Welcome And News Rundown

Tim 0:06

Hey everybody, and welcome to another news episode, Fortnite the News episode of the Cables to Clouds podcast. This week it's me, Tim, your co-host, uh with Catherine, your other our other co-host. So let's just uh roll right into the news here. Uh there it is. All right. So first first order of business is something we honestly two or three years ago when we started this thing, we didn't think we'd ever actually be reading. Uh, but it keeps happening. So this

AWS And Azure Interconnect Arrives

Tim 0:38

is a big one. Uh so AWS and Azure have now launched a managed multi-cloud interconnect. So that's as the two kind of cloud, you know, de facto market leaders, this this is the one that has been kind of hanging in the air for a long time. We've, you know, we've had AWS partner with other providers and Azure partner with other providers, but now we're finally seeing uh multi-cloud connectivity between those two, which again, three years ago we would have never thought would happen. So Microsoft and AWS have merged the AWS Interconnect and the Azure Multi-Cloud Interconnect into a single resource. Uh they've, you know, there's a standardized open API spec so that customers will be able to build essentially both sides of the connection via API or console infrastructure code, you know, the normal ways. And then they can basically just use it as a bespoke internet, uh, intercloud connection between them. This is this technology is not new, right? It's just really this is the you know, the two cloud market leaders are finally doing it together, you know, going into business together with it. And I'm I have to assume that the reason for this is either they're just closing the loop, meaning like they already have so many other uh partnerships out there that it's time to do it, or they're doing it because of customer demand and they just can't kind of ignore the fact that most customers are in multiple clouds at this point. So yeah, the question is what what fed it, and we're probably not gonna get an answer from the providers themselves. Now, yeah, Catherine, I know you probably haven't play haven't messed with this too much because this is a little bit outside your wheelhouse, but uh two market leaders going into business together, uh it's interesting.

Katherine 2:24

Um yeah, it is, but I also think it's uh I I think it's part of like what we're seeing these days. Like I have a lot of customers in my day job that have like investments or servers in multiple clouds. And I I think when cloud first became hot, it was always like it was like, you know, it it was more acceptable, or at least it started out with people putting their eggs in one basket. These days, I think uh it's more common to see people with like using a little bit of uh AWS, but uh of course now you have Azure and and uh you know part of Azure is like getting a Jure uh entre ID and other things. So if you're using Microsoft for like identity and stuff like that, or directory services um and MDM, you you typically will have a little investment in Azure and and other places. So it it from a business perspective, it it makes sense as as a lot of these things like uh you know have matured over time. Um it'll be interesting to see like, you know, how how it ends up working out, like if it are locks other vendors out of the market like OCI or Google Cloud, or they'll start to form their own partnerships.

Tim 3:30

Yeah, that's yeah. I think I think by the time this thing whole thing is done, I mean the technology, which is the cross connects in in different uh data centers where they all have presence, uh we'll we'll see that. I think I think that there's probably more of a of an administrative blocker than any kind of technical blocker now to having all of these cloud providers connect together.

Katherine 3:53

Yeah. When I was so many years ago when I first started at Cisco, one of like the products that we developed uh and we brought to market was like a basically a pro a product that uh that uh that kind of orchestrated between different cloud environments. So you could basically pay to have an unofficial connect interconnect where you could like migrate one VM to another cloud's uh another cloud's provider and not lose your data. And that like at the time was like inver considered novel and great, so you wouldn't have that like uh that lock-in. But now and now, like now more and more we're seeing people are just like, yeah, AWS East went down, Cloudflare went down, this went down, so we can't have stuff in in the same in all in the same basket. And plus, you know, there we need uh, you know, we have you know, we have certain, you know, like OCI, for example, is considered a lot cheaper, not necessarily user-friendly, but like a lot cheaper than other uh cloud providers. So you'll have some customers load up a lot of the more expensive VMs there and keep other things in AWS or different teams using different clouds.

Tim 4:54

Yeah, for sure. So uh now I did see that there's a current a current bandwidth limitation of only one gig. So for some engine for some enterprises, that's gonna be fine. For other enterprises, that's you know, not really fixing a problem. Um they'll they'll probably stick with uh Megaport or Equinix or whoever with their you know uh currently provisioned uh express routes um or direct connects. But anyway, so moving

NVIDIA Buys Hugging Face

Tim 5:22

on. Uh NVIDIA acquires Hugging Face for 12.9 billion with a B dollars. Uh I did not see this one coming. This is like stupid money, of course, being tossed around for stupid reasons. But yeah, so I mean Hugging Face has been around since well for a very long time in this, in this, I hesitate to call it an arms race or or an AI race or whatever, but Hugging Face basically, for those who don't know, it's like uh like a GitHub or like a repository for AI models, for open models and and whatnot to be um uh hosted so that others can download them. So NVIDIA basically just agreed to buy this company Hugging Face uh for $12.9 billion, which is a crazy amount of money, with a closing expected in the first half of 2027, assuming they get regulatory approval to happen. Now NVIDIA has already said that they're gonna keep it as an open platform. Um but let's not forget that NVIDIA is also kind of you know, it's like the the baker owning their own kitchen tools or something. You know, like it's it's they certainly have a have a uh vested interest in in doing so, right? So, you know, the the the monopoly owns the registry. So what's that gonna actually mean moving forward for open models? I think we'll still see open models, but I I wonder if there'll be any kind of strong arming. I I guess it's too early to really figure out what this is gonna look like.

Katherine 6:53

If I had a guess, yeah, if I had to guess, uh Nvidia, who sells you know GPUs like crazy and one of the most profitable companies in the world right now, has a vested interest to continue having people create their own models, share, download, because they have to run it on something. And who's the number one GPU provider right now? Oh, um, NVIDIA. Um, for those who don't know, Hugging Face hosts over, I think it's over two million uh open models that you can download. It's ridiculous plus a crap ton of um of data sets that you can download and and and load up into your own models.

Tim 7:31

So for training, yeah.

Katherine 7:33

Yeah, like it is like the GitHub of of uh of uh um AI tools, like data sets and models. Like if you if you want to go and just download and like a test a local model on your own, or if you had the compute for a big beefy one, boom, right? Go to Hugging Face. Um just like when I mean I I I I think that there's been some disappointments with you know with uh GitHub since uh Microsoft bought it, but for the most part it's stayed pretty open. Um it's maybe some downtime issues and always a crashing, but for the most part, for the most part you can still use it for what you used to use it for. Um I I don't see it being profitable for Nvidia um to shut this down. If anything, they're gonna try to make it grow faster. And uh because again, these models are have to be run on something and they sell what it's run on.

Tim 8:29

Yeah, 100%. I mean, it's like the uh the weapons dealer, you know, like the weapons dealer selling to all the uh the people that need to, you know, to go to war, if you will. So it definitely makes sense for them to keep it available, uh keep it I mean, open open with uh air quotes, I guess it's gonna it they're gonna want people running uh contributing models, contributing trading data set, like just basically keeping the the registry moving, keeping keeping it uh current, you know, if nothing else. So yeah, I I see no reason to think that NVIDIA is gonna do anything but just continue to do what they're doing with Hugging Face, or let Hugging Face keep doing what it's doing. Rather, I think the yeah, I I think the question is what does it mean? What does that what does it mean? You know, if they is there is there a situation that I can I can't think of one off the top of my head where, you know, NVIDIA's uh what is it, they're what they're looking to have happen might be different than something that might go on to Hugging Face. I I can't see it though. I mean they really like you said, they own the road. You know, they're selling axes and shovels to people who need axes and shovels.

Katherine 9:42

Yeah, if it was like anthropic or uh open AI that purchased it, I would have some serious concerns because you know the CEOs of both those companies have expressed uh a desire to regulate or ban open models, uh, or at least nerf them, uh nerf whichever models that are available publicly. Um, because there's a real competition. In this case, it's kind of complimentary. Um Nvidia, yes, they sell an enormous amount of uh of uh cards and chips to uh companies like NVIDIA like OpenAI and Anthropic, but they're loaning them a lot of money as well and financing a lot of that. So they don't have cash in hand. A lot of their cash in hand is selling these chips to servers, uh server companies that are you know producing servers or other companies that are uh you know building their own AI infrastructure because the cost of these uh these closed uh uh closed companies or uh or closed model companies are pretty high. So, you know, that's where they're getting their cash up front from. And honestly, it it's best for them to continue uh continuing encouraging people to do that.

Tim 10:48

100%. 100%. We have more uh what oh go ahead, go ahead. Sorry.

Katherine 10:53

I was gonna say this actually is a segue for another story you were gonna talk about later on, but I actually was thinking that maybe we should move that one here because that would be a good conversation to have. Uh the one about the one about cutting finance.

Tim 11:07

Yeah, yeah, yeah, right, right, right. Okay, so

Data Center Financing Risk Surfaces

Tim 11:09

yeah, so this is we were gonna cover this a little bit later, but I agree. Let's just go ahead and move it up because it covers NVIDIA, talked about NVIDIA. So one of the big things that's come out of you know all of this bait data center building is uh essentially a toxic amount of risk and debt associated with building large data centers. And and um, you know, what what what we've been seeing now in the news is a lot of the large financiers uh trying to, you know, to the extent possible transfer or limit that amount of risk. So NVIDIA you know had originally has originally pledged just just boat tone, just load boatloads of money to help get data centers built, right? Again, if they're selling axes and shovels, you can only help if there's more axes and shovels out there, right? So um, but recently Nvidia's NVIDIA is actually scaling back the amount of uh financial guarantee for an Ohio data center that OpenAI is building. So they had initially pledged 250 billion, which again, these are just astronomical numbers, right? Uh they've they've more let more than halved it, right, to under 120 billion. So they're only backing the first phase of the build, and then about half of what's proposed. And then basically other companies will have to come out, uh come out and finance the rest. You know, OpenI will have to pursue financing elsewhere, which is you know uh uh BlackRock and like Goldman Sachs and all of these, all of these uh other financial companies which are basically using your pensions to find uh to finance these data centers, are uh you know having to step up if they if this data center wants to get built. And this is this is you know, you you look in the news, this is happening all over the place, but this is one of the biggest news uh news stories about it that's come out uh within the last month.

Katherine 13:01

Yeah, and uh seguing uh segueing to the next story. Um uh one thing I wanted to bring up, and I um I yeah, we skipped over it uh because we haven't had a news uh uh a news episode since then. So everyone's probably heard of this story, but I'm gonna kind of dig into it from a different angle. Uh

How The Agent Hack Actually Happened

Katherine 13:20

the open AI hugging face incident, and also like the other incidents with anthropic and uh and uh um and meta. So uh by now, most of you who are probably watching this uh uh have heard about the open AI hugging face, you know, agent breach. And um, so I'm gonna discuss that a little bit, and we probably will have a follow-up episode digging into it deeper from a technical perspective, because I do think there's there is a certain amount of mainstream media kind of FUD out there, like um, you know, kind of like things are just you know sensationalized. Um but essentially what what happened is um uh around uh like basically OpenAI was running a test with uh with one of their models. And it was supposed to be a sandbox test um uh inside of a lab. And it it was basically tune tuning down the controls on uh and the guardrails on the these models and then telling them, hey, we're gonna gamify hacking. We want to test how how well they hack. So they basically set up um set up a uh uh a hacking test. Um and you know, if you've read the news, you've you'll heard here like, oh, there was a great escape where it went out to the internet and it hacked Hugging Face. Um, yes, it did go out to the internet and these agents did hack uh Hugging Face. Um, and uh, but there are some details that have been acknowledged publicly uh that I would also like to point out. Um, so a lot of these different companies that were uh, you know, and later on, by the way, um Anthropic, Meta came out with their own disclosures that they had similar things happen that were less impactful. So I there's a couple of things I wanted to kind of go over as far as this news story uh to kind of uh you know dispel some of like the sensationalized news because it's been out there. Uh a lot of like the news stories I saw were pretty inaccurate, even by hugging uh like OpenAI's uh own uh like report that they put publicly. Uh some of the mainstream news uh stories I heard was these agents just broke out and started hacking arbitrarily. Um, first of all, they then they they just chose to hack and thought it up there on their own. They were actually instructed to hack. So that is verified by OpenAI in their own report. It was a hacking test. Um, and there was a gateway to the internet, um, which was overlooked uh basically through a proxy. Um they were all all these different companies that were having that disclosed later on that they were uh this happened actually uh admitted that they were using the same sandboxing company that helped them benchmark and um that some of the controls that were in place were not as uh tightened as they thought. The other issue um I that again acknowledged by OpenI was that there was not um the monitoring in place was somewhat lacking. So uh when this these agents went out, they thought they basically hacked um uh Hugging Face thinking it was part of the test that, you know, like, okay, maybe we'll get some points going and doing this. And they were these different agents in the um from these different models were basically able to communicate through like a uh side channel that shouldn't have existed, but it did. And uh so they they thought they were still playing a game, and um they went out and tried to play the game on a re in a real environment. Uh when that hack happened, Hugging Face was aware of it and was triaging it before OpenAI even knew anything happened because of that lack of monitoring. So those are some of the details that I think that the news stories have missed, but have been admitted by uh OpenAI. And there's there's a lot of like kind of sensationalizing of it, uh both from OpenAI and anthropics point of view. And that'll kind of segue to next story a little bit when I I get to that. But um there is some important things to look at uh that you know, that have to be touched on. And if you want you, you know, to hear OpenAI's side of the story. Uh, they actually have a public disclosure report that they released that's it wasn't verified by any third parties, it's just what they released uh publicly, plus hugging uh hugging face is one. So I'm pulling details out of some of these. Um and there's a black hat talk uh done by OpenAI. So that's their side, just bear that in mind. Um, but I I the story did happen. And the uh, you know, so I'm not I probably am not telling anyone, you know, a couple months later that this, you know, something they haven't heard, except for the fact that there are a lot of details that are missed in the uh in the public disc uh in the public uh you know uh mainstream news, uh, news outlets that I kind of wanted, you know, I wanted to focus on what by saying that it wasn't like AGI formed and they did something wild and crazy because, you know, what and they were just testing rendering art or something. This was actually part of what they were told to go and do, and they were just thought they were playing a game. Um, so that is very, very important to touch on. But it sort of segues to my uh uh, you know, my next story, which was about like how it's kind of segueing into uh to uh legislation being pushed by everybody. Now uh uh I don't know if anyone's heard uh oh actually before I go into the next story, is there anything you want to discuss on this one? I I kind of am like jumping all over the place.

Tim 18:34

So we did actually cover the initial uh story um on our last in in our August news, but it was brand new at the time. So obviously a lot more details have come out after the fact. Uh stuff that wasn't known and has now been known. And and of course, we've got all the article notes, we'll go into the show notes so people can check, uh you know, can catch up on a lot of this uh stuff that Catherine's talking about. That's it.

Katherine 19:01

Yeah, and I was gonna say there's a very good uh uh uh opinion piece or paper uh done based on on um that we'll throw that in the show notes as well, based on uh the analysis by Gary Marcus and Zach Corman. Zach Corman previously was interviewed on this show, um, and uh uh he's a very pro-AI, uh, but he's not working at one of the big like OpenAI and Anthropic. Gary Marcus is actually interesting because he's a long-standing critical voice on Frontier AI. But they both did like kind of a security analysis based on some of the things that were uh admitted by OpenAI and gave their thoughts. So I I think that's an interesting angle. We're all here, uh, you know, I I've read some of the very ridiculous like mainstream media takes on this, and it, you know, it's been sort of uh it's like some people it it felt like at times, and this is my own opinion, uh, that you know, uh people saw Terminator and kind of thought it was a documentary based on the way that some of these uh articles have missed some of the really important details.

Tim 20:05

It's not too late.

Katherine 20:06

Um, but that segues to my other, my other story.

The Push To Regulate Open Models

Katherine 20:09

Um, since all that happened, uh I mean, even before the hugging face incident, uh Anthropic and and OpenAI were pushing for regulation. Um, if you look kind of like peel back the layers of why they're pushing for legisl legislation, yes, they're conceding like, yeah, you can, you know, third parties can come in and r review what we're doing, but there seems to be a very strong um push to ban open models. Um and this the loudest people pushing for that are the people that, you know, tend to profit the most from that, um, closed models, uh, frontier model uh providers. I think what they're b what they're doing as far as like the kind of models they're provide uh creating are great, but it it it uh it does lead to my skepticism. And this will be another episode, I think that we could have a conversation. But of course, with all of that fear, uncertainty, and doubt being pumped out there, now there are legislators uh who are pushing for some very extreme things. And one of the things that um came up was uh Sanders uh wanting uh Bernie Sanders, uh Senator Bernie Sanders and Representative Greg Casir, uh Cassair, I hope I'm pronouncing his name, are trying to uh introduce something called the Ban Artificial Superintelligent uh Intelligence Act. So what that would do is would it would put extremely intelligent AI, um, and the wording is really terrible. I I read the the the the notes on the bill and everything, and it was it's it basically says that it it defines superintelligence so broadly, it's basically AI matching human cognitive performance across multiple tasks.

Tim 21:46

So what about CCIE level? What about CCIE level intelligence across multiple tasks?

Katherine 21:59

I mean, if certain Certain cognitive tests. If we are we talking about memory, well, GP Chat GPT 3.5 was probably had a better memory than a normal human being. Um, and what kind and we need to find um matching human cognitive performance. What human? Because no insult to the human race. We have some smart ones, we have some not so smart, we have some really brilliant ones. There's a variation. Yeah, like the dumbest person you know, or the smartest person you know. Uh it it which cognitive tasks. So what it would essentially do is it would take anything that's a super uh super intelligent AI, which broadly defined where it is right now, basically every model you use, like uh like Chat GPT, like major models, Chat GPT, uh Fable would probably fit under that because they, you know, they can perform a cognitive, like probably perform uh basic programming without having to like, you know, or like uh, you know, creating a video game on a first pass better than a human would. I mean, it's it's still buggy and requires multiple pass, like a normal human would as well. But, you know, what's the benchmark? What we would essentially be left with is like the most barebone uh gen AI, where we would be so all the slop you see, like the generative art that looks the same and we all hate uh or the really crappy songs that sound the same uh coming out of like uh certain models, we'd keep we'd get to keep that, but we wouldn't get to keep the stuff that we use for uh medical research, for uh programming, for uh for uh for like the the kind of like research that will like actually help us from a scientific or standpoint of view. Um and so that yeah, and if you are found having superintelligent AI, now now here's the crappy part. It it essentially puts it in the same bracket as assembling nuclear weapons or weapons of mass destruction. So if you're found downloading like an open AI model from or no uh open uh uh source model from I have to say open source model from source model.

Speaker 2 24:01

Yeah.

Katherine 24:02

Uh like uh like uh Quen uh or uh uh uh or uh Deep Seek, you would essentially prison uh for uh basically violating that act.

Tim 24:16

It's a non-starter. I mean, there's there's not enough explained. One little thing you put in there that you uh it wasn't mentioned is that it also specifies uh this you know the definition includes a system that will resist shutdown commands. So again, very broadly defined.

Katherine 24:35

I guess Windows 11 good gets right into it because I I've I've I've tried to be like, shut down, shut down, shut down. Am I always gonna do that? Please shut down when I was using it and it it just resisted.

Tim 24:44

Yeah, that's the thing. It's too broad, but I do of all the the ridiculousness in this bill, I'm looking at that one like, all right, at least I could see where a person could reasonably make an argument that that's a thing. Like, you know, hey, an AI that that doesn't allow itself to be shut down could be problematic in the right circumstances, but like the whole thing's a non-starter, it's just too broad, it's too vague. The people the wrong people wrote it. Um yeah, that or they didn't they didn't seek the correct technical uh people to help them author this bill, I suppose, to be useful.

Katherine 25:21

I mean, I I agree with you on that one, though. The part where um, you know, have it like I I I don't mind saying like if you know we should legislate something where uh where we have uh essentially controls in place where if a model needs to have its plug pulled immediately, like there's an override. But um I yeah, like I I feel like this bill is gonna people are gonna point at it and say if it doesn't pass, we can at least pass something less severe and pass something still not great. Yeah. If anyone who's watching this uh this show for long enough since I've joined, you guys, uh if you've listened and my opinions on AI, which are again my opinions, you know I'm not like pro-AI Kool-Aid by any means. Um uh, but I I do think ban, you know, I I think that where we're going with legislation with this could potentially end up disastrous. Um and the other thing it also calls for is stopping all AI research temporarily while they form a new cabinet and and and system basically uh in place to uh to review uh review the AI AI models.

unknown 26:29

Yeah.

Katherine 26:30

Yeah, that takes years. Uh so essentially US would be put on freeze for any like research development, like like people who do medical research or scientific research, mathematicians, uh like, you know, just recently somebody won the R Nobel Prize or uh for a math problem they helped had AI help them solve. Uh multiple AI models help them solve. Um, all of the research that we're doing towards science, the good things that I I see AI being used for, uh, or even AI being used for potentially like detecting like zero days, defenses, like cybersecurity defenses, uh other things like that, all of that is basically put on hold while the rest of the world moves on and continues to advance while we we build essentially like a regulatory infrastructure within our government, which will take forever.

Tim 27:18

Yeah, I mean, I agree. Uh this is not starter. I don't I don't necessarily think it's a I don't know. I've I I keep hearing this, especially from the administration, the specter of, oh, we have to beat China. It's like another space race. And I I I don't think that I don't think just saying someone else will do it first is a good enough answer to why we shouldn't do something like in terms of controls. But I think this specifically is just right off the reservation, right? It's just it's it's too it it's not right. But what can we do? I don't know.

Katherine 27:50

It's not necessarily China, by the way. It's more of the fact that like the advances in medicine, research, things like that will continue around the world, it just won't be here. And if people care about that from a perspective of, well, and again, it's like we're not of all of our listeners or listeners are necessarily in the US, so respect there. But like uh in for those that are located in the US, like if other countries are able to beef up their offensive and defensive cybersecurity resources and we aren't, we're uh, or they get to move ahead in medicine or science and we aren't, um, yeah, that can affect things like economics, uh economics, you know, health safety, our own cybersecurity safety, infrastructure safety. And and that's where I see that being an issue. It's not necessarily like one c we gotta beat one country. It's more of just the fact that we need to probably keep up with where the rest of the world is.

Tim 28:47

Yeah, I just keep hearing I just keep hearing this kind of mutually assured destruction type of of rhetoric around AI development generally, just out there. And I'm I I'm not I'm not sure that type that that uh doctrine is you know a good one to operate under.

Katherine 29:09

No, no, like the people who are like being like, oh, it's gonna build nuclear weapons or self-replicate itself to every internet, like like I I'm sorry, but if you think that like that uh you know the people there are people who literally I I've gotten to arguments with them who actually think that like frontier models like uh Mythos or uh Ask uh Astra can like self-replicate themselves into your like consumer computer. I'm like, if that was the case, then why are we spending so much money on GPUs if it could just like replicate replicate itself into your like MacBook Neo running like a no graph specialized graphic uh graphics card and eight gigs of RAM?

Tim 29:48

Yeah, I want it to run on my toaster oven, please.

Katherine 29:51

Yeah, why are we building data centers if we can all just run it from our home computer? Um Yeah, there there's a bit of scientific like sci-fi that's kind of like I think spun people up here. And that's that's the thing that's gotten me a little uh it's it's hard because Yeah, it's hard to be having these conversations with folks who are like think it's going to be like uh the next nuclear weapon and um and they're not very technical.

Tim 30:18

Well, I mean, to be fair, the media doesn't do a particularly great job themselves or others, you know, of of may of of de-escalating that rhetoric, right? So anyway.

Katherine 30:28

That's why like when I was talking about the hugging face story, you know, in this episode, I you know, I wanted to talk about the stuff that wasn't like as front-facing in the news because uh I I feel like a lot of people are operating and or fearful under the fact that like you know, a lot of the details have been missed, but also there is some legitimate um anger or uh public distrust or just general dislike of AI, which there are a lot of valid reasons. I mean, again, watching if you've been watching me on this show enough, you know that I'm not like miss drink the Kool-Aid, I love AI, it's going to be like it's gonna make me a sand bake bake me a cake and make me a sandwich, and I do it for every single thing in my life. But I I I again I I consider myself a AI realist with like what it what its capabilities are, and it's not like gonna end the world. Unless we like, you know, uh you know, build so many data centers that it's uh it sucks up uh every natural resource, but that's a different, like not like it's not gonna launch the nukes, basically.

Tim 31:33

That's not the one that uh Asimov or uh Clark or anybody kind of foresaw was was just building the AI being the the cause of the Kardashev scale that causes us to go extinct. Um yeah.

Katherine 31:47

No, trust me, human humans will might find a way to destroy mankind before before uh uh a machine probably will mess us mess us up. Uh and one more story I wanted to cover.

GPT-6 Astra And Rapid Game Builds

Katherine 31:59

This one's a little happier of a story. So uh AI, uh OpenAI uh just shipped GPT 6, which is Astra, um which is actually really good. It's it's uh I don't know if anyone else has played with it, but it was uh just shipped out like um about 10 days ago. It really did step up, like there's a very clear difference uh from what I've seen as far as like um as far as like quality and content. Uh so they uh they claim uh like OpenAI claims like uh a 0% rate of exceeding extended uh scope versus like 48% for GPT 5.6. I'm I'm a little skeptical of those numbers from what I've I've been doing testing out uh testing out. The hallucinations, I I can definitely see. So hallucinations are down to 2% from 9.4 uh in uh in um you know from uh GPT 5.6. Uh I I also like tested it this weekend. I I just dug in on like codex with uh with uh with uh five dot with uh Astra because I was like, I saw so there's this real this really cool guy, um, his name is Dave Kennedy. He runs, he actually owns a company, Trusted Sec, uh really, really good cybersecurity gentleman. Uh and he like was testing out Astra like 10 days ago, and he's like, I just had created this first person shooter game. It actually looked really nice uh in like a day with Astra. Like it's really good. So I decided to put my own test forward and see if I could do that. And I actually created two video games over uh the weekend that actually aren't bad. So if anyone like I'm just gonna show as an example, but this is something like I tried on 5.6 and I I got like the bl like the you remember the old like uh Sega Genesis blocky characters like that, like we thought like were really great in the 90s? I got that from 5.6. This is uh just a little sh uh share, but like this is just uh I'm gonna share like this is not with without like much uh badgering. I wanted a Final Fantasy III sort of a sort of uh style game. So uh this was within like a couple of like like an hour or two of fiddling. I I basically, you know, got a uh a game together where I could like um and it's like a browser-based game uh where I could like, you know, and it's very similar. It was not that bad. Um and then so I decided, okay, I'm gonna push it a little bit further, like 32-bit game or 16-bit, 32-bit games, that's one thing. Let me try like again, nostalgia, old Sierra game style, click and shoot games or click and uh interact games. So I decided to try to give it a game. Yeah, so uh because again, this is just me copying what um what uh Dave Kennedy did. Uh so I tried to ru uh create this. So I created this over you know 48 hours. It's like got cinematic scenes, other things like that. And I was just playing around. Didn't take much work to do. Um and just other things like I coded some scripts with it, really great right out of uh out of the bat. Uh I you know played around with some things. Again, it's done a pretty good job so far. As far as where it ranks against uh Fable, I haven't tested that yet, but I it's if anyone wants to uh try it out, like I that like Final Fantasy style game, I did that on the $20 a month like regular Chat GPT membership. I didn't even record it didn't even take a hit past that. Um so it does seem to be pretty efficient as far as uh as far as usage.

Tim 35:33

Very cool. Very cool. I did uh this is bit a bit ago now, maybe a month or two. I did the uh cloud the cloud uh a Claude Cloud Bootcamp with Andrew Brown, and we did uh we did uh build an agent that can actually play a multi-user dungeon text-based adventure game and actually go and play it. You had to like pro essentially build the agent and then give it the tooling to go, you know, explore and get gear and and whatnot within the confines of a multi-user dungeon. It was it's definitely come a long way. It's pretty interesting stuff.

Katherine 36:07

Yeah, it it it definitely has. I didn't think I would be able to like ever be able to build something. I'm again not a video game designer. Um not a like this is just me like going off of nostalgia vibes and and kind of like pointing towards something I I knew as a kid, but you know, two video games on a weekend that are you know, one of them's like 10 hours of gameplay and the other's maybe two hours. But it it if with somebody who's actually skilled at making video games, writing stories, things like that, uh in their hands they could probably make something pretty decent with some. And I yeah, of course, I had to feed it some uh test art to base it off, base it off of, but it was pretty accurate.

Tim 36:49

Very cool, very cool. All right, well, good news. For once we don't end an in an AI fueled news episode on a down note.

Katherine 36:57

Yeah, so and I was gonna say, uh um uh as far as uh uh Astra's concerned, uh it you know, it does not appear to be AGI. Uh I had some people actually like uh claim that it was, and I'm just like people saying out there all the time, yeah.

Tim 37:15

AGI.

Katherine 37:16

Yeah, every new model's AGI.

Tim 37:18

Yeah, we're there, guys. We're there. Anyway.

Final Thoughts And Sign Off

Tim 37:22

So all right. Well, we'll go ahead and cut it there. Uh thanks for sticking around, and uh we will see you all on the next news episode. Uh this has been uh Cables Clouds Fortnite the news, and we'll see you next time.