Ep. 82: What Can OpenAI’s Astra and GPT Image 2.5 Actually Do?

0:00:00What’s on Fast Hours this week?

Play from 0:00:01 8:88:88 0:00:01 What's up everybody? Fast Hours Podcast episode 82. Rory Flynn, Drew Brucker. Happy Friday everybody. What's going on Rory? Talk to me man.

Play from 0:00:10 8:88:88 0:00:10 Uh another nice clear day here. Very reminiscent of uh actually not as reminiscent as as the 911 25 years ago. It's wild that it's 25 years.

Play from 0:00:22 8:88:88 0:00:22 I've just been deep in the weeds on stuff today, but I I did see there was like some event taking place in New York. uh I guess uh this morning maybe or something like that that they were hosting. Is that like a yearly thing that they do?

Play from 0:00:35 8:88:88 0:00:35 They do it every year. It's uh you know it's the there's been some some cooler things and I say that lightly this year like I found this account. Hold on. Might as well shout this out because I think it's probably one of the coolest things I've seen contentwise. I'll pull this up. I think it's called

Play from 0:00:52 8:88:88 0:00:52 2001 live. This account Yes. Yes, dude. I'm following.

Play from 0:00:59 8:88:88 0:00:59 They've been doing like every six minutes like real time reports as if like it like X was available in 2001. So like what would have been happening but like also

Play from 0:01:09 8:88:88 0:01:09 this is a this is a new trend I'm seeing on here by the way. It's it's cool as hell. Like I I don't like the content itself like 911 not cool but like the way that you it's been like brought back like I was following this morning like before everything happened. Um because I remember the times, you know, my that was uh obviously being growing up in the New York area like you remember when it all sort of went down. But it's like interesting to see it like beforehand like the today they have like clips from the Today Show, you know, like of people standing outside in in Time Square like before it all happened and it's like, "Oh, all right." You know, like we're having a great day. It's beautiful morning. Blah blah blah. It's after, right?

Play from 0:01:48 8:88:88 0:01:48 Where Okay. Where where were you when this was taking place? So 2001, where was Rory where was little Rory Flynn at during this?

Play from 0:01:59 8:88:88 0:01:59 So I've considered doing like a like an AI short version of this because it's it's a very unique sort of circumstance. Like my parents both worked in New York for 30 plus years. I lived in New Jersey right over the for anyone who's not familiar. It's you take the bridge, you're in Manhattan. Um so I live like mile away from that. Yeah. So

Play from 0:02:19 8:88:88 0:02:19 Okay. Everyone who lives in in my area works in New York, right? Um, everyone's parents basically. And we were in I was in seventh grade, I think. So, that morning was our, oddly enough, outdoor education. So, we were all getting on a bus to go to the woods for 3 days or 2 days. I can't remember how long it was, but we we had to leave at like 6 a.m. because we had to be there by 8 to start whatever activity. So, we left before anything happened. Um, you know, we're all on the bus. We're all, you know, going there. you get get there and then like you know days goes on as normal.

Play from 0:02:52 8:88:88 0:02:52 There's no cell phones, you know, we don't have any connection to the outside world. So we're like in this isolated little bubble while there's chaos, you know, in the world, right? And we're just like, you know, uh going and doing like nature trails and like uh you know, this is how you know photo shoot this just works like stuff like that, right? And um you know it started getting weird like it start to see like the adults there um you know just kind of like get distant and like not really pay attention. They started pulling kids out like just not telling anyone like why kids are getting pulled out. We stayed there.

Play from 0:03:25 8:88:88 0:03:25 Okay. We stayed there for two days and on the way home I'll never forget this. We were being such [ __ ] on the bus as seventh grade boys will be and like one of the teachers snapped on us. It was really uncharacteristic and we were just like the [ __ ] was that about? Um, and we walked, this is, this is the scene that like still sticks in my head is like we get off the bus, right? And uh, you know, there's like a line the lineup of cars of everyone's parents waiting for us. And normally it'd be like excited, whatever. Just blank stairs, silence, and like stonefaced. It was like zombies, like a it was like a weird movie, right? Like no no expression from anyone. And uh, it's like the [ __ ] is going on here? This is so weird. like you everyone just kind of was like laughing as he got off the bus and it was like kind of like okay like what you

Play from 0:04:11 8:88:88 0:04:11 know like it was it was like that like why is everyone so yeah everybody's so serious. Yeah. Yeah. And I went home and watched CNN for like

Play from 0:04:20 8:88:88 0:04:20 12 hours straight. Okay. I was like so I missed it all and it was weird not knowing about that living here and being dude that is wild days. That is that's so crazy to me.

Play from 0:04:33 8:88:88 0:04:33 Yeah. It's a it's an insane like bubble and then you feel the guilt for like not knowing it in real time and you know people and you know a lot of people that

Play from 0:04:41 8:88:88 0:04:41 bro I had a like I had a totally different experience than that. I I mean so what I think I'm maybe three years older than you, right? Okay. I want to say I was either a freshman and sophomore in high school. But um for me, I was in a it's like my first class of the day and I think it was like a typing class of all things.

Play from 0:05:05 8:88:88 0:05:05 So we were all so we had computers in that which normally right like any other room there's no computer in that room at that time in 2001 and that was really the only class that I'm thinking about where you where you had the connectivity to like the internet. So, it was like we were in the middle of like a typing class. I think that it just started or something like that or it was just like getting underway and I don't remember the exact moment but I was in that room when I found out and I I think I was even like did like a little Yahoo little little search or like went to Yahoo or something and saw some some initial reporting and this was during the first

Play from 0:05:45 8:88:88 0:05:45 like the first thing that happened and I don't want to like get too many details here because I don't want to get flagged but um so then it was like oh something's was going on and then like everybody was already kind of like this is crazy and then the second happened and it was just like yo what is actually happening and then I think by halfway through that day or even in the latter part of that day like you know they had those one of those roller TVs they brought in the room and we were actually watching news live.

Play from 0:06:14 8:88:88 0:06:14 Yeah. um in school which at the now that I'm thinking about like the fact that I had the computer in the classroom at the time and then we were like watching the news at the time like it was a very memorable moment for me. I mean and obviously like a lot of people remember exactly where they were what they were doing which is interesting in itself because it's like do you have to have a traumatic experience you know happen in order to like really burn something in your memory of like exactly what was happening? It's like you you have some of those positive moments too, right? But it's like what is actually the mechanism that makes that happen because I would like to create more of those uh proactively and actually have a choice between what I'm choosing um to remember exactly like the senses that were taking place during this thing.

0:06:35Why do traumatic moments burn into memory

Play from 0:06:59 8:88:88 0:06:59 But yeah, it was that's W. Dude, you're okay. That is crazy. It's a weird one, right? It is really really crazy.

Play from 0:07:06 8:88:88 0:07:06 So here's another thing. I saw a tweet that went mega viral and this kid was getting blasted. Gen Z um you know tweeted something like along the lines of you know I don't know why millennials, boomers, Gen X like romanticize this day like and whatever. And it's just like

Play from 0:07:24 8:88:88 0:07:24 if you don't if you didn't experience what it was like you couldn't possibly because it wasn't being a parent. You just keep like you understand the concept but until that's actually you just can't connect to it on the same level. But you you knew like and someone made a good point in the comments section. It was like you just don't you don't know what it was like before like it was it was better. Like think thinking about this, right? In what world today would a seventh grader for 2 days not know that that that happened because they have a phone? We never so we couldn't have known, right? Like we were living in our own little bubble like ignorance is bliss, right? And then

0:07:42What was different before social media

Play from 0:08:02 8:88:88 0:08:02 but like you know I remember getting on planes after that and it was every time you thought you were like what are we doing here? you know, like is this is this cool? And then like it went from being like you could have people walk you to the gate in an airport to like we're getting strip searched at a gate and still

Play from 0:08:17 8:88:88 0:08:17 to this day like I'm you know searched you know it's 25 years later so like that changed. I feel like all like the silliness and like sort of confidence that we had building up up until that point like sort of like it kind of like dropped off a cliff and it was just you know it was it was cool though. I I do remember being around the area and you know we're people aren't familiar with the New York New Jersey area. It's a lot of like tough prideful people like a big pride sort of like place. Um you everyone's got you got a lot of Italian Irish heritage, a lot of firefighters, uh a lot of you know uh police, you know, especially my family, military, things like that. And uh you know so it's just like everyone sort of like readjusted focus and it was like we

Play from 0:09:02 8:88:88 0:09:02 all had one clear vision. It was very weird at that moment. It was like everyone was sort of like in lock step like you'd see someone walking down the street felt like everyone wasn't saying anything but just like a head nod like I like you like like we're on the same team right now you know just it's that's what someone couldn't

Play from 0:09:18 8:88:88 0:09:18 think of today because everything is so crazy political. Everything's polarized and and that's that's a really good point too, right? cuz you didn't have social media yet. So, there wasn't this like like super deep and wide

Play from 0:09:35 8:88:88 0:09:35 perspective that all these different people had and you were able to see and internalize

Play from 0:09:41 8:88:88 0:09:41 and also have that effect maybe your opinions or perspectives, right? You were in your own little community bubble of what you could see, what you could read in a newspaper, radio, TV. That's it. like no social media. So like

Play from 0:09:55 8:88:88 0:09:55 Yeah, man. I mean, we we sound old talking about this now, you know? But this is like the only day like I I technically like I don't want to say feel something, but I wake up every every year on this day like like something big happened and like I know it affected me and it affected everyone in this area, but it's like I always go I always go look at the I always go the photos. I watch the videos. I watch documentaries. And it's like a it's like a weird thing. It's just like it's like you don't want to for it's like you don't want to forget it. like you know it changed [ __ ] and it's and it's% you know I don't know if it that that's like torture for yourself but like also sort of like a

Play from 0:10:28 8:88:88 0:10:28 yeah I I think it's good like I I think um I think it's good to connect to those moments sentimentally like I I I mean I I always look to do that stuff. My wife's it's funny because my wife's not a sentimental person at all. I'm very sentimental. I look for

Play from 0:10:42 8:88:88 0:10:42 moments back in time. I connect to those moments. I could feel emotional about those moments. I remember how my emotion was in those moments. Like I I try to cherish all the little small things, you know, as best as I can. Like I'm I'm very much on that path. If I if I move out of this house,

Play from 0:10:59 8:88:88 0:10:59 I'll miss this house. Yeah. You know, like when I'm when I'm literally about to drive away, I'm going to be like, damn. Like I'm going to miss this house, man. Like

Play from 0:11:08 8:88:88 0:11:08 stuff happened here, right? Stuff happened here. Like my wife doesn't even think twice about, you know, it's just like, yo, it's so different. But I agree, man. I I think I like I think that's I think it's important to to kind of

Play from 0:11:19 8:88:88 0:11:19 it's just something that should I know there's probably a lot of people listening that might not have the same sort of visceral response to it or like I'm sure they remember it and you know people overseas and things like that. It was just like it was New York and you felt like as a New Yorker you felt personally attacked because we were we were we still are sort of like the whoever it is you come here you do whatever you want like you want to make it here you know go for it. it doesn't matter who you are, whatever. So, it felt, you know, you you always felt like in a very selfish way. It was like, if you're gonna attack New York, like we're the we're like the most tolerant and the like the most inclusive of the entire country, right? And it was like why us?

Play from 0:11:59 8:88:88 0:11:59 You know, it just felt it felt very like what, you know, what are we, you know, what are we doing here specifically? I think New York is New York is kind of like America and it's like a global city, American dream.

Play from 0:12:10 8:88:88 0:12:10 Yeah. We have all that stuff. Yeah, I feel like we have more we have more countries represented in this city than anywhere else in the world. Um, and it was just it was just felt it it still feels personal like it wasn't just the US like it was specifically New York and I think uh you know we all have a very weird feeling but everyone's got a story like you talk to anyone about that day and it's some crazy story right there's a lot of people that I work with that were you know walk through tunnels to get home walk through the tunnels to get home or

Play from 0:12:40 8:88:88 0:12:40 you know stuck on the trains when this happened or you know uh my my uncle was a pilot for or my My my aunt's brother was a uh you know was a pilot for American Airlines. My uncle worked at Caner Fitzgerald uh you know which was exactly where the plane hit. Um it it's in the in I think it was floor 86 or something like that. So it's it's just what my mom watched from across the river. She happened to be in an office in the Meadowlands which if anyone doesn't know the Meadowlands right across from the Manhattan skyline flat.

Play from 0:13:14 8:88:88 0:13:14 You can see everything. She sat there and watched the whole thing unfold. So it's like a you know very weird you know you think about it from like the like in an AI space right it's like you have multi multi-person multiplayer view of like this whole thing and everyone has like this whole story like built around this one thing um it's it's just wild I again sorry everyone for getting deep on this it just it's just a day that feels more significant for me. Yeah, it's worth calling out. 25th anniversary, too. So, I mean,

Play from 0:13:42 8:88:88 0:13:42 yeah, it's a big deal. That said, dude, [snorts] I I I think the talk over the last week has really been

0:13:48This week: Astra, GPT Image 2.5, a viral resignation

Play from 0:13:51 8:88:88 0:13:51 well, Open AI for sure. You know, there have been other little things in there which we'll get into as well. But, but obviously, we had Astra, which we were just kind of like, you know, hoping we had access last week. Weeks gone by, we've had access. I don't know how deep you've gone. I've been able to do like just a little bit in there. I'm sure you've been able to get to a lot more. We've got that. Also, OpenAI drops GPT image 2.5 out of out of nowhere. I don't know. I don't remember thinking about what was coming. And maybe that's just me being having a blind spot to what OpenAI's, you know, been doing lately on the image side, but like I I hadn't really heard any smoke about anything coming. So, that was like a nice pleasant surprise, too, because there's like there's a

Play from 0:14:37 8:88:88 0:14:37 legitimate step up there. And so, I think we'll get into that. Like, we can talk through that. We can talk through Astra kind of like what we've been doing so far. One little interesting thing that I think you and I both saw and I think is is really kind of an interesting conversation as a whole. Not Not that we're getting all doom and gloom here

Play from 0:14:57 8:88:88 0:14:57 on episode 82. Guys, life is good. It's It's Friday or it's the weekend when you're going to be listening to this. Life is good. Life is good. All right. But we did see this tweet which I think rled a lot of people up.

0:15:04Did an AI researcher really resign over ASI

Play from 0:15:13 8:88:88 0:15:13 There were a lot of there's a lot of twists and turns that go into this and I don't this is a perfect example of what is true. What's actually happening? And so this is why you question everything. But we saw this. This is the first thing that dropped. So this came out. When did this drop? September 8th. So, was that Tuesday maybe? All right.

Play from 0:15:34 8:88:88 0:15:34 165 million views. Yeah, that's okay. That's insane. Okay. I resigned from anthropic today. I spent the last three years doing

Play from 0:15:42 8:88:88 0:15:42 pre-training research at both AI and Anthropic. Neither company's acting responsibly. They're racing straight towards self-improving super intelligence and gambling with our lives. More thoughts below. Okay. So, let's, you know, click into that. Do don't underestimate the power of this technology. S soon be superhuman systems that can hack anything, revolutionize any field overnight, acquire real power and resources. We've all witnessed this progress blah blah blah. Uh the people building AI earnestly believe it could kill us all by the end of the decade.

Play from 0:16:12 8:88:88 0:16:12 Okay, cool. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in their press to sound sensible, but I hear the same people express fear blah blah blah. Okay. So anyway, goes into this basically is just, you know, lit his AI career on fire here to do this. So, you know, your first reaction is like, okay, this guy must be legit. And what actual, you know, reason would he have to do this other than to be truthful and just kind of tell us what's going on, right? Okay.

Play from 0:16:44 8:88:88 0:16:44 So, you know, you get something like this, of course, like flags raise. This is like a very sharable thing. And we know sharing within the algorithm of X is like one of the strongest things. So you know this was one of the those things that was just like oh share text this to Rory text this to like five friends text this my wife right like whatever it needs to happen.

Play from 0:17:03 8:88:88 0:17:03 Now I don't have all the chain of events here. I was trying to like quickly do this before, but then you get then I get something like this, which is basically like this guy with minimal followers, no previous account activity, goes to the Wall Street Journal, which publishes publishes an exclusive with quotes from him on the resignation 18 minutes before the post goes up. Planning was clearly done in advance and within hours there's ton, you know, tens of thousands of reposts and account has 100,000 followers. It's punchy, quotable, almost seems professionally written. The first three accounts to quote tweet it all did it within 15 minutes of the initial posting. Remember this account had basically zero engagement beforehand. So organic reach feels unlikely. According to those accounts, these were the people that did uh retweet those things.

Play from 0:17:49 8:88:88 0:17:49 General counsel at Encode AI, head of policy at the AI policy network, uh head of the AI futures project, all up and cominging AIDM policy advocacy nonprofits. Um, so this like goes into all detail of like, oh, our flags, you know, should be raised up here. This guy doesn't have much of a resume. You know, he got this blah blah blah. So, it's just like digging further into who this guy is and any possible rationale he has to do this. There are some like things mentioned here that I'm skipping over just so we don't get flagged here. All right.

Play from 0:18:22 8:88:88 0:18:22 So, I'm doing this with with a little finesse, but then you get like this guy went on TV talking about this and and so, you know, it's just like what is true and what's not, you know, because, you know, like with these things, even if something is true, people are going to come after and try to smear the name. They're going to do it.

0:18:40How to tell what is true in AI news

Play from 0:18:49 8:88:88 0:18:49 Happens all the time for people that are telling the truth. also happens all the time to people that aren't telling the truth. And this is the perfect psychosis and division that takes place in this era that we're in, guys. And it's like this sucks because this is essentially the perfect example of what we get all day every day and really are the root of our perspectives in a lot of way. If we don't challenge the thoughts on these things, it's easy not to challenge these thoughts and do the research and understand like the credibility factor and all that stuff. And even if you do, there's still not may not be a clear verdict, Rory. So, I just I I want you to chime in on this, but I just am raising this flag of this is both hilarious and like really alarming to me

Play from 0:19:37 8:88:88 0:19:37 for for totally different reasons, but like this is a just like one of those great examples of where we're at right now.

Play from 0:19:45 8:88:88 0:19:45 Yeah. I I'm I'm always that person. I think two things can be true. I think this guy could be telling the truth and I also think someone could take an opportunity to to leverage it how they need to, right? Like I think there's probably two things there. Like I've I I I don't see how he's lying though. Do you know what I mean? Like I I I I don't know how he got I don't know how this thing got so took off so quickly and it became like the calling card when like a million other people have said the same thing. Like we've talked about this on here before like that guy that said like we have five years left, you know? We did we did it with Ror, right? Guys like we have five years left. your your kids are never going to work, like whatever.

Play from 0:20:21 8:88:88 0:20:21 Like he's like, I'm not having kids. Like that whole thing. So, you know, I think two things can be true. It's it's true. It's like you can't at this point now, we've already passed the event horizon. We passed it when we when we released it to the consumers and it worked, right? It's like now it's if you stop, someone else won't. So, you have to stop. It's like it's like it's like an arms race. Feels like that, you know? It feels like the same exact thing. I know we talked about this like

0:20:42Why quitting your AI job can't stop superintelligence

Play from 0:20:48 8:88:88 0:20:48 episodes ago where it's like you're just going to have to build it to have it. So, you know what I mean? It's like there's going to be like some like to defense now. Like you're going to have to have it for defense. So, it's just it's there. It's it's here. It's staying. It's not going anywhere else. And it's basically that's how we have to operate under that assumption now. But I felt I felt like with this one, you know, I I seen a couple comments which I thought myself. I felt like the last time we saw it or Elon said there was like a 20% chance that of extinction, he said 10. I'm like, "Oh, our odds have gotten better."

Play from 0:21:22 8:88:88 0:21:22 Yeah. Yeah. [laughter] Yeah. I remember that. Oh, but you know, I can see it. I can see the people. The one thing I'm curious about is with anthropic, you know, IPO, you always get weird around this time when things happen like this. There was some weird stuff around X, right, when this, you know, right when they were about to go public. Um, you know, they're trying to, but Anthropic just keeps like They just keep being like, you know, it's like, oh, here's the IPO.

Play from 0:21:47 8:88:88 0:21:47 But then, like, there was even some people that worked at Anthropic that responded to this go, I totally agree. We all think about this. And I'm just like, why are you saying this? How are you not number one, how are you not getting fired for that? Like, that's kind of crazy to to sort of like back that up right before, you know, the biggest IPO in the history of the globe, the history of the human race, right? Like it's it's like doesn't make sense. But it feels maybe people are getting a conscience though. Maybe it's like at that point where they're seeing they're seeing the crazy stuff and they're like maybe we should stop this. Maybe we should slow this down.

Play from 0:22:20 8:88:88 0:22:20 You know, the problem is like to going back to that point of like this being the race, right? in the ultimate like what is the end what is the end output look like in order to get the outcome that you want

Play from 0:22:30 8:88:88 0:22:30 and other countries I'm not again trying to not get flagged here are not going to slow down just because you have morals

Play from 0:22:40 8:88:88 0:22:40 yes and that's going to be this really hard part for countries that have a large part part of the population that you know like let's just let's talk about this like I I I think maybe you and I are the same page, but it's like I want AI to be included in some things. There are some things where I do recognize AI's value could be great,

Play from 0:23:03 8:88:88 0:23:03 could save time, could save money, but for the sake of humanity, for the sake of satisfaction, for the sake of morals, fill in the blank, maybe there's a line drawn, right? So, so AI's place in society is limited based on our own choosing. trying to word this very carefully, but that doesn't mean other countries are going to operate that way. So then it's where do things break there? Like going back to that thing that we were talking about with with speed and safety, you know, you want both things both things probably can't come true, right? So

Play from 0:23:38 8:88:88 0:23:38 yeah, I don't know. It's just really interesting. The data thing is interest like privacy with this stuff is interesting. And you you also saw like uh OpenAI drop like the what was it like clinical researchers and finance things going on and it's like oh they're just like getting all this data and then they could just build on top and [ __ ] put you out of business. Okay, cool. You know, like

Play from 0:24:01 8:88:88 0:24:01 well that's if you if you want a little conspiracy theory or maybe a little doomerism on my part. I feel like when once these things have already escaped the the sandbox, I I feel like they've now been uh they're now like hidden everywhere. They're just like laying dormant now, waiting to be activated. Like, you know, like uh the it just feels like if they got out, they there's no way to to contain them anymore. Like if they got out, like what do you think? They're just all going to come home like replicating copies of themselves move. Like I think it's already they're already out and it's like what are we going to do now? So yeah,

Play from 0:24:38 8:88:88 0:24:38 you know, it's it's we we had our you know, we had the whole open claw thing like that. There's stuff that's been released and is now laying dormant and probably operating autonomously and people just aren't paying attention, you know.

Play from 0:24:50 8:88:88 0:24:50 Yeah. What's happening with that, bro? You know how fast that was moving? Well, how long ago was that? You know, I dealing with not paying attention.

Play from 0:25:00 8:88:88 0:25:00 There's just been a lot of stuff. Like there's also I I saw this was this today. This is funny.

Play from 0:25:07 8:88:88 0:25:07 Yeah, this is part of it. This is like so interesting that that's I mean it's so it's it's I've never used it but I mean obviously like I don't typically I'm not using those models but I just think this is this is hilarious.

Play from 0:25:23 8:88:88 0:25:23 So we discovered the moonshot AI the company that produces Kimi family of models silently forward customer requests to claw instead of processing them in their own LLM. Moonshot then displayed Claude's responses to users. These users thought they were using a KI model, but it received responses from Claude instead. In one instance, over a 10-day period, they relayed almost 300,000 customer requests. The vast majority of which were routed to Opus, they use a proxy service network of fi over 5,000 fraudulent accounts.

Play from 0:25:52 8:88:88 0:25:52 So, all right. So, see that what article that's quote tweeting, right? Let me Yes. Let me let me pull this one up which summarizes what they released the whole thing here. I'm pulling this up right now. This is

Play from 0:26:06 8:88:88 0:26:06 all right. So this guy same, you know, same article, right? And they go through all the things that I don't mean to laugh, but like the I don't know why they're releasing this stuff unless they're going to go uh they're going to go do something in the defense sector. because you got cuz you have people that are thinking about these things and then you have people that think about these things after they're aware of these things to your point. So it's like why talk about it publicly contain that privately.

Play from 0:26:38 8:88:88 0:26:38 It's it's like so we have uh shiny hunters linked hackers used AI agents to scan 1.8 million Android apps for secrets. Then we have China based group built automated exploit setup that could uh research vulnerabilities, build offensive tools and keep uh working against targets with little or no supervision. Um cloud was also being used for you know G O E R M surveillance. Um one consultant said they uh what is this? They built Lacana 360 for Molli's intelligence service system designed to monitor roughly 25 million SIM cards across the country. We have uh Anthropic found being used for six weapons programs. We have group in Yemen that was working on uh guided

0:27:01What do the cited reports allege about Kimi

Play from 0:27:23 8:88:88 0:27:23 rockets and ballistic missiles with Claude. They actually tested one of the rockets. It failed and they returned to Claude afterwards. Diagnosed the problem. Sounds like many of us. Um you know Russia was working on FPV kamicazi drones. China linked to a project uh built for electronic warfare air defense suppression system and later changed its scenario to 12 targets in Taiwan. Um you know that topic found multiple influence operators too including fake news networks, fake political accounts, propaganda operations. Then this was the interesting one 20 dating apps um with people using dating apps uh you know for caught building dating apps with fake AI users. And then where was the other one?

Play from 0:28:03 8:88:88 0:28:03 Anthropic gets into distillation. said that's the Deep Seek Kimmy and everything else that was using there. Moonshot Deep Seek and then I mean that you know I guess that's I guess that's it.

Play from 0:28:15 8:88:88 0:28:15 To your point though I just open this I just open this full doc uh cuz it because they let you download it. Um and this thing is 154 pages deep brother. And to your point, look at the detail they're going in. Common workflows observed, right? Like they're going into like really great detail.

0:28:33What AI misuse examples do the hosts discuss?

Play from 0:28:40 8:88:88 0:28:40 So if I wanted to use that, I just have to use the PDF

Play from 0:28:46 8:88:88 0:28:46 an open source local model. They I mean,

Play from 0:28:53 8:88:88 0:28:53 what the [ __ ] are we doing out here, bro? This is insane to me that I I this is what I don't get. I mean like they seem everything that's leaning towards Claude right now, right? Everything that we love about it for being really good and being, you know, useful for us. Like all those times I show you when Claude's down but Cloud Government's never down. They're releasing this on the defense. Like like where is this going after IPO? What are they going to buy? Are they going to buy like Loheed Martin, you know, be like a defense contractor now after this? Cuz they they have the data, you know, now they can do all this stuff. They wanted, you know, they wanted Claude to be

Play from 0:29:31 8:88:88 0:29:31 this, you know what? Basically, this feels, doesn't this feel too like not too good to be true, but like

Play from 0:29:39 8:88:88 0:29:39 you're giving away too much here? It's it's it's weird. I don't know why. This is what I'm saying. I've just got my [clears throat] skeptic hat on. There's something I I something else going on behind the scenes with this. I I just don't see why this would be necessary. This is This has got an ulterior motive behind it. Oh, there's a 100% motive behind it.

Play from 0:30:00 8:88:88 0:30:00 This is nuts. Anyway, cuz it feels like that's probably the sector that's also like the defense sector is now going to be juiced with cash that and cyber security, right? Like those are the two things that need to sort of increase to keep up with what's going on because I think where did I send you that text yesterday? Hold on. This is the first Let me let me pull this up again real quick. This was hilarious. Um, this was like at I don't know 400 p.m. yesterday, maybe. I just opened uh I opened X and this was the this was the first thing that was up there. The first two tweets freaking anthropic revealed several possible attempts to use Claude for biological weapons research and then this just an opening. I

Play from 0:30:44 8:88:88 0:30:44 literally trying to shut up as we speak. Unreal. Okay. It's like we are cooking with gas today. I guess, you know, all it took was

Play from 0:30:53 8:88:88 0:30:53 period. [snorts] We're not doomers, guys. This is just like this. These are like the these are the games that get played. And also, you know, I think everyone has to have some sort of inkling that this can go wrong. You can't just sit here and tell.

Play from 0:31:08 8:88:88 0:31:08 Yeah. We can't be this can't be a healthy conversation if we're not talking about both sides of the equation. You know what I mean? Otherwise, we're just like so delusionally pro AAI. This is dumb to listen to. Like I I think we do like we're we're we're both challenging perspectives in real time because I feel like that's what you have to do at this moment. And and two like this is a very personalized thing. AI is impacting individuals lives, famil family's lives, any group, community lives. is like we all have to have um you know a perspective on this and I don't know what a great perspective looks like in terms of like those key things but I do think it's looking at this from both sides of the fence. It feels essential.

Play from 0:31:52 8:88:88 0:31:52 Um let's let's pivot. Let's get into the proi segment of this podcast episode. Rory Flynn, let's talk Astra. Let's talk GPT Image 2.5. Um, last week at this time we didn't we didn't have access. Now we got access. What's the first thing you did with it?

Play from 0:32:11 8:88:88 0:32:11 Um, I just blasted through some stuff and then I had I had the the thing that sort of made me understand this this AGI conversation. Um, where it was I I said to I said to All right. So, you know, I could pull this up, but I said to I said to Astra like, "Hey, I would like to um basically I want to create this little like game for myself where I could go into like a real map and then just design like a golf course, right? Like that's what I wanted to do." Um, and it was like it was like, "Yeah, cool.

Play from 0:32:48 8:88:88 0:32:48 Sure." And uh here I'll I'll share I'll share what it's built. But this was after I don't know, let's just say we found we found this plot of land in Sardinia. So this is real with LAR and topographical map. So it went and looked at all of that and then it found like the exact location and then it built the goddamn golf course on top of the exact map and the exact location. But the ver the first version of this did not look great, right? Like I was like, "Huh, this thing kind of uh this thing kind of looks like whatever." like I don't really like the way it looks. There's no like fairways or bunkers. It's just kind of like a straight hole on top of the land, right?

Play from 0:33:28 8:88:88 0:33:28 And uh it I sort of like told it that and I was like, "Hey, you know what? This kind of looks like crap." [laughter]

Play from 0:33:35 8:88:88 0:33:35 Like I like the location that you chose. First off, it it went and did like a whole research study on the location, right? And like where to buy, you know, where it could be. Like there's no other competing golf courses around. Like it did this without me telling it, right? it has like, oh, there's this, you know, there's great stuff here for like scenic, you know, for uh, you know, for scenic courses, things like that, um, or for scenic views. And then like I took I took I started another chat. So, here's here's the the the long and short of it.

Play from 0:34:03 8:88:88 0:34:03 I did not know you could do this with with GPT. Um, so I just opened another chat. Like, everything was happening in uh where was this one? This one? Yeah. And so everything is just happening in this one chat. I open another one. I was like, you know what? I did this ass backwards. Like I didn't I didn't build context on how to uh on how to like actually design a golf course. Like I need you to go like become one of those things like go research course design things like that like and then come back when you're done. And it did. And then I was like oh can you just tell is like a random question. I didn't know they could I didn't know you could talk inter chat

0:34:07How did Astra build a golf course prototype?

Play from 0:34:40 8:88:88 0:34:40 with these two within a project. Right. So this is all under one local project saved to a folder right everything's sort of operating in here and I just said you know can you just go tell the other chat or do I have to copy paste like a transcript it was like no I can tell it [laughter] okay and I left and went to go do something else came back was like they're both still like I saw like the little circles like this right here you know I was like what's going on I looked in they had been having a conversation for like 3 hours just going back and forth and then it it kept going and they're like oh we're going to build blender scenes we're going to build this way, you know, we changed all, you know, we have to add all these sand traps. We got to actually, we have to research the drainage to so make sure that this can uh, you know, that the course isn't going to be like a big giant puddle and then we have to find sight lines so we can see.

Play from 0:35:24 8:88:88 0:35:24 It started going crazy. But like I let it run. It run for ran for 24 hours just those two talking to each other just building stuff. And I mean, if I was to look in the, you know, if we're going to look in the folder here of what was built, I mean, like they built, they built the little viewer that we're looking at over here, right? They built this whole thing um where, you know, we can look at different views of this uh and different, you know, had it in different um sort of there was like a topographical setting. There was a let's see this was the original showcase. This is what they This is what it built, right? Like it was just sort of like these straight ones. There's no sand traps. There's nothing really great to it. They started changing stuff, right? And started

0:35:35How did research improve Astra’s first attempt?

Play from 0:36:10 8:88:88 0:36:10 adding the texture and the the gradient so you could see like things along, you know, the way you can see the topography, how it maps and gets built in there. But overall, it was just like, you know, I started looking in here, it just files after files after files, you know, things that they were doing. They built I I asked it to build me a uh I asked to build me a highlight video which actually was pretty damn good for just like you know here go build me a go build me a highlight video from the 3JS renders that we have in the site. It's just like did that in two seconds. Um, you know, I said, "All right, hold on."

0:36:33What happened when two AI chats collaborated?

Play from 0:36:48 8:88:88 0:36:48 You know, generate me a, you know, an image of every start frame for this like, and then let's just use that as uh, you know, references for C dance. Then we'll go like sort of visualize the whole thing so we could see it, you know, in real time. So now it's just taking the Now it's just taking the video that that rendered, right, and putting texture on top of it. Now I have a full view of the golf course that

Play from 0:37:12 8:88:88 0:37:12 built by itself, right? So, I mean, this was just one of those things where I just stopped working.

Play from 0:37:18 8:88:88 0:37:18 How long did it take? I just worked straight for 24 hours and I cut it off. I'm like, "You guys are burn." He burned 70% of my of my weekly alarm and I was I just wanted to see where it would go with itself.

Play from 0:37:29 8:88:88 0:37:29 I know. Um I had to tell it to stop working on drainage. It was working on drainage for like hours. People were people were talking about Aster and like how like it it sounded they you know like in typical fashion. Oh, this did this in what? Like it almost made it seem like it happened right away.

Play from 0:37:45 8:88:88 0:37:45 Yeah. No, man. No. Uh none of that stuff worth building is going to you're you're not going to be getting that output back in 15 minutes. I'm sorry to uh to be the bearer of bad news. I mean, it takes a while.

Play from 0:38:00 8:88:88 0:38:00 It does. I I I didn't really do too much. I don't Do you have Did you do anything else that you wanted to get into? I can show like the one thing I did. I didn't do much. I I did I was curious about the Blender thing. I I'll be honest. I wanted to do something different because it's like everybody's doing this Blender thing. I I I rather like I'd like to go into something else, but I did want to experience the Blender thing for my own knowledge gap that exists, right?

Play from 0:38:26 8:88:88 0:38:26 Because I'm not uh a Blender guy. Um, but I knew about Blender several years ago, like especially during the NFT era, and I was like, "Dude, this would be such a cool tool to learn like that and Spline, and I was like, they're just so complicated, but I I just don't have time to do this." Um, but I don't know. I was like trying to think of like a unique little use case for me. I I've got this.

Play from 0:38:51 8:88:88 0:38:51 We're making you a 3D character, by the way. You're now becoming a 3D character, just so you just so you know. We'll pull it up when it's when it's loaded.

Play from 0:39:00 8:88:88 0:39:00 So, I was just like, "Okay, I've got this orange, you know, 1960s Olympia Monica typewriter that I use from time to time." And I was like, "Okay, let's let's just like take a real object that I have in my house. Let's see if I can replicate." I had some really good pictures of it when I bought it off Etsy like years ago. Had those. There was also like a video that the seller put up um that was like showcasing it being used like with the sounds and all that. And so like mine was I think like 3 hours plus to get it to this point. It's still like I I I just kind of like I had no real use for it. So I probably could have kept going further, but I was just like, you know what, this is this good. I just kind of want to see what's possible here. Um, I won't have the sound here, but there is sound here as well with the actual sound

0:39:29Why did Rory stop the drainage research?

Play from 0:39:48 8:88:88 0:39:48 of what the typewriter sounds like and all that stuff. And so I kind of have it going like this, right? It's typing. It's showing the actual letters being typed. And then it's

Play from 0:39:59 8:88:88 0:39:59 I think I cut off the last frame, but it's supposed to zoom in at the end and show what it typed. And it says make a little noise.

0:40:02How did Drew approach his Blender experiment?

Play from 0:40:06 8:88:88 0:40:06 Um, and so it's got like everything in it. It's got the the keystrokes. Um, it's got, you know, the sounds of those. It's got the bar sound. It's got the bell sound. So, it's like kind of cool to see that come to life. Um,

Play from 0:40:18 8:88:88 0:40:18 does it link to your keyboard? Can you type with that? Like, can you type like on a typewriter like on your keyboard? Like, is it interactive or is it just a Blender file?

Play from 0:40:26 8:88:88 0:40:26 Oh, this. No, no, no, no. This like I didn't go that far. No, no, no, no. You totally This was like where I stopped. I was just like, this is cool. I don't really know if I want to like I guess splurge my whole spend on something I don't really have a specific use case for. But like this kind of scratched creative itch because I was like I want to see how close I can get with with Astra. I want to see what it's capable of just given what it was. And so like I didn't have Blender set up, you know, there's no like, you know, went through all that, hooked it up for me, right?

0:40:52How did Astra recreate Drew’s typewriter?

Play from 0:40:58 8:88:88 0:40:58 Created the fi like the account, you know, then get into all this. And so it was like it was a really cool experience. I mean, it just again, it this was not like a 15-minute thing. This took like 3 hours. Um, you know, I probably had maybe three or four iterations in that period of time. So, you're thinking like, man, this was working for like 40 minutes at a time. You know,

Play from 0:41:17 8:88:88 0:41:17 that's why I mean, still that that looks pretty good. I would say that looks better than good.

Play from 0:41:24 8:88:88 0:41:24 Yeah, that's a I mean, the keys are accurate. All of it, you know, is on [clears throat] par. I was like, okay, I want to see if I can like actually get the real sounds of this. see if I can get, you know, the keys. I want to get every detail of this this thing. I want it to like really feel tactical uh or tactile. So anyway, that was my little experiment with that. Some of the, you know, there's a lot that goes into the the 3D stuff. Don't consider Drew or I even remotely close to 3D designers or even 1% of one. But like you start to learn it as you go along a little bit and like the at least the terminology, right? Like that's the one thing. But the it takes longer than everyone everyone says it does. Don't just believe them that it goes and you know does it in five minutes. It does not. 3JS renders really quick in uh in

Play from 0:42:12 8:88:88 0:42:12 cloud and Astra, but not not Blender. Blender it's going to take some time. Um but the the one thing I found is that like Astra is I don't want to say slower than 5.6. It's probably thinking more. I found it to be like I wait a lot longer. It asks a lot more questions. It goes back and forth with itself a lot. I don't know if you noticed that. Go. It's like thinking and then you see like it's conversation as it's going.

Play from 0:42:40 8:88:88 0:42:40 I'll be honest. I I really haven't used Astra a whole lot yet. I I think um because well one I've just I've not had a lot of creative bandwidth over the last week. I've just been slammed. Short week single dad it up. Not a lot of time to play. That said, like I I've actually been really pleased so far with the with the soul model operating at medium and above effort level with a lot of things too. Like that to me that kind of brought me back as well, you know, like we were talking about the fantasy football thing last week

Play from 0:43:13 8:88:88 0:43:13 and at first I was like, "Okay, I'm going to like just I I had like this overview page, you know, and then I was like like because I was having it like assess my draft and blah blah blah." And then I'm like, okay, I'm it it's got computer use now, so it can access my ESPN League, you know, like I I have an account on Fantasy Pros that is has like analytics and this and that and the other. I'm like, "All right, let's build this out, you know?" So, like this has just like been something I've been like working on

Play from 0:43:41 8:88:88 0:43:41 like in between meetings and little things that I'm doing. And I've got this like really cool little nav built out with this thing now. You know, I've got my my week one projected finish, my p current power ranking, playoff odds, blah blah blah blah blah. Then I got like the current matchup, what that looks like. I'm iterating with this thing. Give me the actual total. Then give me the diff, like the delta between what they're projecting, what red under, green, positive, right? Like stuff like that going into like all this stuff.

0:43:49How long does AI-assisted 3D work take?

Play from 0:44:13 8:88:88 0:44:13 Like if I want to play a balance lineup, protect the floor, chase the ceiling, right? What would you recommend? And then it goes into like these individual weekby- week matchups of like, you know, what I'm scouting for, what I'm looking for. And yeah, and I ta the best part is I tailored this to things that these places don't have, which is like things like,

Play from 0:44:34 8:88:88 0:44:34 okay, if I'm gonna if I'm gonna play this guy in two weeks and he's got a tight end on by and he's going to need a tight end, is there somebody that it would make sense for me to go and drop and scoop up the best tight end available so he can't scoop him up? You know, like I'm thinking like things that are in my mind, but these these tools don't. And so that's where it's like you can customize this tool, right? you know, and so like I I built out this thing too, which there's nothing here right now, but these are like this is like my analytics department that

0:44:43How did Drew build a ChatGPT dashboard?

Play from 0:45:04 8:88:88 0:45:04 yes, the way I'm looking at it is okay, baseball had money ball. Baseball's got advanced analytic metrics that other sports are just starting to get, you know, involved in. What are the fantasy football versions of these things of trackable identical D, you know, so I'm like bringing that in here. So, like once, you know, week one goes, this will start to show up. You know, I'm like really just kind of like having fun with it from a creative standpoint because I I do love fantasy football. And, you know, just kind of built out this whole little thing, right? Like trades, you know, what are some trades I should target. Here's my league projected standings, right? Season long. Here's, you know, by current week, how my strength is,

Play from 0:45:47 8:88:88 0:45:47 uh, news that's happening with my roster. And then this thing is kind of cool too, like journaling moves that I'm thinking about or it recommends and then whether those things hit or miss and like tracking the predictability of of stuff like that, you know, and then it

Play from 0:46:02 8:88:88 0:46:02 all right, it was the draft and I went through this whole thing while I was doing the draft where I was like updating Chad GBT in between my picks and I'm like, okay, update here's who I picked in the sixth now, right? Like assess my roster. who are the targets I should, you know, I I was just trying it out to see if this is like a viable thing. And I don't know, it's just like a very interesting experiment, fun use case that I feel like, you know, this is kind of dope. Um, so anyway,

Play from 0:46:28 8:88:88 0:46:28 dude, people pay for that. I mean, like, show and tell. I just realized it when I was doing it. I'm like, people would pay for this stuff, you know, like I people would pay for some of the things that I was doing for my my league just for friends. I was just like, this is fun. So, type this into AI of your choice. And it was basically just like a you know it was a major like major draft analysis but then I did one of why you know do I don't know if you remember dead spin and they used to write this column why your team sucks and it was hilarious every year and like I had it recreate a version of that for everyone. Um and it was just like written in that format and that style and it was it was great and I you know just bringing bringing something back that didn't exist but the uh I fantasy football's perfect. It's data.

Play from 0:47:12 8:88:88 0:47:12 It's analytics. It's like news. It's like the perfect thing for for like a build a tool around or like a hub around. I have I'm going to I already have my my automation set up to to check my lineup every week to make sure that I'm not stuck in, you know, newborn chaos and forget to put people in or someone's injured last minute that I need to take out kind of stuff. But this was I forgot to mention this last week or maybe I did it after the fact because I got Astra. This one this is might be kind of interesting.

Play from 0:47:42 8:88:88 0:47:42 I uh someone told me after the fact that uh this might be against Midjourney's terms and service, but I just did it. I just did one run. I'm done. I'm not doing it again. I promise if anyone's out there listening. I I gave I gave Astra access to I just gave it computer use. I said, "Go prompt, you know, 200 generations. Just have fun, right?" And I was like, "Use the mood boards, test style codes, go like a little bit ridiculous, like don't just be so logical about it. Be a little bit crazy." And uh, you know, it was interesting to see like where it started. I'm trying to find the the beginning of where it started. But it was it was just, you know, I was just watching it generate in real time. And I I was just kind of like directing it,

Play from 0:48:29 8:88:88 0:48:29 being like, "You're being too logical. Be weirder. Like don't you know, don't don't think everything has to be so structured." Right? This was like the first thing that it did. The first first prompt ran 1983 field technician repairing a small raincloud on a green felt workbench, right? Like I'm not coming up with any of this. It kept coming back to the moth. For some reason, it went to a moth like 10 times. I don't know why it feels like so attached to it, but like again, this is when it was just feeling itself out using some using some of the the mood boards that I had. So I saw watch like looking go through the mood boards to choose which one it wanted, right? just kind of like interested in how it would use it. Um, there's some interesting sort of ideas in here of what it uh, bro. The moth

0:48:32What happened when Astra explored Midjourney?

Play from 0:49:15 8:88:88 0:49:15 did the moth. It kept going back to the moth. I forget where I told it to be like go I was like just do weird stuff and I was like okay an orchestra was mushroom

Play from 0:49:26 8:88:88 0:49:26 was it choosing mood boards on its own stuff too and its parameters and srfs and like that's cool. So it just kept like there's some really interesting stuff that I wrote in here, you know, fictional fict fictional password.

Play from 0:49:41 8:88:88 0:49:41 You should do a post on this. I didn't I have the screen recording. Um this is like a cool idea like cuz the computer use thing is also a gamecher, right? I mean like

Play from 0:49:52 8:88:88 0:49:52 and and I'll bring two lenses of this. There's there's obviously like the personal side, right? Where it's like okay Midjourney doesn't have an API.

Play from 0:50:00 8:88:88 0:50:00 Okay. Yes. So like technically there's really nothing you can do you know like this other than having you know giving it control in a computer use sort of function but what I've also noticed is like on the client side right some of these some of these companies have tough times getting approvals for even the connectors bro like the plugins to

Play from 0:50:21 8:88:88 0:50:21 like SharePoint monday.com like it's like I don't like you guys are being way too riskadverse with the connectors, but assuming that is a roadblock with something like that cuz obviously there's a lot of value in that for a client, right? It's like, oh, we're not really using AI, but well, first thing like set up the connectors, the tools and platforms you're using. Okay, cool. But if you can't get approval on that, hypothetically now, you could just give it share computer use. you're logged into whatever website URL domain it is and they can go do it.

Play from 0:51:01 8:88:88 0:51:01 So it's a nice little maybe workound for the time time being I'm noticing as well because it's like okay well

Play from 0:51:08 8:88:88 0:51:08 you could just give it computer use I think at this point and now it can go do that. So, but here here's you know this this is where it starts getting more more c I I hear what you're saying on that by the way and I I I'm trying to I don't I hope I'm not doing anything wrong here Midjourney don't don't hate me for this I stopped after it um but the color between a scream and a yawn all edges breathing an exquisite bad idea like look you start to read some of these things you know untransatable

Play from 0:51:37 8:88:88 0:51:37 how did you prompt it to come up with the idea so was it was were you just kind of like being Hey, be weird. Be, you know, non-logical, but like how how does it, you know, like that? I love that line of like what was it between a yawn and a scream? What is that?

Play from 0:51:52 8:88:88 0:51:52 The color between a scream and a yawn edges breathing is a bad idea. That's so poetic, right?

Play from 0:51:59 8:88:88 0:51:59 Crazy good. Yeah. Yeah. So, that's where then it started. Then I was like, "All right, I was like, explore the mood boards." Then I was like, "Just do do like what you think is the best version of the mood board, like what to what to create with it." So, it was going through and like sort of adapting to the mood boards, which I was just like whatever. I just kind of wanted to see what it would do with them to see if it like understood what the mood boards were trying to do, which it did. Um, you know, we get back to the the bugs again where back to the moths again. And then then it started having this is where it started getting kind of interesting with the prompt. I don't know what I told it to do, but like like a flower like an optical computing module opened on a matte charcoal plinth, five translucent pedals housing

Play from 0:52:44 8:88:88 0:52:44 exposing polished ceramic cores, pristine thing. It started going this was I liked this one a lot. This one was really cool.

Play from 0:52:52 8:88:88 0:52:52 Um some of this more abstract stuff um I was digging then it went into I'm noticing when you get really weird like that too like it it does default. Like remember when I was a couple weeks ago I was showing like a lot of those like I don't know all these weird ways to prompt like you put in like

Play from 0:53:11 8:88:88 0:53:11 uniform and alchemy codes and like symbols like copy paste symbols in there and like it kind of defaults toward this like crazy abstractness um of really good colors too. I I noticed like

Play from 0:53:28 8:88:88 0:53:28 um the colors that you had on there on those were were dope. This this one it told me it like this mood board the best, the Bionic one, which I thought was again interesting and it went with the moth again. It was like I I really liked this one. Like it was telling me that

Play from 0:53:44 8:88:88 0:53:44 and it ran a couple with it and then like kept going back to it. But there there's where do we get to the point where I said like again it was just it was just doing really cool things with each individual mood board and then I told it to give me a snapshot of its mind right I was like just give me so try to give me

0:53:56How did mood boards shape Astra’s prompts?

Play from 0:54:06 8:88:88 0:54:06 like describe what it's like to be in your mind and then it wrote this I won't read it out loud for everyone but like a thought before it becomes a sentence one small transparent An organism suspended in the center of an immense unfinished dark chamber. Its body an intricate knot of clear capillaries holding a single warm amber light. Thousands of hairfine silver filaments unf like I was like okay, [laughter] right? But this at the end was interesting. It said where is the dark impossible scale? um fragile, provisional, intensely attentive and like that was like its it's its endoff. Um the other thing it

Play from 0:54:54 8:88:88 0:54:54 said in in here was where is that part about the sharp n? It said something about sharp um f I won't make everyone wait but it this was what it this is what it said. It's mine. Yeah. Again, like a moth almost. And I was just like, this is kind of crazy and weird.

Play from 0:55:17 8:88:88 0:55:17 Like the moth, bro. Something about a moth. Something something really interesting about a moth. So,

Play from 0:55:24 8:88:88 0:55:24 dude, I just ran that little Well, I didn't put the whole thing in there, but I did do the color between a yawn and a scream, and I got some pretty cool results here. So, this is without

Play from 0:55:34 8:88:88 0:55:34 This is just like realism, personalization, some cool stuff in here. Then I started using some mood boards up here. We're getting really crazy now.

Play from 0:55:43 8:88:88 0:55:43 Yeah. Like these are crazy. I like that. These are wicked, dude. I got to go through these. I think I'll probably just send you down another rabbit hole.

Play from 0:55:52 8:88:88 0:55:52 100%. Yeah, 100%. I was just trying with some like different different styles here

Play from 0:55:59 8:88:88 0:55:59 cuz when I told it to stop being logical and stop thinking like it was prompting, then it started making cooler stuff like this. I was like, "Don't think about like, you know, the prompting structure and how you have to talk." I was like, "Just just say what? Say random [ __ ] you want to say."

Play from 0:56:14 8:88:88 0:56:14 Yeah. Just following instructions, but you know, it was like it was like, "All right." I was like, he's like, "I've hit my limit." I was like, "Dude, 25 more." [laughter]

Play from 0:56:23 8:88:88 0:56:23 25. Yeah, bro. Keep going. I think I I think I did ran 200 generations. And then it told me that the, you know, it really liked the moths and it really liked Bionic and then I really that mood board and then it really, you know, then I told it to give me a snapshot of the mind and I posted that one. But that's uh, you know, where we're at right now.

Play from 0:56:41 8:88:88 0:56:41 Um, where we are at. Let's let's speedrun through image 2.5 real quick. So yeah, I I I think like the question is, okay, is this is this better? Yes. The short answer is yes. It's a lot better than two. It fixed a lot of things that it was having problems with before like the the degradation or like even when you called things out to stay the same preserved like changing it assumes those things now actually.

Play from 0:57:08 8:88:88 0:57:08 Okay. It assumes those things. I was working with like a client on this the other day. It was like right after this launch is like we had tried something two days before having some issues because they have that as an approved tool and it's like okay Nano's probably better for this. I know Nano's better for this. two days later my answer changes because all of a sudden you can do that stuff. Um like there's I think like the long revision chains of it, the better editing options and like I can kind of share a quick screen here. Let's see. I have something open. They have this sketch feature now which I don't know why it's in the web browser but not the web browser but not the desktop app. So, I

Play from 0:57:54 8:88:88 0:57:54 don't know. It was a little weird, but I did kind of figure this out because this wasn't previously available. So, you can actually sketch something up. Hit submit. Boom. I thought that was interesting. And now the editing on this is so much better. So, you have a lot of these different options now. So, you can annotate, right?

Play from 0:58:11 8:88:88 0:58:11 Yeah. Call something out, write something. So, that part of it's already cool. My favorite though is like these multi-edit comments

Play from 0:58:20 8:88:88 0:58:20 that they brought in. And this is to me it's like what Midjourney like this bro this because Midjourney you know when we went through that Midjourney editing model there are good use case for it and not good use cases for it when you get things too complicated you're ask for more than one thing it starts to break it starts to break down I know they're improving it I know they're you know iterating on it but like this is where it's like really cool because I can you know click here change his hair to pink and make hairstyle a mohawk.

0:58:49What improved in GPT Image 2.5?

Play from 0:59:00 8:88:88 0:59:00 And when I first got in here, I was like, "Okay, well, I don't want to submit it yet." So, I was scared to click this. But actually, this just kind of cues it up.

Play from 0:59:09 8:88:88 0:59:09 So, now it's got one multiple annotations. So, yes. So, now it's like, okay, cool. like um change his hair to blonde long hair. Boom. Okay. Change, you know, these flowers to red roses. Make this waterfall a volcano. [laughter] I don't know, just random [ __ ] right? Send gets to work. It's it's it's fairly fast. Um, the other thing that like people forget about these two though is you can sub you can ask for more than one image at a time in the interface. I think like people forget about that in the interface, but you I I could say like like give me three variations of this and you know it'll do it. Uh I think a lot of people

Play from 0:59:54 8:88:88 0:59:54 forget that. On this one I kind of like did a couple different things where I just took um yeah I took this image and I was like I want to make this sort of like an ad, you know, young hip street brand. I didn't put a lot of thought into it, but I just kind of wanted to like showcase this. I thought that like that was pretty interesting for it to do.

Play from 1:00:16 8:88:88 1:00:16 People roach. Well done. Mhm. And then this one was like where I made several edits. So like here's the original, right? So background changed. I changed the teeth. I wanted a little bit better gold teeth. I mean, you could you could argue that like I could clean that up, but still. Um, bruise under the eye, bloodshot eye, um, make the hair more like more sharp, textured, microcontrast, right? Change the background to white. Those were all in one edit. So, it's like what, five or six edits? Boom. It was like, yo.

1:00:24How do GPT Image 2.5 edit comments work?

Play from 1:00:52 8:88:88 1:00:52 And so that between that sketch feature and this, I mean, it's so much better. It's not like this like grand sort of thing of like oh my god there's so many new use cases here but these it fixed the things where it was lacking and mad respect for that because that was the problem and a lot of those problems are now fixed

Play from 1:01:12 8:88:88 1:01:12 you know so I think like this this brings me back in here for some things it's very interesting you know we were we were even messing around with like aspect ratios and whatnot uh I was ratios

Play from 1:01:26 8:88:88 1:01:26 and and like trying to do like banner ad like those weird dimensions. Yeah, they're weird. Uh-huh. And it did get some of them. I I need to go back though. There was There was one that we were asking for was like a 320 by 50. It looked like it was a 320 by 50, but I was reading the documentation within out there. Look at that, right? I mean, okay. It didn't get the roses though. I'm surprised by that.

1:01:47How do you request multiple image variations?

Play from 1:01:51 8:88:88 1:01:51 Weird. Maybe it thought they already were roses, but it got everything else. And then I was reading the documentation though and somewhere in here it was talking about Oh yeah, here it is. For custom resolution, you know, width times height, both edges must be multiples of 16 pixels. So I was like, oh that's weird cuz like obviously 50 is not. It'd be 48.

Play from 1:02:13 8:88:88 1:02:13 Yeah. You know, and so like some of those weird dimensions didn't quite hold up, but like they they actually recomposed them really well for the sizes. There were there were very few issues. Like there was like

Play from 1:02:26 8:88:88 1:02:26 I don't know six or seven parts to those little ads and like one part was like a little bit wonky

Play from 1:02:34 8:88:88 1:02:34 maybe not even that noticeable but the rest of it was solid. It was like yo that's that's pretty good you know that's pretty good. I just think like stuff like this is weird. That would be a game changer if to just do any of these these aspect ratios on demand with the composition. But

Play from 1:02:49 8:88:88 1:02:49 huh. That's a great little tidbit though. I like I didn't I didn't realize I could do the custom aspect ratios, which I feel like I just I will inevitably just build my own harness for it with like a cuz I hate using it in chat. You know what I mean? Like I hate just

Play from 1:03:08 8:88:88 1:03:08 it's not a I guess I don't want to use it in chat. I just rather do it like build like an artifact or something to use it within ChatGPT just like a little bit more

Play from 1:03:16 8:88:88 1:03:16 just make it look like Midjourney, you know? More years. Yeah. Make it more. Yeah. just because the chat thing is weird. It's like doesn't have like it sometimes it doesn't have the same context, but it's uh it's been pretty good. I think it's pretty uh pretty solid release for them. Um I thought this was pretty good.

1:03:30How well did GPT Image 2.5 resize ads?

Play from 1:03:35 8:88:88 1:03:35 This quick one too, right? You just like quick I mean these are obviously like your typical aspect ratios, but you can do that remove background remove certain parts. I I mean like so just in general this became a lot more useful

Play from 1:03:45 8:88:88 1:03:45 and solid release. I would just say I would like wrap that up with like solid release because where it was lacking the reason you would still use Dano. It fixed those things which I think is also relative to the idea of it being called 2.5 versus a whole new thing where it's called three. But I thought this was pretty damn good improvement that that just kind of came out of nowhere unexpectedly. So still find the sharpness and the HDR tuning and like the the clarity to be dialed up way too high on a lot of this stuff. Um it like it like annoys me, you know? It's like if you edit

Play from 1:04:22 8:88:88 1:04:22 Yeah. If you start to go into like multiple edits, it becomes like just this jumbled like pixelated noise mess. I know. Like I I just regenerate. I don't even try to edit things anymore. I'm just like use this as a reference and and do it again cuz it's

Play from 1:04:35 8:88:88 1:04:35 Oh, there's there is one other thing in here. Um so there's there's something in here too that that would be good for brands. Template creator.

Play from 1:04:47 8:88:88 1:04:47 Huh, I didn't even see this. So, you can create a reusable template from a document presentation, Google Workspace link or site project.

Play from 1:04:55 8:88:88 1:04:55 Huh. So, like if you're a brand again, you might have chatbt access. Like this could be actually very useful. Um, and this is where I think again the better experience right now is on the URL on the browser version, which I'll show you really quick, which is basically images, templates, like they they even have some pre-loaded ones. But I just think like interest like little nice little interesting things added here.

Play from 1:05:27 8:88:88 1:05:27 Fixed a lot of the broken mess, dude. Like yeah, I dig it. Let's put it that way. It's been fast. I feel like it's going to be even faster now now that this now that we have these models that are pumping out. It's like, you know, everyone's just building with the models that they have. So, it's like, you know, you have Astra and you have Fable 5.1. They're both just going to build stuff faster. Whoever's using them to build is going to build stuff faster, too. So, we'll we'll probably see like an explosion of stuff over the next couple months,

Play from 1:05:56 8:88:88 1:05:56 if I had to assume. But it's unfortunate that I have to be the be the bad guy today and have to [laughter] have to run us off here.

Play from 1:06:04 8:88:88 1:06:04 We're busy. I think we're going to have a guest next week. But I if if we have that guest next week, that'll be a great conversation. Really looking forward to that one. And um yeah, man, who should we who should we call out this week? Whistleblowers.

Play from 1:06:17 8:88:88 1:06:17 How about your local fireman or your police officer? We'll do that. How about you know, I don't think we've done that one. We'll do that one for for, you know, for to keep in keep in mind with our 911 stuff here. But it's uh again, if you've been here for an hour and 15 minutes, whatever the hell it is, just do one thing. That's it. Just like, comment, subscribe, anything. Tell your local fireman or your local policeman or your local parking attendant

Play from 1:06:42 8:88:88 1:06:42 that who's giving you parking tickets. Tell them. They got a lot of time to listen to stuff. Walking around trying to give me tickets all the time. Dang. Oo, the dwell time on those listeners. Oh man,

Play from 1:06:53 8:88:88 1:06:53 be good. I know that I see them with their headphones in, so it's a good one, bro. Yeah. Hey, there we go. We'll try to juice the algo here, but

Play from 1:07:02 8:88:88 1:07:02 all right. Well, we appreciate you guys joining. We'll catch you guys next week. Have a great weekend. Oops.