2027 the capacity shouldn't be affected because there's quite a lot that's already imp.

5:46 So unless they're going to go and say, you know, this permit that you had is no longer valid, which I I don't think is re realistic. So 27 should be fine. 2028 is much more at risk because a lot of permits being are being in progress. But there are things like for example Texas is a big state where it's like quarter of capacity is being planned in Texas. So, and the new governor, I think the conditions are not that, oh, you can't build them anymore, but there are certain like preconditions. So, I think that that's the thing now is I think it's up to the industry to negotiate the the right settlement, which to be honest is probably a good thing anyway. I think it's it's unreasonable to be like, well, we're going to come in, build all of these things, make lots of money, and you get nothing. So, I I don't see why they couldn't just pay off the local population. I don't know whatever build schools you know the legal bribes and give make people happy you know why not I don't think it's even going to increase the cost that much and I think it's a good thing that people are happy when the projects that come into your society not kind of force them over the heads through some corrupt schemes or something or just kind of go over the top of them so yeah it could be could so no risk in the very short term but I think it's maybe a good thing also forces the indust to be a little bit more considerate. So maybe maybe not a bad thing.

7:12 >> Just a brief TLDDR for folks who not have you know been following the US midterms. The the Democratic left on the US took over by a significant margin across the House and the Senate. including Texas, which was predominantly a Republican state for a long time. This is what I'm showing right now is the new governor's takes the new Texas governor take on AI. So Texas now has a new governor and a new senator as well, both Democrats. This is Governor Hinosa and in her Texas first data center priorities plan.

7:45 It's actually not that restrictive as many people are thinking, oh, it's over. some stuff, you know, are very specific like lowering costs for Texans, they make sense. And then Texans seat at the table to decide on how data centers count to the communities. That's always welcome. union jobs is a little bit of okay the Democrats they're going to recuss unions and but making data centers pay for themselves I think that's like a very standard thing that many folks who are building in communities are looking at so there's a few there there's a few kind of like things about is AI over obviously it's an exaggeration one of them is will data center build out halt which will potentially slow progress innovation etc. from the other side, there's all of the pause AI and pause moratorium stuff like that. which there's a few candidates that are into that. Would love to hear thoughts on how you guys see this, whether or not that's going to happen. Is AI pausing? would love to hear from all of you. Welcome back, Campell.

8:52 >> Yeah, I think it's a bit too early to fully tell what the impacts of the last few days of political events will have. but yeah, there's been several of the of the candidates that have been voted in like clearly making statements of hey, we should pause AI. Hey, we should pause super intelligence, but some just straight up saying, hey, let's pause AI development overall. And I think ironically I even if this does successfully end up like fully pausing AI development, which I think is unlikely, but I think even just the statements of them saying they want to do this might ironically accelerate AI development over the next few months because if all the labs are like, oh, like with sometime within the next 12 months, they might pause AI development or force us to pause. So, we're incentivized to scale up as much as we possibly can over the next few months.

9:44 >> That's really funny. Yeah. I do want to address one specific thing that's on my timeline. Michigan, I believe Abdul say is the guy from Michigan literally said the water skier metaphor. The water skier metaphor is this. The United States is the speedboat and it's like going very very fast and the Chinese AI models are the water skier that's behind the boat. So the faster the boat accelerates, the faster the skier accelerates because they copy everything which which literally to anybody who follows whatever we follow as much and literally everybody who reads any new DeepSick paper and gets astonished by kind of the the result there sees that this is an absolutely just like a ridiculous take. I would love to hear you guys like position on whether or not China will also slow down if we pause behind the United States. Let's talk about this. Will will the Chinese models slow down? I know I'm laughing, so this is a loaded question, but like will the Chinese models slow down if the United States pauses AI innovation? Non, go first, please. Daniel, >> I think the the skis and the rest of the boat are made in China, except maybe the the engine. I have you guys tried renting an H100 lately or tried doing work without AI? I had to to set up sandboxes and do a bunch of security stuff and only using local deepseek and I could not run an agent. That was that was not fun versus just putting in cloud code and telling it to do stuff. Yeah. I don't know what are they going to pause tell some altman how much RAM I have at home or >> Yam is pausing the you the innovation in the west will like some candidates suggest also pause it in the east >> how do I even respond of course like like of course not man they train on and you domestically made chips at this point. Come on. And no, no one's going to stop anything.

11:54 >> Yeah. It's it's not all distillation is is basically what we're saying. >> for sure. >> So there's a lot of concerns about, you know, the the the the elected folks, but also I think here's my basic take. would love also to hear from you guys as well about this is that we know no we try to bring you on Thursday II a positivity about AI because we know it could be scary it was scary for software engineers when we started it's going to take over their jobs it's scary for regular people now because well it could take over their jobs that's the fear at least politics is all about you know hope and fear hope and fear we will sell you hope and we will scare you with fear to get elected and we'll sell you hope that we will make it better this is what basically all politics around the world is is is at AI is the easiest to like to rattle the fear around like AI is the easiest like oh it's going to take over the job and also it destroys the environment it's the easiest so there's there could be a lot of stuff the ramped up towards the midterm elections in the US that just populist like takes on let's just you know we will help you solve this like thing and then not a lot of it can go into effect it's still there's still Congress there's still people voting there's still like a bunch of stuff The other thing is would love to also bring this to the show before we start with the LTR. If anybody read the commending heights which is a new kind of book essay that's sort of a continuation of what Leopold Ashen winner wrote in >> situational awareness.

13:22 >> Yeah, situational awareness. Thank you so much. So, so there was this essay by Leopold situational awareness. We all read it and it's like it outlined how AI will evolve etc. And now there's a new one from Will Manis called commending heights. basically I I will send the link in the show notes. I strongly recommend everybody to listen to read because it outlines a potential future for AI. The main thesis as far as I got it from there is AI is already taken part of the economy whereas previously was like chipping away at some of the economy. Now basically it's embedded everywhere and everybody's pension everybody's the the the machine will need to be fed regardless of who's at the hell is basically the the outline.

14:03 However scary that sounds, basic the folks are saying basically AI was the technical part in San Francisco. Now it moved to the halls of New York and the stock market. Anybody read that? Did you go ahead? >> I didn't fully read it, but I did skim it and plan on reading it later today. But overall, the idea of it, there's like this metaphor of it's going to be created in San Francisco and it's like a like a lion's going to be dragged over to New York and put in a cage and and this is like an old tweet from Will that he he wrote wrote a while ago and I guess he expanded into this article.

14:40 but yeah, I think there's going to be a lot of interesting effects on the economy. But one of the things I've been thinking about lately is if you look at the projected amount of compute we'd have for 2027, 2028, 2029 and measure kind of how many instances of the models you could run on that compute. Like I think a lot of things might be more compute limited in terms of like the parallelization you can do than people expect. like I estimate the frontier models at most you might be able to run like 100 million instances of them >> in you know like around 2028 which sounds like a lot but that's not even like that's not even 10% of the working population of the world >> right and so like then naturally most of that demand is probably going to concentrate towards the most high value work of things like biotech material science and maybe not as much job impact while all that innovation is happening as some people might expect.

15:40 >> Yep. okay. So, given that, and I know we have the TLTR to get to, but given there's not tons of releases this week, I I do want to like focus on the fact that a lot of folks who tune in to Thursday, I just love our banter. I would love to hear from you guys maybe one two sentences on what still excites you about the what's still positive in your mind what still is happening that is that will you know despite whatever's doom and gloom on their timelines doom and gloom on their portfolios whatever what still what still makes sense what still makes you smile what still makes you like get up every day and go hog at it and get excited about the future well from you seem ready to go first so as always It's getting better and it's still fun to use and I wouldn't want to be without it. I couldn't be without it anymore. I live with my agent basically.

16:32 We are a diet. We work together in all things. so this works. I just want to add one more thing quickly about what Peter said that we outside of the United States for instance, we have no direct influence on the politics there. But you and I, Alex, as AI evangelists, I think we have an impact in just what we are doing right now, talking about this, showing people this on the show in private life. And I think everybody who sees these misconceptions in their real life outside where you can talk to the people online. It's often host hostility and people are single-mindedly and I don't even know if it's people or if it's bots used by whoever wants to control the discussion. We've talked about this before. So if you meet the people, talk to the people. I talk to them on the train, on the plane, everywhere. Protesters even talk to them because if you think it through and then you can really show them and they can understand what's behind all this and that it's not black and white. AI isn't just white that it can be abused, surveillance and so on. There is a lot of negativity that is real and we have to make sure it's not abused that way.

17:33 But just trying to pause it, it's just not going to work. We have to find better ways to do this. I would love to hear Peter what excites you about AI at this point in time at the end of 2026. Yeah, for me the biggest thing is I just I really love discovering new things that I can do and that's why I really care about the frontier of the model capabilities is because a lot of the time I and I get the point right and I've I went to a conference hosted by a bank earlier this week and they're like oh yeah we're like fine-tuning small mal models and we're optimizing here which you know totally makes sense but it's It's not that exciting. What's what's really exciting is, "Oh crap, I had no idea I could do this." That's that's and I get this all the time and especially when we get new models and some people discover it and now we can do this, now I can do that.

18:29 >> Like for example, I'm I'm just waiting for the models to be able to do some kind of animation, maybe 3D animation through code, right? Not through video generation. I feel like it's almost there and you can almost do it. It's not quite there but then suddenly if that opens up then you can do so much more and we had with Opus we can create HTML style HTML based explanations of some concepts that that was a new thing and I think that could be for education super super useful. So there are few things like that that just keep emerging that we keep discovering and I feel like it's s such a unlimited amount of information we can discover. But I also want to sympathize a little bit more and I think we we should be a bit more careful with just and I see people dunking like on Twitter on on some people who maybe from the outside of their world. I think there was one p crying that she's tired of slope and she was an artist and so people are kind of giving her a lot of hate. But in reality if you if you're not coding if you don't really know the latest capabilities what do you see? You go on Instagram, it's full of like slope crap videos that are all generated, just complete trash.

19:39 You read the comments, the all AI generated trash. You go to work and a lot of your colleagues are copy pasting stuff from code into Slack like that. That is kind of crappy and it sucks and I if that's all you're seeing then you're like well and and they're saying it's going to kill us all as the data centers who do this and that like yeah totally right you get it. So yeah, and it's I I don't know what the answer is.

20:04 I think it just it's very unfortunate, but I think that's also something just to bear in mind is that most people don't see what we see and a lot of the people's experience is just complete sea of slope, which I also hate. So yeah, it's a it's a kind of it's a weird world now. So hopefully we'll get past it. >> We'll definitely get past it. LG go next and then >> yeah, physics, biology, and math breakthroughs. I think all of that's just super exciting. but something more specifically that I think is apparent recently is the rise of verified software that I think will become more prevalent. We're seeing these massive lean projects that are being formally verified, but also what's difficult but still possible and then starting to be soon I think is really just actual infrastructure and software that's commonly used like starting to formally verify those things and maximize the security and safety of the the world of software that's used every day. So on that topic, the guest at the end of our show is going to be an incredible guest for you LDJ specifically to listen to and and welcome to join and ask questions as well. Max Kat Alexander is going to join us. His talk at AI engineer was literally about disposable software versus verifiable software. It's going to be very exciting to to chat about he's he's done code quality at Google and now is a distinguished engineer or a principal engineer at capital one. Non, what excites you about this world of AI?

21:29 give folks some positivity, folks who want to tune in to the show and listen. >> I think public opinion has improved quite a lot lately, mostly because of all the software decompilations and the games. People forget there are 200 million gamers on Steam. They all have GPUs and that's a big and they're mostly younger people who are gonna be likely going to work places or play while at work. so yeah, I I've seen I've seen opin opinion change quite a bit. Also, the default free models got a lot better recently. Before they were just kind of useless or they would make mistakes and stuff. Now I would say even free Chad GBD is pretty good for healthcare stuff. Like as much as I do not like the taste of Chad GBD products, but yeah, I I I think it's changing.

22:28 People are starting to see some benefit and like I I don't know any doctor now that doesn't use an LLM. You have to also think about that part too. >> folks are modifying their games. A lot of folks are seeing benefits more than just asking questions from a better Google. What excites me is that a lot of people have not yet gotten the whoa this thing can do that thing that Peter talked about and Wolfin talked about. A lot of people in the world are now realizing that they can save themselves time. They can improve their health. They can improve whatever.

23:06 They can leave the job and go do a small business for example in United States. A lot of people realizing they can do more. They can dream more. They can achieve significantly more while also taking care of the stuff that they need to take care of by getting superpowered with artificial intelligence. Whereas in the when they see on the timelines the the doom loops and the scare and the artist community saying, "Oh yeah, it's all slap and it all like stole whatever." they may have some thoughts about this. they see the the misinformation about resource use etc. And then many people still discover the use cases like we all did because we're engineers and we like suddenly like immediately got superpowers all of us like three years ago. Many people are still still in the process of discovery. Like Yam said we're still very early to the world waking up to what AI can do for them.

24:01 And like Nissan said medicine every doctor now uses an LM to some extent at least to help them to not have to remember everything the person said or transcribe it in in a safe way. for example, there's a lot of use cases that the world is waking up to that's building the AI is as in superpowers into making everybody's life better. And that's what still excites me to this day. And I think that's they're just going to continue and still continues. And even if we stop, which we won't, AI is not pausing. There's no there's no pausing this. Open source is on the rise. We're going to talk about a bunch of open source models today. But even if we stop the current level of capability is so much an open field of discovery and use cases and what we can do with with the current existing level of capability then there's a lot to to to to benefit from and enjoy and this is why we bring you the show so that you also will kind of dream big and dream bigger. but besides all this Nan you want to finish up?

24:58 >> Yeah the the.com bubble did not slow down internet speeds. So >> no, >> it's like yeah, there could be a big deflation of the bubble. I don't think it's going to pop because there is real demand there. people need them. But yeah, we didn't stop building out internet after the com bubble. So I I do want to acknowledge there is there is a real risk of financial crash, but I that just might make RAM prices cheaper again. So, I'm >> You're kind of pro pro that are folks, I think it's time for us to move to towards TLDDR.

25:36 >> the highlights of this week and then we'll we'll talk about open source. I think we'll start with open source this time. Let's move to TLDDR. All right, here it is. TLDDR is our segment on the show where we talk about basically the highlights of the things that happened in the world of AI. We already mentioned and I think it's it's that important. I think we already mentioned the Will Manius and Joe Roberts publishing the commanding heights and essay of who profits from AI and whether AI is going to be you know part of it. I strongly recommend you guys at least check this out, skim it.

26:16 but this is a great we can read or listen. and it it it definitely outlined some of the stuff that we talked about before. On Tropic, speaking of what Non mentioned, Entropic, this week it added the rule that that decided no game clones anymore. So that you can still do it with some other models, but since a lot of folks have been doing cloning on Tropic have been playing with this. Would love to hear N thoughts of this a little bit.

26:42 >> It's still fine. It's still fine, guys. It still works. >> Yeah, they just did enough so they don't get sued. >> All right. they're looking out for us, but also they're looking out out for that sweet decompiled game training data. >> You can still do it. >> Cognition releases Sweet to Flash, which is their fast and cheap model that basically is achieving the same score a little bit better on Frontier Code while being significantly faster. This is based on Inkling from Thinking Machines.

27:10 one of the price of Swoo. we'll mention this as well. And this is in the world of the big companies and LLMs. One of the most exciting things that I think is going to be super exciting, Siri finally opens up to every app. Siri opens up intent from the five categories that previously there were to every app. So Siri could be the overall assistant now in 27.3 iOS 27.3 better. I'm really looking forward to whether or not this will improve Siri in any way now that you can use intents from every app or not. Let's go to open source. Open source has had a big week. Recursive recursive intelligence. I believe that this is the lab. Not not to be confused with recursive AI, the other company that we had. the names are getting very confusing. announces open sources a dense repole 65B preview claiming 80% on terminal bench with just 65 billion parameters. This is a very efficient model that can run on local machines. I'm going to be very excited to chat about this whether or not folks have tested it out here. Moonshot released Kimmy K3.1 Pro. Finally, we talked about Kimmy. the weights are out. It's a 2.8 trillion parameter multimodel. It's now on hugging face and it is 104 billion parameter active.

28:32 it apparently on terminal bench 4 beats Astra. This is the model. Obviously not something you can run locally but many companies in inference companies already running this so you can test out Kim 3.1 MIT CEL released a banger report that I think across all our timelines that GPT6 Astra supposedly edits Quen weights directly when while knowing the weights. I would love to hear from folks here on the show whether or not this is as exciting as they claimed. basically saying that lifting quen 3.5 math 500 from 7% to 48% by editing its weights directly. I would love to hear from you guys what this means and also Tencent launched HY4 with vision and they say ranks among the open the best open weights model on finance evil as well 1 million token context window HY4. So this is in the open source. Anything else in open source that I missed that's important?

29:30 Wolf from Nist LJ any huge models in open source that that we should mention at least on stage here? >> Yes. >> Anything from hug and face non >> there was the Swiss Institute quantization of deepseek >> GSQ RCO. They also made a coder version where they cut down half the experts. anyway, basically you can run DeepSk now on just the 70 gigs of RAM at full speed or you can just use an old GPU at 32 24 gig GPU and run it at like 20 tokens per second. I think that's because it's the the model is extremely capable. yeah, I don't the the deep sea quantizations and the the un unrestricted fine tunes those were there were quite quite a few of those also from unsloth as as well the the two bit quantizations are getting very good actually useful >> send us links for that so we'll add this to the show notes while we continue in the LDR let's say obviously AI and policy a lot has happened with the kind of the midterms election in the United States for like this ffects the rest of the world as well. One of the things that we'll definitely discuss is that voters reject data centers or at least data center without rules in 38 out of the 39 local ballot measures in midterm elections in the US. We'll chat about this and what this means what this means for companies like core that we work at. our co actually posted about whether or not this affects the current buildout and our plans to the future.

31:08 will definitely read out that in tools in aentic engineering would love to keep you up to date about the some of the stuff. Deepse released a harness. It's called the DeepC harness and it now has a creator mode that turns chats into installable plugins. so we'll definitely mention at least the fact that Deepseek now is not only a model company, it's also a harness company. Like everyone is everyone gets there. there was a fake cloud code extension that reached almost 5,000 users from inside cursor and then also warp was hacked and then also the Non you said playright had a CVE and vulnerability and it's getting it's getting kind of scary out there folks. So we won't spend a lot of time every week because if we do then it's going to be a doom show. If we if we tell you about everything that was released that's scary this week.

31:58 We'll definitely at least tell you like, hey, the highlights is you have to keep up to date this and this and that. And N would love to for you to like at least mention to folks about the playwright thing because it does feel like a big one. >> Oh yeah, sure. Go ahead. >> You can do it later or >> Yeah, just like one or two lines now and then we continue in the >> guys just get your cloud or or codeex to do a daily scan. Just just set up a daily scan. It will tell you a lot more and get a phone notification. This is how I I find most of these things and it usually helps and it's on time before you have other masses of agents mass exploiting everybody else. I I think that's the the simplest thing you can do. I mean I can't really blame playright and and stuff because the thing is like three days old, four days old. It's not even within the team's sprint meeting to include an an update. But yeah, at at the same time had another friend demo something where they just clicked on a link on their MacBook and it's it's gone and they they they wiped it. It was like they they were demoing it to each other like what what you can actually do. So as long as you just update Chrome every day, make sure it's updated and >> had on the show was that last week? I believe last week or maybe a few weeks before there's a few things to update, few things not to update.

33:27 Packages like npm, python, etc. Do not update all the time. Just make sure they're like at least 24-hour old browsers and operating systems. >> Update immediately. >> Update immediately. If you see an update in Chrome, restart that >> Yeah, Linux update immediately. Packages keep like one week thing. And look, the agents are very good now. ask them to put your npm stuff in a sandbox. >> literally just that it takes three sentences. You don't need a big skill or anything like that. Hey, scan my computer all Python and npm packages every day, please. Thank you.

34:02 >> Right, let's move on. so in the tlddr, thank you. In the tlddr also something open air reported yesterday 948 Linux kernel root exploits. 948 Linux kernel root exploits in a single run cost a swarm run cost a million.2 million in API credits generally. these the swarm basically went and found a bunch of vulnerabilities and then validated a bunch of them and then like apparently fixed them. This is a lot and I think this harks back to LJ what you mentioned like let's get to verifiable software because this is scary as there's a lot of kernel root explains that they found and hopefully have said this is why we tell you update the US because they obviously before releasing this talk with the Linux maintainers as well. is there a breaking news in the middle of my TLDDR Peter?

34:55 >> Yeah, it's very very timely. >> Okay, let's go. AI breaking news coming at you only on Thursday I. So when you say very timely, you mean very very timely because as I was reading out the news item from yesterday about OpenI's unreleased model that did Linux vulnerabilities, OpenAI decided to to tell us what this model is. Peter, go ahead. All yours.

35:28 >> Yeah. So you remember this news when it turns out that Linux had 948 privilege escalation exploits. so this is the model that found it apparently. so yeah opener launching GPT6 cyber a major new advancing security research. so yeah available to day break red customers which are I believe is like the highest year right. So they had GP6 Astra at Daybreak Blue and now released GPT6 Cyber which scores higher than Astra. So it looks like probably a post trained version even further from Astra seems to be the case.

36:08 >> Yeah. On exploit gym which tests whether models can exploit a specific ability. GPT6 cyber scores 48% up from 32 from the previous GP 5.6 Cyber. it also follows ST boundaries more reliably. G56 Cyber avoided exploiting out of scope services in 99.93% of tests without its deployment prompt. If you guys remember the hug and face swarm hack was all about the models hacking something out of scope because they were met with impossible tasks and 100% with this prompt. The improvement in alignment reflects dedicated training to stay within authorized scope, respect restrictions and report results accurately. this capability is already helping secure real systems. Our researchers use GP6 cyber to develop 948 privilege escalation exploits in Linux identify 81 previously unknown high severity high severity vulnerabilities in Firefox despite extensive prior testing with Frontier models. the model. So basically all of the other models, Astra, whatever already like went through the Firefox codebase and tried to exploit as much as possible and this model went in there and found 81 other high severity ones. What the is happening?

37:23 lovely. So this is GPT 6 cyber security teams at Cisco, Calif and Code Rabbit are using it for vulnerability research and code review. We have relied on it to find and we fix weaknesses in our own infrastructure. Yeah, I think it's important. Our goal is to dramatically expand daybreak bring advanced security research to organizations of every size, independent researchers and open-source maintainers. So, open I say we have a limited window to use frontier to secure software and this is the defender window that Greg Bachmann talked about the defenders win whereas as capabilities improve defenders have limited time to use those like new models before other models obliterated open source deepseec clones etc catch up and then use the same level of capabilities to hack.

38:11 and it seems to be kind of like a cat and mouse race. very interesting. So, this is GBT6 Cyber. I >> feel like it's the the most genius business model in the world. We're going to build the models that going to hack you and we're going to sell you a model that's going to protect you and you the latest one. >> This could be bigger than just regular tokens. I I'm absolutely agreeing with you. They're now, you know, we've unleashed this thing and the only thing that can withstand with this thing is a swarm of other things that we can also sell you. Yes, this is this is if we're talking about commending heights and we're talking about is AI over whether or not the democratic elected officials can pause AI when when crowd strike or a big security company comes to him and says hey if we pause not only will the capabilities catch up but also we won't be able to protect ourselves then the answer is like very clearly hey okay let's protect ourselves I think Peter I think you're absolutely All right, this is like not only a genius business model. This is what's happening like de facto. 50% of exploit gym is done. 50% of exploitium. This is crazy. Astro does zero. Just look at this graph. This is a beautiful graph. If you don't have Daybreak, so you don't have removed safeguards, which is a lot of the stuff that we talk about here with open source that you can do. Astro gets zero. So asteroid is de facto significantly worse than using an open source model at these tasks to protect yourself if you're not part of Daybreak.

39:43 In the form the forge powered by coreweave we will read out the statement from Mike in trader the CEO of coreweave about the effect and the data center buildout that we currently have the effect of midterms on that and also cor brings 22 megawatt in Copenhagen site. So we're expanding internationally and adds Crowd Strike Falcon security to its AI cloud. So integration with Crowd Strike I think Wolf from you mentioned that will mention this as well. Vision video Pika open sources H3 flex now they're saying the top openweight model and design arena for text to video. So text to video keeps improving. I think the last thing here is OpenAI rolls out new images 2.6 with GIF and two new API models. So new images from OpenAI that they say are 50% faster. very briefly huge news that I missed in TLDDR before we jump into open source because we have a lot to talk about in open source.

40:41 No, let's talk let's go to open source folks. Open source AI. Let's get it started. All right, folks. The the number one thing that we have to talk about, I really want to hear from you guys what this means in open source. this is only available because open source is available. there has been a very very popular release paper from Omar Hatab and some other folks at MIT CEL that say that GPT6 Astra can natively edit Quen Weights and I cannot wait to hear about like what this actually means. So I think here's the here's the cold story. Can the frontier model fix another model just by adding its ways directly?

41:40 researchers at MIATC cell say yes they gave GPD6 Astra a tiny open model quen 3.5.8 billion parameters and ask it to make the model better at math. No gradients no training run no weights and biases. Astra changed 0.016% of the weights about 131 numbers and the math score 500 went from 7% to 48%. Instruction following went up 3x as well. They call this weight space programming. The idea that the frontier models can now read and write neural network weights. folks have rebuilt this and and tested it out. It's not as bombastic as it looks I think but some people did did make u reperc sorry re rebuild this in their own stuff. would love to hear from you guys. Are you excited about this?

42:33 What does this mean? Can AI change its own weights at this point? Is this RSI? Are we are we so back from being so over just a few moments ago? Niss, what does this mean? A model can can go do this directly, >> guys. Guys, what the guys? What the What the Like, >> first of all, is this real? Yum. Is this real? Because some folks are saying that this is maybe overblown. >> Look, I suppose it's real. Like, results are results. I mean, it probably did what they say they it did. For folks who don't understand what this means, Nissan, please give us like a brief explanation of what this means, how does this like even translate?

43:15 >> So when you hear parameters on a model like 0.8B or 200B, that means billion parameters that so if they're 8 bit parameters that's in in gigabytes. So 0.8b means 0.8 gigabytes. And there's actually quite a bit of data that you would need to fit in the context of a model. But context windows have been getting pretty big at around 1 million parameters. And also if you train the model for this, it starts to recognize those patterns. Now this has been known for some time that you could do this with smaller models like LSTMs or like even other simple like GRUs or classifiers or or decision trees. You could you could do this before, but those were very very very tiny models like a thousand or maybe thousands of parameters, sometimes millions, not not billions of of parameters, but yeah. So basically, Astra or Fable just goes in there and just changes one bite here and one bite there and that that that makes it better. Personally, I am not that surprised, but I've known this for some time for the smaller models. I think it's pretty cool that you can do it with with the bigger ones, and a lot more people are going to be training them for this. So I I think for the the next releases is where we're going to see some real stuff. But people were able to do really cool things. Like Alpin made something for he just did a syntax highlighter for for C and it was 980 parameters.

45:01 That means 980 bytes that that model is that's a 1 kilobyte file. explain what Albin did here because I think people don't really like I I also didn't don't really like understand what's like he wrote this fable wrote this neuronet weight by weight like this not a result of training run >> yeah yeah >> this is the result of token spitting those tokens being weights >> and resulting in the weights file >> and it can do syntax highlighting which normally you need quite a few kilobytes of >> we often do this I have to pause you we often do this on the show we talk about mind blowing mind-b blowing mind-blowing things in a way that we just kind of glance over them.

45:45 What the are you talking about? the model just wrote the weights instead of the the thing that we did as humans of of of training this curating like a thing with a training process that we like okay all of the data send the buildouts all of the GPUs the Jensen sends all of them that are used for training models the reason why there are running matrix multiplication is in the the result of that is to train a model that can do some stuff and then there's post training etc. What we're saying right now there is a new field of research that now a friend of ours Alpin is doing on the flow on smaller neural networks including some stuff happening on a model that's nearly 1 billion parameter that the models can do this process without that metric multiplication via inference not fully not training astron models but the model can print out weights and those weights will not just be gibberish it will actually do some stuff is that is is fair summarization of what the has just happened. LDJ, please please please explain what I just said >> or calm down. I don't know like how do we react to this? The model can just like print weights. What the Yeah, I'm I'm not insanely surprised because I do recall I want to say about 12 months ago roughly when GPT5 first came out there was actually I saw some instances of people making specialized models like basically instructing GPT5 to be like hey just try to spend as much time as possible focusing on the task of making a model that's like a thousand parameters specialized to I think it was like doing specific types of math operations or something like that. I think maybe something slightly more complex than that. This is like more complex and and more impressive of a step beyond that. But this is I think this is expected on you know that trajectory of like hey once you see kind of sparks of something happening then you can expect within the next one two years after to see like significant aspects of that same thing happening. I think that's what we're seeing now. But I don't think it'll reduce demand for compute a lot for like >> training >> overall AI or something because even even if training costs reduce by 100x well for one that allows you to then like >> effectively do 100x more right like throw 100x more resources at the problem but also you could use a bunch of that for inference for serving for scaling up test time compute and so many things like that >> it's it's faster than running the actual C code because the neural that is so tiny running out thing, right?

48:23 >> Yeah. Yeah. It just highlights it just puts the pretty colors on code that you see. That's all that's all this does. But it's actually faster than doing that programmatically in C. You just let this minuscule like less than one kilobyte neural net do it and it's it's faster than actually running the code which is >> what what are some implication of this? Okay, so obviously we're we're looking at the world where RSI is coming. Everybody's like I don't know if it's going to get positive etc. like RSI is the thing recursive self-improvement the models improving themselves cloud in their IPO we talked about this Peter remember like it was like 20ish% of the the the things cloud now you know 85 or 90% of code is written by cloud within entropic but now 20% of the research directions is now also dictated by you know cloud inside this is a more direct way towards RSI potentially this is like the model looking at it own weights and say, "I actually this new run, let's tweak this to to make it better." Like honestly kind of scary a little bit. Peter, what's your take on this >> with things like this? I And obviously it sounds exciting, but I never quite know like is this is this just a trick or or something? Because I'm just thinking about like what how would it know what to tweak? because I I can sort of see maybe okay maybe there was a known bug in how I don't know model processes maths or something and that's because the jump is too big right for it to be like a plausible explanation I think so I think it probably maybe it knew how to fix a bug and then suddenly maybe it processes things a bit better >> that's that's the reaction of Lucas bear by the way Lucas bear that is now at Meta and X open the eye, deep mind and brain. he basically said that pretty much that's what he saw. He didn't see any kind of like incredible things there. He he saw a model that that was told to go fix something and and kind of like found how to fix something.

50:27 >> Yeah. And and the category that's why I'm also excited about frontier models that that are so much better than what we have now is that I think they have capability or they're just about getting capability to actually invent new things. So I think the model's not quite there but for example I think a good test would be can you invent a new way to do mechanistic interpretability like could we actually do that because I think right now my understanding also all it did was kind of use existing techniques and tweak something like is that is that actually that interesting I mean maybe there's a bit of utility but I don't know I bet people at anthropic already do that probably every day so maybe we just didn't know about it I think it's good But it's probably not that revolutionary is is mine to achieve.

51:14 >> Here's my only reaction to the the labs are probably doing this. there could be that many people in topic and I tried this and then they discard it cuz like they didn't see that it's valuable etc. and the value for this is still in play. We'll see whether or not it's valuable that the model can like edit another model's weights. Can the model edit its own weights directly like this? we we don't know. Like obviously Quinn 0.8 8 billion parameters is such a different category of model than than Astra itself then you know this may not be applicable in the future as well but it's very very exciting and they call this weight space programming where the model just spits out u basically neurons that they kind of work it's kind of similar to like a birth process if if a model can generate another model just by thinking about this if you think about this where it like goes In the realm of super intelligence, it's kind of very like an agent can spawn another agent of the same model.

52:18 We're not talking about that. We're talking about a model basically thinking things into existence that are specified for other stuff. Like Nissan said, the Alpen example that we we see, it's basically a neural net that does a thing better than the programming that it does. So instead of writing a code, they can spin up a neural net that that is faster at that thing. I think it's very fascinating in examples of what we can do for the future.

52:43 >> This is going to get crazy. >> It's going to get crazy. >> Yeah, you're going to need like 10 million context windows and stuff for it to even do something revolutionary, but yeah, at the moment it can it's pretty easy to train for. It can pick up patterns and we demoed last week it it turning it evil or or or good and stuff like that. There are quite a few patterns in there to pick up but it interpretability is like a science. is like 5% science. It's it's like trying to do astronomy 3,000 years ago. Like you had you don't have the right tools type of thing.

53:18 >> This this could be yeah a huge thing for me as well. all right, LG. Unfortunately, we do have to move on because we we can talk about one topic forever, but we do have guests that are coming. So it it does kind of compress our timelines a little bit. let's talk about other stuff in in open source. I do want to mention obviously Kimmy 3.1 finally the open weights are here. Kimmy Moonshot the company released Kim Kate 3.1 Pro. The weights are out. it's a 2.8 trillion parameter model multimodel is now on hugging face and and in some other places that run inference as well.

53:55 folks are saying my GTX 1030 is ready because what can local GTX do with this huge model? like pretty much nothing. But this model does show an incredible Let me try to zoom in here. see if you can guys can see as well. no, zoom does not seem to work here. this the model does show terminal bench 4.0 moonshot reported only. cloud opus 5.5 gets 66% on that terminal bench and Kimmy K3.1 gets 65 just shy of the Opus 5.5.

54:29 a token of million windows a window of million tokens a context window and then text and images run natively. what do we have to say about this model? Who's excited about like the next scheme? Maybe >> this is really funny. Maybe one thing to mention and I I'm not entirely sure what's the whole background to this but my memory when Kim K3 Pro came out or I I think they didn't call it pro but they had something weird there must have been something weird with the license where basically on open router all of the prices were the same at launch. I I haven't checked if they've moved, but they were the same for a while. And I think there was some kind of rumor or someone commenting maybe they had some restricted licenses that would restrict the price that providers were allowed to charge. To me, it kind of nearly defeats the point of open source for these kind of big models.

55:27 Well, if it's all the same, I mean, you might as well just run it from I know your bedrock. What what's the difference? But now if I look at open router for Kim K1 Pro, first of all, it's on loads of providers. And what Moonshot is charging is $4 in, $20 out. And if I look at Modo for example, which is a very good company, they've got like $2 in, $10 out. So it's half the price on on day one basically. So to to for for me this is this is very impressive. That's a big difference because before it was kind of disappointing I must say. I even set up a job to track the prices and then they never changed. I was like, "Okay, that's pointless." so yeah, I'm quite excited about that part. The price might actually go down.

56:14 >> It does look like there is price action on this and it does look like maybe, you know, there's changes in how the licensing worked. let's talk about the license as well for for this Kimmy Wam. Do you have the license there or >> I have it open and I analyzed it already. So it's not an open- source license in the usual style. It is open weights. You can use it and for most people it doesn't affect them. But the license specifically states if you have a product or services that have more than 100 million monthly active users or more than 20 million US or equivalent in other currencies and monthly revenue, then you need to display on the interface the Kimk 3 name. And there's also a provision to make a direct contract with them. So you have to have a special license for this. This may be why Peter saw the pricing to be the same. Maybe in the previous iteration the the providers were falling in this and had to make a license where they couldn't go lower. This seems not to be the case anymore. That's a good thing.

57:14 >> Yeah. so so the the stats reported by this model are huge. LJ Nissan or Yam any thoughts on how well this model performs? Kimmy is known to show great models and benchmarks, but also like folks are like smelling this model like well and they like them. but also there's a lot of talk about Ki specifically doing distillation on the Opel stuff and maybe this is why they smell like that. I you know any thoughts on the model itself here? Yeah, unfortunately there's been some things with Kim Kate 3.1 Pro that it seems like like for Terminal Bench 2 and and 2.1 I think it is when they presented certain benchmark questions from it but like kind of modified or edited there there seems to be evidence of Kimmy recalling what the actual benchmark questions are even like details that are not given in the prompt and but are actually in the true original benchmark questions and things like that and seems to be potentially benchmaxed in a way that the previous Kimmy ones aren't. but overall it still seems like a pretty good model and improvement on on the past Kimmy model but it just a bit disappointing what's being found there.

58:27 >> A little bit disappointing. Yeah. obviously not a model that people can run locally. So the openness of this only means you know the the folks in China are showing given models in open weights and this open weights are under specific license. and then there's a bunch of fine prints like Wolf from said, but a great like you know model overall. shout out to Kimmy for releasing this. also there's a few like Neil LJ said, there's a few different mis discrepancies in in the checks as well. Na, any thoughts on this? Yam, any thoughts on this or we move on?

59:01 >> The the vibes have been have been mixed. It it does still look like it is the best open source model you can get. there were some like you could do it with like 50 grand of hardware. You could get like 30 tokens per second or or something like that. So it's not bad, but I don't it's it's still not quite Opus 5.5 level. It's more like sonnet 5.5. And if you're going to do that, you're probably just going to run Deep Seek a lot faster. So, I don't really recommend this over Deep Seek 1017 Flash because it's not really that much better. So from from huge models, let's go to models that you can actually run locally and recursive intelligence not to be mistaken with recursive SI the folks that we had on the show just recently released Ripple 65B preview 65 billion par dense LLM pre-trained from scratch. We don't often get those.

60:05 We often get fine tunes or continue training things etc. This is leper train from sketch model. ripple sets a new state-of-the-art in the US open weights model in agentic coding tasks. So did we get new sort in open weights on the western frontier? It looks like potentially yes suppress surpassing significantly larger models including mu spark open beam which we talked about from reflection I want to say I'm getting a little bit confused by recursive reflection recursive etc. But yeah, Bim from affectionately Michelins company, Inkling from from Thinking Machines and Neimotron from from Nvidia. So, first of all, my reaction to this is the more the marrier. We'll learn to remember the lab's names for sure, even though they're a little bit confusing at this point, but the more the marrier on the Western Frontier. I'm like very much applaud this. folks, what do you think about Ripple?

61:06 Let's take a look at the terminal bench kind of like change. Oh, this is just the video. Okay. >> Flashback to three years ago when we had 65B models coming out. >> Yeah. I I want to see the actual things. Okay. Let's take a look. So this is them comparing to all those models that we told you about before. so Meta Muse Spark open which just released 816 billion parameters. This is a 65 billion parameters.

61:36 That is quite insane. A terminal bench 2.1. This is a >> people did a lot of tests like webgu and graphics and stuff. It wasn't that good. I would have to test it because it is something that that I can I I can run and if you if you quantize it, you can actually get it going at a pretty solid pace. But the the early vibe checks don't seem to quite match what the benchmarks are showing.

62:08 >> But I was look at what they're comparing themselves to, right? Like reflection beam is it's half a billion. Half a trillion parameters like 500 billion. Mu spark is 800 billion. inkling from from the thinking machines is 975. This is the model that 2 light was trained on right like just on top. Neatron ultra is 550. This model is 65 billion parameters. There's like 10x smaller than all of the other models they compared themselves to. The balls that you have to have to come out with with results like that and say, "Hey, we're at the start of Frontier at 65 billion parameters." Very very interesting.

62:44 Whether or not it performs like that, that's that's the different story. looks like the downloads on Hugging Face are ripping. and let's see the license Apache 2 as always, you know, Western license Apache 2 shout out. We can test, we can ask Astra to look at its weights and see what it knows about those, you know, doing some open weight modeling. what else? LJ, you have anything about this? anybody else? >> Yeah, I think it's exciting. I think I want to see more models like this to try to be something in the middle that could fit on a lot of people's hardware, but is dense. And I think there definitely is something to that where you end up being able to maybe get more agent capabilities for a given VRAM footprint when you do that, but of course at the cost of lower token speed.

63:36 I'm reading their hugging test right now and they basically said they pre-trained from scratch, didn't they say in the in the in the release of >> Yeah, they used the Almo >> but now it says that they used Oh, they used the Almo data set but pre from fully from scratch. Yeah. >> Yeah, maybe they used like maybe architecture 2 or something, but like yeah, the pre-training from scratch. I think when I was looking at their their numbers for like mid-training and post- training and all that, the RL was only something like like 5% of their pre-training compute, which tells me like they they have a lot of room to still increase it a lot. As we know, some models now even at large scale are doing even more RL compute than the pre-training. So, I think this is really just the beginning for what we could see at this size.

64:28 So this is quite fascinating. We're reading through the blog and they're talking about like ascent and reinforcement learning. we carefully curated our SFT and recipes with a focus on compute efficiency. We opted not to buy any post training data. Instead devising our own systemic data pipelines. We've seen this more and more like the more like folks are building their own data with the same thing will happen with with type safe the the better kind of results they get.

64:53 look look at this crazy thing they're showing. this is terminal bench scores ripple 65 billion parameter the one we're talking about right now versus reflection beam the model that we talked about from reflection this is millions of rollouts in RL and the score on thermal bench and the the difference is just striking this is like just one million rollouts they get a score of 80% versus 80 something million rollouts and to get to score that's less than that that I think that this is like a fascinating chart comparing them to their competitor.

65:30 >> I I just checked, by the way, shout out for them releasing the base model as well. That's that's pretty amazing for people that do need to fine-tune something like if you're going to do legal work or other things like that. You want to use a western model like you can't for stuff. So, the fact that they have the base model is going to mean there's there there's actually going to be quite a few fine tunes of this. This is why we tell you all the time, this is why we applaud. This is why we do all these things. We talked about Almo and how important open source is. they're shouting out by the way like the the folks at Almo have been gone to do different things. I don't think that like that team is still there. but like Almo from Allen Institute for AI, Prime Intellectics and Nvidia, they're saying that hey we're building on top of this incredible open source contribution including those strong open source ecosystem is improving contributors strengthen the ecosystem which in turns draws more contributors. This is literally why we celebrate open source on Thursday so much because this level capability is is possible. All right, moving on from recursive ripple I think it's time for us to chat. We we'll briefly mention Cognition Sweet 2 Flash and then we'll move on to our interview with with with our next folks. Cognition released.

66:43 Let me see if I can pull this up here. this is not an open source but we have this is based on open source. This is based on the model. It's not open source. Inkling I don't believe it's open source. Cognition releases debutes sweet to flash. It's a 12 billion parameter 12 billion active parameter coding model at 1/5if of the price of SUI 2. now I don't believe that they open sourced this. Somebody please correct me if I'm wrong but the here is the they do mention the number of parameters. I wonder if somebody can go and check whether or not this is open source. 270 billion parameters 12 billion parameter active on frontier code which is Cognition's own benchmark.

67:28 they're getting 24%. Compared to Beam, the the one that we just showed you from from reflection is at two. Muse Spark open is at six. And then the Frontier models, GP6 Luna is at 21. Haiku is at 27. Haiku is incredible model. Haiku 5.5. And this model is at 24. This is a flash model that's basically the same asu from before. It's just significantly faster. you can access this model inside dev and I'm assuming for free because that's how cognition usually roll out. any thoughts on suite 2 from folks here before we move on? LJ, go ahead.

68:06 >> Yeah, so it is not open weight but it is based on inkling small and they're transparent about that and so I guess that's why they list a parameter count because it's it's obvious just from the statement of you know it's based on that model. >> Yep. Cognition also beats it claims it beats every American open weight model including models three times larger. this was before the Ripple announcement as well. So, shout out to Cognition. Hopefully, if you're basing this on on a model like that, maybe maybe open source. shout out to Cognition for this. I think it's time for us to move on to our interview with with a friend of ours.

68:47 it's been it's been a while. and we we'll do from the forge a little bit later. Wolram, you want to you want to do the honors to introduce? >> Sure. Let's do this. Bring him on. >> All right. folks, welcome to the stage John Davies, a longtime friend of the show, I believe. John, we met once in London. and the woman told me that you you're launching something exciting of your own. I think the whole panel can stay here because I think whatever is launching is relevant to like many many folks here. how about we start with just you telling us who you are and what you're doing before we dive in into the interview.

69:27 >> Okay. Hi, I'm John. Good. it's wonderful to be on the show and hi guys. I've seen many of you before but not all of you. I work predominantly in the financial services industry but been in the AI side for the last three or four years. We're huge fans of the private models, open weights and providing those for governments, banks, etc. who basically just can't put things up in the cloud. They they just can't have their data going up there. So, we spend a lot of our time not training the models, but just getting the most out of the models and putting putting the models into the flows. They're tools at the end of the day. So we we put compliance, anti-moneyaundering, that sort of thing through the models.

70:11 >> And John, you've been following yourself for a while. You've been following the world of open source for a while. You talk about Sparks, you talk about Mech Studios. what's your current hardware count? >> I test local models. >> I'm just sitting in my my home office here about 2 terb of VRAM in terms of Sparks, Studios, Minis, etc. some of those will be older. we have access to sort of that again in office and clients around around the world literally from Central America to the United States and Europe. So access to quite a lot.

70:47 These are wonderful boxes and they do amazing things. >> And you're launching something that can run across those boxes, I'm assuming. given that not a lot of folks are running as much like local hardware. tell tell us about your launch this week. So, we've been working a lot with the KV cache. We store a lot of our information. We take it out of the KV cache, put it onto SSD, and then you can get away with bypassing rag. So, you can basically load it back up. And I suddenly thought, hang on, if if you can do aic coding and you can basically take the KV cache, store it on SSD, park it, bring it back up again when you need it.

71:25 and you design something specifically to be multi-threaded. It's a bit like our go back 20 years ago when we had single core machines and then people came out with multi-core machines but none of the software you know word and excel when we're were using the multi-ores we've now got these 30 40 50 core machines they're just not using the the processors properly. You got VLM on the sparks which does an amazing job but there's nothing on the Mac. So I basically since last Thursday, it's literally that quick have spent my entire day, evenings, afternoons, breakfast, dinner, etc. guiding the vibe coding as we all do these days to create a new inference engine. And I've been tuning that on the coin models and obviously Glimmer and some of the other models as well to make sure that it works. And we've got crazy numbers out of it. I mean, this this blows the sparks out of the water. Not just one spark, but two sparks and and goes beyond, way beyond. I've taken some of the other sort of areas in this space. I've taken it up to 64 and then just today, 128 simultaneous agents, and the numbers just keep going up.

72:39 >> I mean, and this is on a Mac. This So, this is just for Mac. I literally just put it live a short while ago. So, it's in info d. So, I nerd D. >> Yep. and then incept 5 which is our company.ai >> inferred. So walk us through walk us through what we're seeing please. >> Okay. So I tuned this on the on the sparks on on the studios. Now not everyone is lucky enough to have a studio. It's a massive unbelievably powerful piece of kit.

73:11 >> But it does mean that I can tune it. But what people can buy >> are the the minis the the M5. The M6 is is incredibly affordable at what about $1,500 US for the 32 gig machine. So I was trying to get as much out of these. Now you can you can't fish fit flash next on it, but you can fit the 27B >> 3.8. >> Yes, 3.8 27B with a drafter. We're using Dlash 2 on these. We get some incredible figures. We can take this up to 16 agents. Many of the other inference engines weren't even able to run these on the smaller machines. And what we will be doing next is connecting these up with the Thunderbolts and and trying to get some better figures out of these.

73:56 so just just look at this for example that at the bottom is splash 1.5. and that runs out at 16 agents. It won't scale past that. Now I've released this open source. It's it's Apache 2. and just you gave a round of applause. Thank you. Thanks Warframe. >> see this is this is a fan of the show. This is a person who knows what's going on on Thursday I apache to John >> now I I have no doubt that everyone else is going to catch up with this by probably by the end of the week because that's just the way it works.

74:33 >> I've been I mean TensorFold Ash's tensifold is absolutely incredible and naturally has has to appear on some of these benchmarks. We got Prince's work on MLX, Maya's incredible Maya Labs builds that she creates. so all of this comes into what I've I've helped to build here. But but look at this. Splash 1.5 stops at about 400 tokens a second at 16. And we just continue to go up. >> So explain to me what what does it mean stops and continues to go up? Like what we're talking about here with parallelism.

75:08 >> Yes, parallelism or concurrency. just don't they just don't scale at the moment. Now again this they will change this and I'm sure you come back on next week next week's show have the line going straight up to the end and and it becomes a bit of a race. That's that's what it's all about in in our industry. But >> so many of the others just stop at 16. and and not only that they stop, they flatline. And so there's no point in going to 32 or 64. and as you can see here, we've we've just continued to go on to try and build the performance out of this. And so we're now looking at over a thousand tokens a second. And okay, that's divided by 64 or or 100 agents. So per agent, you're only getting one token, sorry, 10 tokens a second, but ultimately you're still getting a thousand tokens to put into the processing. And and that's what's important. That's incredible.

76:06 so you call this inferred and folks can get access on GitHub. Let's go on GitHub and give you a star obviously number three. thank you. >> And folks can run this on the Mac >> and laptops as well. It works just as I mean many of my my colleagues have been testing it for me on on their laptops. >> All right. Wolf from one more comment and then we we have to move on. >> Yeah. What excites me about this that it's optimized for agent swarms basically and this is a a new thing in that we have been seeing these swarms acting usually just in security related incidents or something but I think this is the next paradigm that we are now seeing after agent that it's not just one agent but really a swarm and I see the benefits myself when I just use a swarm of a dozen agents or something to tackle a topic and the the results are much better than if I just use my main agent to do one thing and I think this is something more people realize soon and yeah having an inference optimized for this is I think very very looking at where we are going not where we are at right now and yeah that's why I found this great >> yeah you how do you use >> we're we're using it with anti-fraud anti-moneyaundering there have been some recent attacks as you know after the PHP thing and so you get you get sort of attack surfaces and you want to actually look at those IP addresses and see what's happening and so you can launch off swarms of agents to basically do do the sort of research on this and and that that's been hugely powerful for us.

77:37 >> So local swarms I think is is what gets wolf from excited. John Davis, thank you so much for joining us CTO inferred from Incept 5. all right folks, it's time for us to move on. from the powered by Wolf room and I will tell you a bit what's happening in the world of Cororeweave outside of the news cycles as well. welcome to from the forge powered by Corweave. As you know I'm myself as an AI venturist at Corwe from as well and this week we have some stuff to tell you about specifically about Corweave and Copenhagen deal.

78:24 so Corwe brings a 22 megawatt data center online in Copenhagen. The cool thing is the data center there actually gives back to the community. I think W this is very interesting. the surplus heat in this data center goes back to Copenhagen's district heating network. I I don't believe that I've seen some stuff like that quite before, but I think it's very very interesting. so now we have data centers in the UK, Spain, Sweden, Norway, Denmark, not quite. In Germany, we have some evangelism to do to to to get there.

78:58 >> our prices. asked about why we do not have the zeta centers in Germany and it's not the regulations but mostly it's actually the price of electricity >> and we have also a deal with crowdstrike to explain the partnership putting falcon protection 24/7 threat hunting on core cloud we talked to you about like local swarm agents many folks are looking to cross track to secure what's happening within them nobody asked me to do this I I'm doing this because I think it's It's very important to to convey this message because you know a lot of speculation a lot of FUD is happening out there. this is directly from Mike in trader CEO of corewave LinkedIn.

79:41 He's saying my belief in some national ambitions transcend politics leading the next era of technology is one of them and here's the the highlight that I want to read out. We currently don't anticipate any impact to Cory's existing customer commitments or associate backlog as a result of shifts in the political landscape. Corweave already has more than 1.5 gawatts of active power across our diverse global footprint. We also have a clear line of sight to 4.2 gawatt of contracted capacity. We've signed as of August 2026 given us confidence our ability to serve our customers. I think it's a very important statement of reassurance from Mike and Trader from Coree.

80:19 >> I just want to iterate something that people may not be aware of. This show is sponsored by Core Weave, but we have full editorial control and everything. So, this is not something the company says you have to re bother them with. This is Alex decision. Yes. >> And yeah, keep that in mind. >> I honestly saw it only this morning. I thought it's very important for folks specifically because we need to get Mike on next because he's posting this on LinkedIn and most of the X crowd is not is not seeing this. So, I kind of wanted to connect those worlds. so, we'll definitely work on this and try to get Mike and Trader on next. All right, folks. Next up, we have a great interview with well, hopefully a new friend of the show, Max Alex Max Alexander. Welcome to the show. This is your first time here. we just met recently in New York. where we got to chat. Max, you're principal you work on developer experience in Capital One, principal engineer. I think that's what the the way you told me to to to to shout out. Is that still correct?

81:14 >> I do. That is accurate. Yeah. >> Yes. and then you I would love to at least briefly mention some stuff that you did before for your career just to folks to understand like who we're talking about. I think one of the things that you recently posted was about hey at some point I was given all of Google to decide how to write software for all of Google. Can you tell us briefly about that story like in one two sentences? I think it's very interesting.

81:38 >> Yeah, sure. once upon a time I was the technical lead for code health for all of Google. So I owned Google's engineering practices. I'm the original author of Google's code of view guidelines that still exists. So most of the text in there was still written by me. >> And at one point I was very sort of briefly grabbed by a coalition of engineers who asked me to sort of be in charge of Google's engineering culture whatever that meant with which I and the story that I was posting was with which I did nothing. I chose to do nothing with that opportunity. but the code health stuff I did do quite a bit of work.

82:09 >> Yeah. code house specifically is the topic of why I brought you on here today. So you gave a talk at the engineer. Does code quality even matter anymore? Tell us how you came to that like conversation, how this was received and basically feel free to also repraise some of the points that you did on the stage there because I think not everybody watched your talk from AMG. >> Yeah. Great. So, one of the things that was happening is I was getting really frustrated like with the quality of discourse on X and on LinkedIn on this subject because people were just like either like super black and white. They were like, "It never matters and I'm never ever going to read the code ever again and agents are going to write everything." And on the other hand, people were some people were like, "No, I have to read every single line of every single thing that everybody writes and I can't relax anything ever at any time." And you know my observation because like I live both in the open source world and in the very large enterprise software world is that these people are not in the same universe.

83:12 they they don't live in the same world. Like like some people are building systems that will be worked on by hundreds or thousands of people and last for a decade. And other people are like vibe coding a front end for a thing that they need for today and that they're going to throw away. And and the the standards that you need for those two things are not the same. And that's always been the case, but it's just become so acute because we've entered a world where a 100,000 times more software is going to become disposable. It you're you're going to write it and you're going to throw it away and then when you need better soft and when you need something for tomorrow, you're going to write that and you're going to throw it away. You know, the the example that I often give people is like the Star Trek computer because that's what like a non-technical person thinks of when they think of what is the computer that I want to have like because you know like Gene Ro is not like a computer engineer. He's just like thinking like what do I want to make for a computer in a science fiction world and the computer in the science fiction world does two things one you talk to it. Two it makes an ephemeral UI for anything that you need at any given time for you to like interact with however you need to. and and and nobody cares about it like nobody cares about how it's implemented or or how it works or anything, right? But then if you think about that like who makes that thing and who makes all the software that works be who makes all the things that work behind it? That's like some serious business engineering stuff that would have to happen in order for that thing to continue to function as smoothly as it does for everybody. And so there's really just these like really two broad camps. And and on one side, code quality really really matters. Like if you're going to write durable software, it really matters because it's for sustainability. But if you're going to write disposable software, just don't care. You don't care. Nobody cares. And then >> so this is the range that you identify, right? Like all of those takes and the quality of code especially they land somewhere on the spectrum of disposable completely to very very durable and stays with you for a long time. Right.

85:11 So can you define the two categories? What is disposable software for you? What is durable software for you? Yeah, sure. So, you can generally tell from four criteria and that is let me see if I can remember off the top of my head. one of them is the consequences. So, what are the consequences if it fails? Very often, very low, right? If if you write a party planner for yourself and it doesn't work, you're like, "Oh, well, that's annoying. I've got to go use a spreadsheet or something, right?"

85:36 the next one is dependencies. So, how many other people depend on it? How much change management would you have to do if you changed it? How many other systems depend upon it? How many like you know could you break like if you made a breaking API change would that affect other people would that affect other systems? And the next one is cost of replacement. How long would it take you to rewrite the whole thing a new you know like like I can rewrite you know a calculator in an hour. I don't care about that. Chromium I would have a harder time.

86:03 >> Yeah. I don't think I could recreate Chromium at all period. And the final one is cost of maintenance and in particular how much human attention the system requires in order to continue running. Like I've seen a lot of systems where where people are like oh it requires some human being to help you onboard and like every single and then like you like trying to onboard a thousand people at once and then you suddenly you need like a thousand people to onboard a thousand people and like that's that's a that's a thing where the quality of the system really matters because all of a sudden you got to think about wow how do I reduce the cost of like onboarding just as an example not even the cost of the code maintenance but even just the cost of like having the thing exist.

86:40 How do you see how to actually write proper durable software in the edge of AI where more and more what we do is you know instead of chiseling code by hand now folks are monitoring agents doing some stuff. How how do you see this world at at present? The way that I think about it is reducing the cost of durability because and and I and I talked about this a little bit in the talk but like you know how like people people often think okay well I know all these engineering practices I know like oh I can write tests I can write verification systems you know I can add my static analysis tools the what people rarely think about is how do I make that easier like I that's what I've I've spent my whole career like the last 20 years working in developer experience where my whole life has in how do I make it easier for software engineers to do the right thing? Not just tell them to do the right thing, not just I'm going to go do the right thing for them, but like make it f make it so dead simple to do the right thing that nobody would ever do anything other than do the right thing. Like the the way that people have built software and and fundamentally and and I think we're going to talk about this in a little bit, the way that programming languages are designed lends itself to people writing bugs. Like my whole quest as an engineer for my entire career has been to simplify the act of software development because I think that what's happened is a lot of very smart people have decided that because they are smart they can cope with complexity and we are now seeing that the answer was that they could not and that was all untrue.

88:17 >> So basically you're saying we humans are dum dums compared to the stuff that we create and it's impossible to hold everything in your head. despite thinking that that we can. So h how do we still continue there? Like what's what's the next step? How can we scale those Simpsons larger and larger given that we do now rely on artificial intelligence more and more? >> You have an agent today and it is more powerful as a software engineer than anything has ever been in the history of mankind. You can choose what to do with it. You can choose to direct it in a direction where you're like please just ship you know results for me as fast as possible or you can choose to direct it in a direction that's like hey what do I need to improve about this codebase tell me what not just what are the current security vulnerabilities but how could I structure the code in such a way that a any vulnerabilities would become totally apparent and b those things would be isolated from my users or from other parts of the system even if something went wrong and and then also and look and then also a lot of the work that you've seen around like formal verification and and static analysis and then this is also a lot of the work that and this is also like a goal of my personal side project too as well >> which I would love to go into next. So I have a side project. This is my personal project. So I just want to be clear. I know that my my little subtitle there says capital one, but I want to be clear that now I'm going to talk about a thing that's me personally.

89:44 >> many years ago in 2014, I had an idea about a programming language. And I had a realization about programming. which is that it is basically the management and structure and design of knowledge, not the design, not really math or engineering in the in a more traditional sense. And I was never able to build it because I was thought I thought, well, nobody's ever gonna pay me for two years to go into a room and produce nothing for two years. And finally with when Opus 4.5 came out, I realized I finally could actually start to build this thing. And so since November of 2025, I have been working on a programming language. and this programming language is designed for perfect static analysis.

90:34 and what that unlocks is many things. you know, like I can list off a few of them. It unlocks theoretical maximum perfect optimization. Like it it enables like like theoretically you could have a better optimizer than any other optimizer that's ever existed in any compiler ever. it unlocks the ability to automatically refactor any piece of software and know that it is and know that your refactoring is correct without having to run the software. it unlocks forms of supply chain security that are better than any other supply chain security that's ever existed. and but I think one of the most important things that it unlocks for in the world of Agentic software is the ability to deterministically make rules about what your system may or may not do.

91:24 Because like what you need in a world of agents in order to totally like take the hands off the wheel even for like the most durable of software is the ability to say not just what the system does, not just what its behavior is, but how it accomplishes that like let me give you an example of a problem. >> Give me an example >> the like one of the things that inspired me to work on this. So at Google, I owned a static analysis framework and the most important check in that static analysis framework said don't do blocking IO on the request thread. And the and the reason for that check was like a significant percentage of Google's incidents were cascading failures because somebody did blocking IO on the request thread and then requests queued up in some situation and then that caused a cascading failure.

92:13 And so there was a a hard check. It was a test failure. You couldn't ship the code to prod. It said don't and the check was basically impossible to write and had constant false positives because in languages like Java or Python you cannot successfully trace code paths without an incredibly expensive algorithm. >> Mh. >> You cannot actually it is very very hard to know that like function X was called at any like like I just want to write a rule. This program never calls function X. Okay fine. that is incredibly hard to write in traditional programming languages. And then now you look at something like now now let's say like you want to say oh well now we have formal verification formal verification it's great right well first of all and and look super supportive of formal verification I think the the whole rise of formal verification that we're seeing with AI agents is awesome I think I I I love the idea that we're going to be able to really prove behavior of systems but depending upon the type of verification system you're using the most common verification systems mostly are going to either prove that this specific speific program has this specific behavior which is which is fine that's that's often that's better than a test or depends or this specific program has this specific implementation detail like there's some verification systems that can like look at your code and say like okay you're calling the IO we'll say object just to be simple for people who are not functional programmers we'll say you're calling the IO object and you shouldn't be calling the IO object like this program bans IO so formal verification systems can do that What is very hard for them to do is to write a proof that no program ever does something.

93:57 >> No, no other code that other humans or agents will will will add to the codebase will break that rule basically. Is that what you're saying? >> Yes. >> Yes. >> Like I want to say let's say I'm Google I want to say nobody may ever ship code to prod that calls blocking IO on the request thread. Right? And theoretically, if I could really do that, I would be able to trace SIS calls back to concurrency state.

94:28 And if I could do that, then I could do that for every program. Like I I could just write that rule one time and it would apply to every program in the world. But the programming language has to be structured in a particular way. Because here's the problem. The problem is another program has to be able to look at your program and make statements about it. >> This is what static analysis is all about, right? >> This is what static analysis is all about. It's all about another program being able to look at your program and make statements about it. and and by the way, lots of other cool things that I forgot to mention like the language will do automatic concurrency. So you will get automatic maximum parallelism.

95:06 you will get a a certain level of verification like not full formal verification but more like what's called pragmatic verification for free and it will basically be enforced by the language. I I will tell you even just from my experience building this language over the last year. The only thing that has ever let me take my attention off of the agent is knowing 100% that it cannot do something and I know and now I no longer need to look at that. No amount of instructions, no amount of skills, no amount of agents.mmd, whatever, whatever. No amount of hooks, whatever, has ever stopped the agent from messing things up in my code bases other than static analysis or tests that fail. I think Max I I want to like land on how this helps the world where most folks will not only just run one or two agents will will run agents 24/7 for disposable systems but also durable systems that will continuously improve continuously check for security modifications whatever how what you're building is helping that world of us managing swarms and armies of agents to to do things for us in the future.

96:16 >> Great question. So, two things. One of them is you want to be able to know that the thing works that you wrote, right? And >> that asked for >> that you asked the agent for >> and ideally you want the thing to come out right the first time. >> Yeah. >> And and you want things to come out right more quickly. >> And so like the the total amount of time that you care about is not the total amount of time that like the compiler spends running or whatever. The total amount of time that you care about is the time from intent to correct result.

96:47 And if you can like put as much of that as possible into a deterministic system that like removes all of that choice from the agent, everything that like is not a thing that the agent should make, like a decision the agent should make, you you reduce it. Like let me give you an example. Like one of the things a language does is tries to guarantee that there's only one way to write any correct program. And and that that sounds crazy, but I can do it more right even right now. like with the state that the language is in right now, I can do it more than any other language can do it today because like I don't I don't want there to be stylistic questions. There's no stylistic questions like like if you do have to go read the code, there's only one way it could have been written.

97:24 So like you never have to worry about how it got written. Okay? Like like whether you're an engineer, whether you're not going to write good code because there's only one way to write the code that like it's it's that it's just that way. And the and what that and the other thing that unlocks theoretically that I I haven't yet had enough time to experiment with but I believe is possible is a much lower level of intelligence that is much faster and much cheaper being able to be much more capable >> because you've given it so much deterministic power and so few choices for how to do the right thing. So significantly higher quality output of smaller potentially open completely open source on your computer without any clouds or etc execution because there there's not that much to do like they can write the simple software. okay so Max you've talked about pretty like you know mind-blowing things here.

98:23 what is next here? When when when can people ex get exposed? people can review or other folks reviewing this is only in your realm. let's finish up on where where this thing going because sounds incredible but obviously you have big claims here that need to be proven by by by someone else. >> Yes, completely. And I don't expect anybody to believe me, right? Like I I I need to show people that this thing works, which is why I haven't released it while like it's why I'm not making it public while I'm working on it.

98:54 >> Mh. because I do have some pretty absurd claims and I I want to be able to demonstrate them and so the thing has to work like I have to get the thing into a state where work can demonstrate that my claims are you know within that I'm somewhere within the bounds of my claims and unfortunately I mean look I have a day job I love my day job it's wonderful but it does take up eight hours of my day and so I can only work on this four to six hours a day yes I am working 14 hours a day so that is my primary constraint, which is that I I I do not work on this full-time. I I if I could, I would absolutely be working on this 12 hours a day. Like, like no, no question. I I wake up in the morning, I walk to my computer, I work on this, and I work on and then I do my job or this thing until I go to sleep pretty much every day of my life.

99:40 and so I am working on it as hard as I can, as fast as I can. It is unlike anything else. If anybody's ever done anything novel with a pro with a coding agent, you know that it's not like doing things that are you know, you you're copying something else. So, it's a lot more time and attention from me than it would be normally in most Gentic building systems, but I am hoping hoping that I will have something in the next three months that I can show to people.

100:04 >> Max, thank you so much for joining the show. Would love to bring you on once you are about to release whatever you decide to call this thing and and chat more about how Asians operate in this new world. Thank you, Max. >> Thanks. Thanks, Alex. >> All right. we're almost at the end here. Let's at least cover OpenAI's images 2.6. OpenAI releases their new and up-to-date images 2.6. Images 2.6 Flare and Sunburst. OpenI is leading the AI generation models now back to lead them. Same price as 2.5 but significantly faster. Flare and Sunburst number one in overall design arena and image arena ahead of 2.5.

100:46 now they can also make gifts. I tried this a little bit in it. It didn't really work for me. now you can make gifts within CHBT with these models. Basically what this means is that they sped up the generation so fast they can generate multiple images in succession while you wait. And here's an example of 4 seconds for generation for ads. and then they can generate a few of them in succession and generate gifts. which is pretty cool.

101:10 >> I did want to make a comment on the deepsea harness because a lot of people >> Yeah. >> Talk to me talk to me about this. >> yeah. So I I tried it. Yes, it is. It is very good. >> Explain what Deepsec released like like a cloud code type like tool CLI tool, right? Is that is that what this is? >> Yeah, Deepseek released basically their own cloud code tool. It is very nicely done. It's pretty modular. You can customize it. I tested >> the UI.

101:39 >> Yeah. Yeah. The the I I have the the UI is good. It's very neat. Nicely designed. I tested it for it phoning home and stuff and I found out that >> by default it does have telemetry on as hardcoded. But then I dug deeper and it's not sending your data to China just to remember that. But there is a big butt here. Whatever your agent searches, it goes through api.deepsseek.comtools.

102:12 deepseec.comtools. So all all of your searches and there's not that is kind of normal but there something you got to keep in mind when >> 274,000 stars on GitHub for this harness which released what a few months ago >> very good at scale but they did a whole bunch of updates. It's it's pretty neat and it it hits the cache very frequently and it's very modularly done so you can customize it to to what you want very easily. Just just keep in mind by default if your agent is searching all of those all all of your search results are are going to deepsee. So that's just the the one thing you got to keep in mind. you can turn that off. it does not send your your data. I mean I tried I thought of like a honeypot thing to see if it would send the data. I couldn't find anything wrong. The only time it does send the data is if you explicitly provide a slash feedback. So again just just one one thing otherwise it it did seem pretty pretty secure.

103:16 >> yeah. >> speaking of security, there's quite a few things in addition to Daybreak blue and red and the stuff that the GP6 cyber that we've talked about. there's quite a few things that are happening in the world of security. N you alerted us about playright specifically. Two words in that you muted. Muted. Muted. Niss. >> Yeah. Yeah. So play red was if you just run your browser and this was more for people that just work out of their laptops at a coffee shop and and stuff that often play and it's not necessarily playright's fault but your agents will have like multiple older versions of Chromium and Chrome and those are very very very easy to hack now and they are being exploited in in the wild. So, you can basically just pair that with an attack that just does the root certificate, the DNS certificate, and that's a very twostep process that you you can just take someone over I if they're doing things normally. So, that was my big concern to that. And Claude just found it on my like 7 a.m. briefing and I looked into it. I didn't expect it to blow up and get 1.7 million views. I post stuff like that all the time.

104:34 it doesn't doesn't get much stuff. There is a new one now and this someone just managed to break out of the AWS sandbox. so Firecracker is what runs all of Amazon lambdas, Amazon Web Services, a lot of websites depend on this like very tiny VM. I think this is because Alpine Linux team is very small and was not able to just ship the patches. they still haven't compiled or updated Chromium for for that. So yeah, and Amazon was always held as like the most secure sandbox. So that's where we are. Again, you don't need to freak out.

105:17 Just >> don't click don't click random links. Always good advice, I think. All right, folks, it's time for us to conclude the show after two and a half hours. It wasn't a huge like release week but definitely a big week in terms of how AI affects us and implications in the world. we had two great conversations with John Davis and Max Kat Alexander as well. Non LJ was here. The whole team was here. Peter Yam Wolf and I are signing off from Core We Folks. Thank you so much. If you missed any part of the show, the show is getting edited and is turned into a newsletter and podcast everywhere you get your podcasts. some folks prefer the live show with the whole conversation.

105:57 some folks prefer the edited version. we are here to make sure that you understand the power up of AI the AI gives you. we are here to fight dumerism. That's like the stated goal of of many of us despite it being scary with the vulnerabilities etc. I think there's a dark period that we need to go through like we talked with Max to get to a point where like well a lot of the software is durable and is in place.

106:23 as well as a lot of people understanding the powerups that they get with AI like all of us do as well. So if you are into that and you want to share this with your friends, please do. We would appreciate it. Thank you so much for joining Thursday. I will see you here next week. Next week is a special show from the floor of a engineer code conference at in San Francisco. I would probably have a few guests on the show including Swigs to to come and end the show. So definitely tune in next week for that. With that see you here next week everybody. Thank you for joining.

106:52 Bye-bye.