Latest / Bad Decisions Podcast / Rive comes to Unreal Engine, Netflix Uses AI generated Assets and Anthropic Caught DeepSeek Stealing from Claude
Transcript
- 0:00What is our beautiful people? Welcome back to the Bad
- 0:02Decisions podcast, where we bring you the hottest, the
- 0:06coolest analysis of everything in the world of tech, AI, and
- 0:11business. It's Wednesday back to you every
- 0:14Monday, Wednesday, Wednesday, and every Friday.
- 0:17And someone said they didn't like your intro last time.
- 0:21Why don't we bring back what is our beautiful Oh yeah, that was
- 0:232 episodes ago. Exactly how are you doing?
- 0:26I'm doing great. How are you?
- 0:27I'm doing fantastic. How are you doing?
- 0:30Hope they are having a great day.
- 0:31It's Wednesday and we are back with a ton of amazing news and
- 0:36updates. That's right, we've got updates
- 0:38for Unreal Engine for motion graphics designers.
- 0:41We got an update on Anthropic and a drama between them and
- 0:46Chinese models. We've got a news piece about
- 0:49Netflix and Confi UI, and we've got a special guest joining us
- 0:53later today. Amazing.
- 0:54Farah, shall we start with the first one?
- 0:56Let's go. So for all animators and
- 0:59designers and motion graphics artists and 3D artists who use
- 1:04Unreal Engine, you should be incredibly excited right now
- 1:08because Rive is coming to Unreal Engine.
- 1:12For people who are listening to us and are hearing their Rive
- 1:15word for the first time. This is a powerful design and
- 1:19animation tool that allows creators to build interactive 2D
- 1:24graphics that work seamlessly across apps, websites and games.
- 1:29A lot of people have been talking about right for the past
- 1:32few months years. I'm not exactly sure when they
- 1:35first launched, but they have become the go to software for
- 1:40people who want to create UI. And here you can see that we
- 1:43have the freaking beautiful announcement that is joining
- 1:47Unreal Engine, not acquired, but joining Unreal Engine on the.
- 1:50Fab Marketplace, yeah, and it would be great for game
- 1:53developers who are designing different levels, different UIS.
- 1:56Also, I think you can use it a lot for broadcast, whether
- 2:00you're doing news broadcasts or any other sports.
- 2:03A lot of the UI elements can be now animated using Drive, and
- 2:08apparently it's much easier to animate complex animation.
- 2:11What's amazing about Drive from what to understand is that once
- 2:16you design your UI elements and animate them super beautifully
- 2:20inside the application, you can then go ahead and take it as a
- 2:24single source of truth into applications, into websites and
- 2:28into games. So you make it once and the
- 2:31design stays and translates across different platforms.
- 2:34And the reason why it's exciting to have it inside Unreal Engine
- 2:38is, as Farah mentioned, for all of those video games.
- 2:41Think about the opening menu. That is the very first thing you
- 2:45see in video games. Think about the map for games
- 2:47like GTA or racing games. Think about the HUD, the HUD
- 2:51overlay for games like Call of Duty.
- 2:53All of them can be beautifully animated and used not just once
- 2:57in Unreal Engine, but perhaps on the website or perhaps on the
- 3:01application that the game might have.
- 3:04Also, don't forget about cars. Brands like Rivian that are now
- 3:09using Unreal Engine to create the interactive experience on
- 3:14the dashboard can now use Rive to create even more seamless,
- 3:19beautiful designs. I was also thinking about
- 3:21Fortnite if they bring this into UEFN.
- 3:25A lot of creator creator LED maps can now use Rive for their
- 3:31maps for the UI, for their HUD, so it could be used in the
- 3:35Fortnite ecosystem also quite a lot.
- 3:37Technically, anything that you can make inside Unreal Engine
- 3:40can now be enhanced using Rive. So I'm even thinking about
- 3:45Sprite sheets here for different Niagara particles.
- 3:49Imagine animating them using Rive and then putting them in
- 3:53your Sprite sheets, having them act as particles because of node
- 3:56system as well. So you can see that a lot of
- 3:59creativity will be allowed for the creators here.
- 4:03I'm a big fan of this. I've personally never touched
- 4:06Drive, so this will be an exciting thing to test out for
- 4:09all of us here. Bad decisions.
- 4:11If you guys have touched Drive and are excited about this, let
- 4:14us know exactly how you use Drive currently in the comments.
- 4:18I'm curious to know what use cases are there for you, the
- 4:22creators? OK, so Faraz, I want to take you
- 4:26to the drama that happened a few days ago.
- 4:29Anthropic posted something on Twitter.
- 4:33It's. I'm going to read it for.
- 4:34You Anthropic, the creators of Cloud.
- 4:36Cloud, yeah, we've identified industrial scale distillation
- 4:40attacks on our models by DeepSeek, Moon Shot AI and
- 4:44Minimax. These labs created over 24,000
- 4:48fraudulent accounts and generated over 16,000,000
- 4:52exchanges with Cloud extracting its capabilities to train and
- 4:57improve their own models. So what these particularly
- 5:00Chinese models have done, they have started attacking Claude to
- 5:04train their own smaller models basically.
- 5:08When you're talking about attacking, I think it's a lot
- 5:11more subtle than that. Just by having exchanges, it's
- 5:14not hacking. Just by having exchanges with
- 5:17the model, they can learn from the responses of Claude and then
- 5:21use that to train their own models.
- 5:23So basically the goal is to transfer the knowledge from the
- 5:26big model to the small model, making the student model faster
- 5:30but at the same time lighter. Exactly.
- 5:32Now, if you go down here, Pratt, if you can just pull down, let's
- 5:36talk about the rest of this message, Anthropic said.
- 5:39This solution can be legitimate. AI labs use it to create
- 5:42smaller, cheaper models for their customers.
- 5:45But foreign labs that illicitly distill American models can
- 5:50remove safeguards, feeding model capabilities into their own
- 5:53military, intelligence and surveillance systems.
- 5:57These attacks are growing in intensity and sophistication.
- 6:01Addressing them will require rapid, coordinated action among
- 6:04industry players, policymakers and the broader AI community.
- 6:08But you know, The funny thing is the AI community was not being
- 6:12sympathetic and the comments are really, really funny.
- 6:16I think the best comment is from Elon and he says in the comment
- 6:22that I have here, I'm not sure if you have the same one.
- 6:23Oh yeah, we have a different one, but I'll read both.
- 6:25Anthropic is guilty of stealing training data at massive scale
- 6:29and has had to pay multibillion dollar settlements for their own
- 6:33theft. This is just a fact.
- 6:36There's another post by Elon Musk which says Bangor.
- 6:39How dare they steal the stuff Anthropic stole from human
- 6:42coders. So someone is stealing from
- 6:46someone who has already stole from someone else.
- 6:49This is the demonstration of that, the public Internet and
- 6:53everyone's data, Anthropic using that to train their data with a
- 6:58straw, and then the Chinese AI companies coming and putting
- 7:01another straw into Anthropic and then taking that.
- 7:04There, there is a funnier one. Go down.
- 7:06Go down. Was it this one?
- 7:09Oh, yeah, yeah, this one. This one.
- 7:11Real data, Claude and Anthropic picking it up and then someone
- 7:14else coming in with a fishing net into their bucket and taking
- 7:17the data. This is pretty much what is
- 7:20happening. What do you guys think about
- 7:23this? It's interesting to discuss
- 7:25because if Anthropic has trained their models on everyone's data,
- 7:30then technically do the Chinese models have the same right to go
- 7:35out and train on everyone's data?
- 7:38But why they or do they not? I mean, I think nobody is
- 7:42deciding in here who's right and who's wrong.
- 7:45But the fact that now, because I think they put a lot of effort
- 7:48in scraping all the data, the Chinese model don't want to go
- 7:51through all that effort. Sure want to go to the easy way,
- 7:54Yeah, and get it from Anthropic. And I'm thinking what is now
- 7:57stopping, for example, the Russian language model or a
- 8:00Indian language model to do the same thing?
- 8:02I think nothing because I think they wasn't.
- 8:05Deep to Seek train the same way. I believe I may have read
- 8:08something very similar or DeepSeek also trained one of
- 8:11their models using the exact same technique.
- 8:14So it's not new. But the fact that Anthropic CEO
- 8:19has cried out about this and said this is a huge risk for
- 8:23America has gotten a lot of people divided in the comments
- 8:26saying that you know what, you've done this first with
- 8:28people's data. You can go ahead and cry and say
- 8:31other people are doing the same with my data because at the end
- 8:34of the day no one gave consent this.
- 8:36Is a funny 1. Yeah, everyone's like, hey, you
- 8:39took, you took. Data.
- 8:40But I'm thinking does it make that for other smaller AI
- 8:44company doesn't make it easier for them now to train their
- 8:47models And in three years time, four years time we will have
- 8:50different LLMS that are trained on these big, big LLMS with the
- 8:54same capability. We doubt them have to go through
- 8:57every single data on the Internet, you know what I mean?
- 9:01We have a lot of mini powerful AIS from different countries now
- 9:05that they can perform not as good as cloud, but maybe 80
- 9:08percent, 70% close to that capacity.
- 9:12I see your point. I guess my concern here is this.
- 9:16When it comes to data training, it must be understood that AI
- 9:23models, these LLMS per SE, need data to become more intelligent.
- 9:29So it is, I think it's it goes without saying that they have to
- 9:35be trained on stuff that humans have created, right?
- 9:38Whether it's an image or a video.
- 9:41Or a code or? Text or audio or code.
- 9:43So all models have been trained on human data.
- 9:49Some companies claim that they have ethical sources, just like
- 9:53food companies claim that they have ethical sourcing of
- 9:57ingredients. Of their coffee beans.
- 9:59But it is. Almost impossible for an average
- 10:02human being to be able to verify the sources.
- 10:05Yet we consume those foods and put them inside our body,
- 10:10trusting the labels. And in 20252026, so many have
- 10:15awakened to the concept that a lot of these companies that sell
- 10:19us food and snacks lie to us. The majority of things that we
- 10:23see in the supermarket are not good for our bodies, yet the
- 10:27front label says they are. But if you actually look at the
- 10:30ingredients where they cannot lie, you'll find out that they
- 10:33are bad for your body. I think the same thing can be
- 10:37applied here where some companies are claiming that
- 10:41they've ethically sourced their data and others don't even talk
- 10:45about it. But ultimately, how can we ever
- 10:47know? Because in my mind, when I see a
- 10:50super intelligent model that can create anything, they must have
- 10:53been trained on the majority of the available data publicly,
- 10:58privately, again, with consent, without consent is such a big
- 11:02Gray area. And we can see that news like
- 11:05this are popping up over and over again.
- 11:07With that said, one thing that is interesting in what Anthropic
- 11:12said, if you take a patriotic, nationalist approach, let's say
- 11:16I'm an American, right? The defense here is that this is
- 11:20a national security concern for Americans.
- 11:23That's the claim here. But they didn't scrape only
- 11:27American data. They scraped the data from the
- 11:29entire world. No, no, no.
- 11:31What I'm saying initially, what they did right?
- 11:33I'm not talking about anthropic. What Anthropic did, what
- 11:36Anthropic is claiming, is we need to do something about this.
- 11:41The policy makers, the governors, other AI companies in
- 11:44the US like XAI and Open AI must do something about this.
- 11:49We should come together as a group because it's a national
- 11:52security concern for another country to steal from our
- 11:56models. They're trying to group all the
- 11:57companies together, which is a smart strategy.
- 12:00But my question is, all the CEOs are fighting.
- 12:03Elon is not going to sit down with him, Sam.
- 12:05Altman is not going to sit down with him, Sam.
- 12:07Altman and him didn't. Even hold hand in the.
- 12:09Elon is talking him on X which begs the question say fine
- 12:15Anthropic took data, ethically or not doesn't matter in this
- 12:20case. What I'm talking about is, is it
- 12:21actually a national security concern and should Elon and Sam
- 12:26Altman and the government come together to actually form a
- 12:29policy to ensure this doesn't happen again?
- 12:31How can they stop it? They I think they would come up,
- 12:35the Chinese model will come up in new ways of extracting the
- 12:38data because these models are available to people worldwide.
- 12:42They're using the APIs, right? Because when I was reading about
- 12:46the the distillation, basically what happens they are using the
- 12:50same APIs and they are training their own data from that.
- 12:53So when they're open to public, so Chinese model can use their
- 12:57models basically. This question is beyond me.
- 13:00I have to do more research on this but I'm curious why they
- 13:04ended the Last Post with this that we have to come together as
- 13:08a community and fight against this.
- 13:10I wonder what would? Be because I think if they
- 13:12didn't post the last three nobody would give a shit about
- 13:15the 1st 2 even with the last one.
- 13:18Last one is like guys, national security and stuff.
- 13:21Please let's think of this. But you see the comments, people
- 13:24actually don't care. Yeah, in the comments.
- 13:27People don't care. I'm generally curious to see
- 13:30what this means for US as a whole being in race in an arms
- 13:35race. With China and Russia.
- 13:37With China and Russia and other companies when it comes to AI.
- 13:40Interesting. Let us know what you guys think
- 13:42about this one specifically. So let's go to the next topic,
- 13:46which Netflix is now using Comfy UI for multiple Netflix games.
- 13:53Of course, the names are not being disclosed because of Ndas
- 13:58or the games are still under ongoing development, but the CEO
- 14:03of Comfy UI went on LinkedIn, explained how Netflix is using
- 14:08Comfy UI in their pipeline and in their workflow.
- 14:12The moment Mian Farah saw this news, we had a discussion about
- 14:16what does this mean for the entertainment industry and what
- 14:19does this mean for people who are in the industry working on
- 14:22titles like games or even on movies?
- 14:25Paras, what do you think? If you allow me, before we get
- 14:29into the debate, which is going to be cool, I want to introduce
- 14:32people to the concepts of Comfy UI and Netflix.
- 14:35Yes, for sure. In case they may not know for
- 14:37sure, Comfy UI is an open source platform that allows you to
- 14:42using a node system, create images, videos using AI and you
- 14:47can connect your platform to models that are available
- 14:52locally on your computer for free.
- 14:54Or you can connect to other models such as Nano Banana Pro
- 14:57and create the most complex node system to then generate images
- 15:01and videos. It's one of the most popular
- 15:03platforms used alongside Higgs field and free it will.
- 15:07Give you V more control. A lot more control.
- 15:10Granular control, consistency, and you can use it for free.
- 15:14This is the biggest point. Because it's open source, yes.
- 15:17So what Yoland and his team have done is great for the community.
- 15:22However, I do have a fundamental problem with this post and the
- 15:25message and I'll get to it. So the post starts by the Co
- 15:29founder Yoland saying that they're now powering the
- 15:33production behind multiple Netflix game titles.
- 15:36Now I personally have never played Netflix games so I had to
- 15:39look this up. What is Netflix games?
- 15:42It's a gaming service offering subscribers 90 plus mobile games
- 15:46at no extra cost such as Dead Cells, TMNT, Red Dead
- 15:50Redemption, the Netflix Edition and these games.
- 15:54Why they came up with the idea is if they have incredibly
- 15:58popular films or series on Netflix, they can immediately
- 16:01turn them into adaptations of video games.
- 16:04But the biggest challenge has been that over the years, we
- 16:07know game development is incredibly costly and takes a
- 16:10lot of time and a lot of manpower.
- 16:13So what can they do to speed up the production so that once
- 16:19Squid Games go by, they can come up with a game within six months
- 16:24and not take three years? Because by that time Squid Games
- 16:27hype has died. This is the reason why AI has
- 16:30become so interesting. But my biggest challenge is in
- 16:33the messaging. So what series Entertainment,
- 16:36which is the company, a game dev studio, which is AI native, that
- 16:40is piring the use of pioneering the use of Gen.
- 16:43AI. The reason why they're using it
- 16:46is because they needed to scale production across multiple
- 16:50Netflix titles. Fast scale fast, usually without
- 16:58AI. Let's just talk about humans.
- 17:00When we have movies and series and games that scale fast, we
- 17:05end up with slop. Quality drops.
- 17:08Precisely. Let's talk about the first
- 17:10Marvel movies versus the latest ones.
- 17:13I'm pretty damn sure a lot of you guys can agree that the
- 17:17quality of the storytelling, the VFX, has dropped dramatically.
- 17:21Not because it's the fault of the VFX artist, but because the
- 17:25production companies and the investors came in and said we
- 17:27need to spin. Up my third timeline, yes, and
- 17:30multiple shows in a year of. Course.
- 17:32And that was prior to AI, yes. So imagine now we have AI slop
- 17:38and human slop added on top of each other.
- 17:40You'll get like a mega slop. That is what I'm against.
- 17:44If they use AI the right way to create better work and of course
- 17:49speed up but not try to you know if they.
- 17:52Invest 80 times faster, less than 200 million games in two
- 17:55weeks. That's my problem.
- 17:57I I believe that you can use AI in great ways to enhance your
- 18:01story, tell a better story, have a higher quality asset, make the
- 18:04game better. But if you want to make 20 games
- 18:07instead of 1 using all generated asset and generated stories,
- 18:11nobody will enjoy. Those games, thank you, that's
- 18:15what I meant. Because ultimately the biggest
- 18:18investors and the board who do not understand the fundamentals
- 18:23of game development and just care about the profit, they will
- 18:28be the ones pushing for more. And when they know there's AI
- 18:31involved and things can be done faster and with less amount of
- 18:33people, usually what tends to happen is them trying to squeeze
- 18:37these studios to create more and more and more content, therefore
- 18:41sacrificing quality. I'm all for using technology.
- 18:45I'm all for using new tools. I'm all for using.
- 18:47AI, but not to sacrifice so much, but to really build
- 18:51something better, build a better story, build better characters.
- 18:55This is why this story is quite interesting.
- 18:58It is important to understand as creatives, as artists, that this
- 19:01is happening. They are going to use these
- 19:04tools. So where do you want to position
- 19:06yourself and what do you think about it?
- 19:08Let us know in the comments. I'm excited to see the results.
- 19:11Actually, you really want to try it.
- 19:13Let's see how it looks, how it feels, and again, does it help
- 19:16the story? Does it help the quality of the
- 19:19game? If it does, why not?
- 19:20Yes, and if it doesn't, then we have to rethink this whole idea.
- 19:24Exactly farhat OK. This is going to be interesting
- 19:27because we are going to call in for a guest to join us right now
- 19:32before we move on to the next topic.
- 19:34This guest we have briefly discussed in our previous
- 19:38episode because we looked at his creation a photo about the
- 19:43adoption of AI in the world today.
- 19:46And that everybody is convinced that AI, we are in the bubble of
- 19:50AI. But clearly if you look at the
- 19:53numbers, we are far, far early in it.
- 19:56Let's have a Damian on the call. So, Damian, we talked about
- 20:01your. Post on the previous show, and I
- 20:04think it really opened our eyes when we looked at it on X for
- 20:07the first time. And what we really wanted to
- 20:10talk to you today is what made you create that post?
- 20:14Because I think it created a lot of awareness in people.
- 20:17I see people like Noah. I even use it in their article.
- 20:20So we wanted to get into the details of it. 100% and I think
- 20:25it's everything is so FOMO, FOMO FOMO that I think it was one of
- 20:30the first times it's almost anti FOMO right?
- 20:32We almost live in an echo chamber when it comes to using
- 20:35tools, new launches, etcetera, etcetera.
- 20:37And it almost brought people back to reality right and the
- 20:39numbers in itself say that 84% haven't used AI .4% even pay for
- 20:44AI bar code cursor, those types of tools.
- 20:47So I think it's almost like an anti shock that had people like
- 20:50wow. So I'm I'm in this world 24/7.
- 20:52I'm almost people are saying we're in a bubble, but we're
- 20:55actually we are the ones in a bubble because the majority of
- 20:58the world aren't even using these tools.
- 21:00Absolutely, man. You know, we talk about this all
- 21:03the time because the majority of our time is spent on the
- 21:06Internet and we're scrolling X, we're scrolling Instagram and
- 21:10these platforms know us very well and they know we're
- 21:12interested in the technology. So all it's showing us is new
- 21:17updates, new tools. And we wake up after six hours
- 21:20of sleep, which is not a lot. I mean, I talk to most of my
- 21:22friends, they sleep for 8-9 hours, we're doing 5 or 6 hours
- 21:26and I still wake up feeling I'm left out of all of the new
- 21:31updates people have already made.
- 21:32Fucking. Videos covering like if you're
- 21:34not using this, you're you're behind if you're not using or
- 21:37like this one person man built an entire industry using open
- 21:40cloud. This one person man built an
- 21:42entire company using cloud code. So it's it's very hard to
- 21:46navigate it actually. 100 percent, 100%.
- 21:50And I think we, we, we always are in the timeline in the
- 21:54algorithm, right. So it's like you get that FOMO,
- 21:57you wake up. It's like, what do I do?
- 21:58Well, I feel like I need to do something.
- 21:59I feel like I need to do something.
- 22:00The world's progressing the world's progressing.
- 22:01But in reality, there's, there's people are aware, obviously,
- 22:05like we'd be sitting here lying to say people aren't aware of
- 22:08AI, but there's clearly an adoption gap, right?
- 22:10Both from the consumer implementing it into their
- 22:13day-to-day life to, to, to better it, right?
- 22:16And then from a business perspective, that's obviously
- 22:18what everyone's trying to, to figure out is how can businesses
- 22:21actually implement AI and, and, and not just be FOMO, FOMO,
- 22:26FOMO. Right.
- 22:27I think that's very important because the moment you are in
- 22:30that FOMO state, you cannot create very well because you're
- 22:33trying to run and jump after everybody else.
- 22:36Just like how horses have blinders on so that they are not
- 22:40looking at other horses racing so they can focus on racing
- 22:42themselves. Humans need to have a blinder on
- 22:47as well. Because unless you don't, you're
- 22:51constantly going to be looking at all of these people building
- 22:53things and then you're going to change your ideas every single
- 22:55day and never actually land on something.
- 22:57And what I love about your posts, it just gave us a sense
- 23:00of relaxation that OK, we're not late, we're still early.
- 23:05There's a lot of hype. We are in a bubble so we can
- 23:07actually relax and focus on what we're building, not have to jump
- 23:10so many freaking times. What I want to ask you is this,
- 23:13because I'm sure when you've done the research, you have to
- 23:16look a lot deeper into how industries are using AI, how
- 23:20yourself you're using AI. Where do you see the biggest
- 23:23opportunity now that people like yourself, us and the listeners
- 23:27are aware of this technology? 100% And I think as I was saying
- 23:31earlier, attention is there now it's about real adoption, right?
- 23:34And I think the problem with AI, it's a lot of it's nice to
- 23:37haves, right? There's no real proper use case
- 23:39there is. But what people are selling in
- 23:42terms of actual implementation, it's it's not actually moving
- 23:46the needle forward for a lot of these mid markets or SM BS or
- 23:49boring businesses, as I like to call it on on my feet, right?
- 23:51So as adoption accelerates is how can we actually have proper
- 23:55conversations of implementation, right?
- 23:58Because people, like I was saying earlier, people are
- 24:00aware, right? But how can we actually optimize
- 24:03A workflow? How can we help client intake?
- 24:06How can we better a workflow beyond just this FOMO and hype
- 24:10and Lai this and AI that? How can we get real now?
- 24:13How can we actually genuinely help a business as an agency?
- 24:16How can we generally help a business optimize workflows
- 24:19right beyond just the hype? I wanted to ask something else
- 24:23also, a lot of businesses, when you go into them and when you
- 24:26ask them, how do you use AI? Most of the stuff is, oh, I use
- 24:29it. I use ChatGPT to improve my
- 24:31e-mail. I use ChatGPT to create
- 24:33documents. But I think with AI you can have
- 24:35a lot more benefit for your business.
- 24:37You can automate a lot of process, You can improve a lot
- 24:41of pipelines within your business, whether you're a small
- 24:43business or a large business. How does a company you think can
- 24:47transition from this ChatGPT phase for the e-mail to actually
- 24:51using these tools in their business?
- 24:54100% and I and I think people sell it is the end game, right?
- 24:59And the problem with AI tools, new agents, etcetera, etcetera
- 25:04is you don't know what's gonna happen in, in, in three months,
- 25:06three days, three years, right? So a lot of these agencies, a
- 25:09lot of people push for that end result that, that full stack
- 25:12system, but in reality, it's finding quick wins, right?
- 25:15How can you genuinely help a lawyer save 5 hours a week,
- 25:18right? 10 hours a week if his billable
- 25:19hours are 100 bucks, it's $1000 a week, $56,000 a year, right?
- 25:23So there's clear ROI there, right?
- 25:26Finding quick win, whether that's to a workflow, an agent,
- 25:29etcetera, etcetera is is where the ROI is for these businesses,
- 25:34right? It's not the whole let's go
- 25:35fully AI first because you don't know what's gonna happen in
- 25:38three months. So it's about finding these
- 25:40quick wins, allowing these businesses to understand, hey,
- 25:43AI is more than just asking ChatGPT about the weather or
- 25:46writing emails. You can connect it to tools.
- 25:49You you can do a whole lot. And it's about finding those
- 25:52those bottlenecks, finding those inefficiencies in the businesses
- 25:55and then automating those workflows.
- 25:57I agree with you on that. I think the biggest challenge
- 25:59I've had, especially now that we are a lot more active on X,
- 26:03looking for information for the show.
- 26:05A lot of people who create content around AI are simply
- 26:09guessing at what businesses need.
- 26:11And for instance, use open Claw and say open Claw is going to
- 26:13automate your whole business, but they don't actually go and
- 26:16talk to the lawyers or the real estate agents and see what the
- 26:19problem actually is. Sure, Open Claw can do a lot of
- 26:22things, but you're just guessing at what their problems are and
- 26:26saying this is going to automate, let's say this is
- 26:27going to automate lawyers jobs. I think I will see a great
- 26:32belief in AI by these different industries when they actually
- 26:35have people they can trust who are going into these businesses,
- 26:38as you mentioned, looking at one problem, not the whole fucking
- 26:42business, one problem that is bothering them and trying to
- 26:45solve that one problem and that's when we're going to see
- 26:47real adoption. Are you working on that
- 26:50specifically yourself in terms of educating people or with an
- 26:54agency? What are you doing in that sense
- 26:56to help educate? Cuz with this post, one photo,
- 26:58you're fucking educated a whole lot.
- 27:00I'm wondering what else you're doing on the side.
- 27:02For sure. And I think you mentioned it
- 27:04earlier, people don't know what they don't know, right?
- 27:07So there needs to be some sort of education to allow people to
- 27:10understand, hey, this is what's possible in this industry.
- 27:13This is what we're doing with this client.
- 27:15This is what's actually possible.
- 27:17Instead of it being FOMO, FOMO, FOMO, FOMO, let's get real and
- 27:20understand workflows, right? So as an agency, I mean, in
- 27:24short, we help companies go AI first.
- 27:25That doesn't mean anything at a high level.
- 27:27But when you actually step into a business and find a quick win,
- 27:31right? Going back to the lawyer
- 27:32example, you help them finding take whatever it is, right, find
- 27:35a way to buy back 5-10 hours a week.
- 27:38Once you're involved in the business, as you were saying in
- 27:40terms of expertise, you start understanding workflows more.
- 27:43You need to be ingrained into those workflows.
- 27:44Otherwise you're just guessing. You're just guessing, right?
- 27:47And that's the problem with agencies is it's a lot of churn
- 27:49and burn one off build. They build up an agent that in
- 27:53three months is not used anymore.
- 27:55It's because there's no true transformation.
- 27:57And as a business it's, it's you need to transform the business.
- 28:00They can't just be, oh, let's do an agent right?
- 28:02Like you have to genuinely transform the business, you have
- 28:05to change the mindset and that comes through education, right?
- 28:08And hopefully I think that chart did that.
- 28:11I think the education part that you mentioned is super
- 28:14important. I think a new category of job
- 28:17would be people who go to these companies, educate them on what
- 28:21they can improve. Like the consultant that we have
- 28:23today. They don't have to necessarily
- 28:25implement anything. They can just come in, do the
- 28:28education, tell them where they can, you know, optimize their
- 28:32workflow and gain their trust. I think one of the biggest
- 28:35problems now we have with small businesses, especially brick and
- 28:38mortar, is the lack of trust. OK, if I give everything to this
- 28:41black box to do it for me, how can I check?
- 28:43How can I verify? But if you have a trusted party
- 28:46that comes in, an orchestrator or an architect that can say hey
- 28:51Mr. Lawyer or Hey the Clinic of Beauty, you have to do this in
- 28:55order to save time on booking appointment, then probably they
- 28:58can trust them more and they can get someone else to do the
- 29:00services for them. 100% and we've seen that we've had a few
- 29:05clients come in and say, hey, come, come for us full time, be
- 29:08that AI operator, right? And I think beyond just
- 29:10knowledge, right, knowledge of these tools, and as someone who
- 29:13lives in the algorithm and lives in the timeline, I, I see all
- 29:15these different workflows, etcetera, etcetera.
- 29:19The fact is these tools now allow you to have the output of
- 29:2210/15/20, right? So beyond just consulting a lot
- 29:25of these AI operators, you're able to execute right on a lot
- 29:28of these smaller work flows using tools like Quad code
- 29:31cursor N 8 N, etcetera, etcetera.
- 29:32You are able to also offer execution right?
- 29:36Beyond just that initial consulting, right?
- 29:37Because as I was saying earlier, once you're actually involved in
- 29:40these work flows as an operator, you go, OK, well, if we automate
- 29:43this part, we can automate this part, OK, the HR department gets
- 29:46500 a thousand resumes a month. How can we find a way to buy
- 29:50back 50% of time right? How can we stream quicker, How
- 29:53can we filter quicker? Etc etc.
- 29:55So beyond just the consulting, there's a whole executional
- 29:58layer in that these tools allow people like myself to have
- 30:01outputs of 10/15/20 people. And I absolutely agree with you
- 30:05on that. Damian, what else are you
- 30:08working on now? Because I'm looking forward to
- 30:10see what kind of posts you have coming up.
- 30:11We have your profile on the screen for people to go ahead
- 30:14and follow. Is there any new research that
- 30:16you're working on that we should be looking forward to?
- 30:19A. 100%, we have a lot coming. Obviously we run an agency, we
- 30:23implement AI into companies that takes a lot of education.
- 30:26So really, really working on finding ways to bridge that
- 30:28adoption to attention gap and yeah, just keep pushing AI
- 30:32forward and, and helping businesses understand that
- 30:35there's a lot of things that people don't know, right?
- 30:38And educating them on what they don't know, then obviously,
- 30:42obviously offering implementation on the back end.
- 30:44What industries do you see the most open to all these new
- 30:49technologies? I would say boring.
- 30:52I'd say boring industries, right?
- 30:53The H FAC guys, the plumbers, the lawyers, the real estate,
- 30:57the traditional industries, right, who live on legacy
- 30:59systems, right? That is in itself an issue,
- 31:02right? Getting these companies off of
- 31:03those legacy systems and or building around those legacy
- 31:06systems. But that is an industry as a
- 31:09whole, boring, right? That's very high level but
- 31:11boring industries, traditional industries are ripe for
- 31:14disruption and that's as an agency who we're targeting.
- 31:17Are they open to having discussions with you guys as an
- 31:21agency usually? 100% And we have it fairly set
- 31:25up in a way where a lot of deletes coming in are inbound.
- 31:28So they're a lot further down the sales cycle and knowledge
- 31:31cycle than most, right? And obviously being on X and
- 31:36seeing all these different agents, workflows, etcetera,
- 31:38etcetera, they have a little more of an idea of what they
- 31:40want. But to answer your question,
- 31:42yes, right. It is harder to to to get a
- 31:44lawyer, not a lawyer, a plumber to implement AI, right?
- 31:47Because they just want to do work, right?
- 31:50It's hard to get them to fully change what they do to move
- 31:53their day-to-day. But like I was saying earlier,
- 31:55if you can find quick small wins, that unlocks a bigger
- 31:58picture for them, right? So it is truthfully a hard
- 32:01conversation to have. But if you frame it based off
- 32:03ROI, hey, we can save you 10 hours, not, hey, here's an agent
- 32:06because that's what they get pitched for, right?
- 32:07But here, here we can save 10 hours here we can sit, we can,
- 32:11we can cut costs here, right? We can optimize this, right?
- 32:14Instead of selling the tools, you sell the outcomes.
- 32:16And especially with these boring businesses, that's that's all
- 32:18they really understand is outcomes instead of the actual
- 32:21agents at technical side, if that makes any sense.
- 32:24I think that is super important, talking about the outcome rather
- 32:27than the features. You can, you can start naming
- 32:30all the tools that you're going to use are different elements.
- 32:32They don't. They don't need to.
- 32:34They don't need to. Know if you tell a plumber
- 32:36you're going to be using open claw, they're like don't talk to
- 32:38me, that sounds dodgy. But if I, if I want to say maybe
- 32:412 hours of your day or I'm going to reduce your cost by 10%,
- 32:45that's a totally. Different conversation, like all
- 32:47of the people that are going to book you will now talk to this
- 32:50chat bot and then you automatically get a report of
- 32:53whose house you need to go go to for your plumbing services.
- 32:55And you don't need to take the phone anymore and.
- 32:56You don't need to take the phone anymore.
- 32:57Like that's how you talk to them and that's how they're gonna be
- 33:00like, OK, you know what? That sounds good.
- 33:02But if you start building out like feature lists for them,
- 33:04this is what we're gonna build for you with these open AI model
- 33:07or this cloud, you're like, no, no, no, no, don't.
- 33:09These guys don't care about that.
- 33:11That's your job, right? You wanna focus on just telling
- 33:15them how you're gonna make them more efficient.
- 33:17And that's very interesting, man.
- 33:20Damien player is the tagline for your X handle.
- 33:24So for people who want to go in and follow and take a look at
- 33:28what you've got there, hopefully more of these anti FOMO pose to
- 33:33give people a real perspective of where we are in the world.
- 33:37I would love to see more of that.
- 33:39Damien, thank you so much for coming on the show of.
- 33:41Course I appreciate it. Thank you guys.
- 33:43Have a good one. Ciao.
- 33:47All right, so Faraz, I think next we wanted to cover another
- 33:52topic which was quite important. I saw it on X, there was a tool
- 33:57on Comfy UI. Are you not going to give
- 33:59credits to who saw it first? You saw it first.
- 34:02I saw it first. I think I saw it. 1st I sent it
- 34:04to Pratt. You send it to Pratt.
- 34:06I saw it before Pratt. We saw it. 1st Pratt is grabbing
- 34:11the mic be real far as I saw it. First would be far as how does
- 34:16he go? How does he know you?
- 34:19Can go and you can trace it back.
- 34:21No, but I saw it first. It doesn't mean that if you send
- 34:23it OK, you did it, thank you. Then you can do the.
- 34:26Intro, Give me the credit. You can do the intro.
- 34:28Thank you. So we've got an interesting new
- 34:30model. We're gonna look at Quinn's.
- 34:32Small that great anyways. Angle 3D camera control.
- 34:38In simple terms, you provide this model with an image below.
- 34:42You'll have a 3D camera that you can rotate in 3D space to
- 34:49generate another angle of that same image.
- 34:52We've tested this with a couple of different images.
- 34:55It works really well with content that is publicly
- 35:00available online and is popular like a Bugatti.
- 35:04How many images of a Bugatti or videos of a Bugatti are online?
- 35:07Many, so it knows exactly what each angle looks like and it
- 35:11generates really good results. We tested it with the portrait
- 35:15of Mona Lisa, trying to see what Mona Lisa looks like from
- 35:19another angle, and clearly you can see that it failed.
- 35:22It generates the angle, but it's wrong.
- 35:25It doesn't work because we only have one image of Mona Lisa.
- 35:28It doesn't know. There is no enough data for
- 35:31that. And it's not smart enough to try
- 35:33to generate and predict what those pixels may look like.
- 35:37Pratt, you may be able to generate the Bugatti for us live
- 35:41and another Mona Lisa if. We can while Pratt generates a
- 35:44Bugatti photo. To be very honest, the reason I
- 35:47got hooked when I saw the demo on X by Framer is that this is
- 35:53the exact control that the artists look for in the future.
- 35:57I actually not really concerned about the quality of this model,
- 36:01what it can deliver right now. But I can imagine the biggest
- 36:05problem that we have in the tools like Hicksville Free Pick
- 36:07or any image generation tool that is that if you want to
- 36:11change the angle, you have to talk to it.
- 36:14You have to prompt it and say, hey, I want a 45° top view or I
- 36:17want a bottom view. Here you can actually have the
- 36:21control of the camera in the 3D space.
- 36:24That is exactly what you have in a 3D software like Unreal Engine
- 36:27or Blender. Add another layer of, I don't
- 36:32know, better image generation or even video generation.
- 36:34This is going to be super powerful in the future and this
- 36:37is what I would love to see in the 3D softwares.
- 36:40I couldn't agree more with that. Can we go ahead and just show
- 36:44everyone one more time the Bugatti model Pratt.
- 36:47So This is why it's interesting. I want to get.
- 36:50Into a little bit more yeah. So the input image is the one on
- 36:53the top left right, so. Quinn Image Edit 3D camera
- 36:57control. On the left we have the input
- 36:59image. That's an image that we have
- 37:00from Google, which is a Bugatti and we want to create another
- 37:04angle with perfect consistency. We can rotate the camera around
- 37:09the car. We do not know what the camera,
- 37:12we do not. We do not know what the car
- 37:14looks like from the back based on the image.
- 37:16But we can generate it now and it should keep the consistency
- 37:20of the hue in the background, the car and the shape of the car
- 37:25and look at that. It did create the top down view
- 37:27of the car. Can we go to another angle?
- 37:32This is fascinating look at that.
- 37:35Back view, yeah. That's the back view of the car.
- 37:38Can you go to the side angle? Side angle of.
- 37:43The car, Yeah, indeed. Put the camera a little higher,
- 37:46yes? OK, let's see.
- 37:53There you go. Perfect.
- 37:54I mean perfect. That's it.
- 37:56No, but it keeps the background, it keeps the floor the same,
- 37:59keeps the car the same. It's not changing colors of the
- 38:03car or anything, but it fails miserably with Mona Lisa.
- 38:06Again, I really love the 3D control.
- 38:10I really wish that we can have these 3D controls in all the AI
- 38:13generation tools because that would give the artist the most
- 38:17amount of flexibility to tell their stories.
- 38:20Imagine you're doing a commercial.
- 38:21You really want to have that specific camera control.
- 38:24What the fuck? Actually, we do have Mona Lisa's
- 38:27nose. Can you zoom in there?
- 38:29With Mona Lisa's flat, maybe she was flat.
- 38:32It's an image. It's trying to really guess what
- 38:34Mona Lisa looks. Like but we don't have any data
- 38:36on that so that's why it's failing.
- 38:38Oh my God, this. Is funny Faraz, you haven't
- 38:41silenced your phone. Or is it mine?
- 38:44It's your laptop on the show. You haven't silenced your
- 38:48laptop. It's closed.
- 38:50Yeah, but it's. Oh, that's what Mona Lisa looks
- 38:53like from the back. Not bad actually.
- 38:57We generated legs for it. Mona Lisa was seated in the
- 39:00photo. Can you generate other angles of
- 39:01Mona Lisa and can we also try one more product after you hit
- 39:06generate? No, no, don't do, don't.
- 39:09Go low angle. Go low angle.
- 39:11Go low angle. Yeah, generate.
- 39:15Can we find another item, like maybe a hoodie or something?
- 39:21No, it's just it's not working. With the, I think with the
- 39:24painting, it doesn't work. We find something in the real
- 39:26world, maybe like a Nike. Because the painting is already
- 39:292. D2D.
- 39:30That's why I think it's having a hard time.
- 39:32We're going to try one more image here, and whilst Pratt
- 39:36finds that image, I do want to quickly take your attention away
- 39:40to the last episode. For anybody who watched and
- 39:43listened, we covered Trellis 2.0, which is Microsoft 3D acid
- 39:47generation model, and we viewed it on Hugging Face, showing you
- 39:51guys what it can generate. And there was a comment by
- 39:53Michael Habeeb which was saying, hey, you can actually get
- 39:58Trellis on Comfy UI and you can use it.
- 40:02Without the permission of the company Microsoft, because
- 40:06that's what we thought we had to do.
- 40:08He was right. We tested it on our computer and
- 40:10it worked marvelously. And because we have control over
- 40:13the samples. Whoa, this is that's really
- 40:16good. That is really good.
- 40:18Because again, Nike shoes. Are 3D.
- 40:20This is in real life, no? No, no, what I'm saying, there
- 40:23are tons of Nike shoes photos on the Internet.
- 40:27You go to the Nike website, you will see from different angles.
- 40:29That's why they have all the data.
- 40:31I'll just quickly finish my last point.
- 40:33We tested Trellis 2 point O on Comfy UI.
- 40:36It works locally on your computer.
- 40:38Generations are fast and remarkably well done.
- 40:42We will have to wait until another episode because we want
- 40:46to do a full comparison of Trellis 2 point O and Honeyuon
- 40:50and really see how much we can push them and what the quality
- 40:53of the models are. And when we do, we're going to
- 40:55bring it to the show and show you guys a full comparison.
- 40:58And that way, Michael Habeeb, you'll get to see the entire
- 41:02display here. We're just going to wait to make
- 41:04sure we finish that. I want to see one more angle of
- 41:07this shoe. I actually am curious about one
- 41:08thing. What?
- 41:09And Nike and Bugatti are famous, so there's tons of data.
- 41:14So it should get it right and it did, right?
- 41:16No, not really. Why not?
- 41:17You see the OR the jumpman? Yeah, the jumpman.
- 41:20Is the jumpman logo is messed? Up but again, it's I just like
- 41:23the 3D control. I don't think this is something
- 41:26that you can use really for your final work, but the 3D camera
- 41:29control is a nice feature. I'm pretty.
- 41:31Sure, eventually all the AI models are gonna have it in.
- 41:33Hopefully and it will help a lot with video too.
- 41:35By the way, if imagine for your camera movement we are talking
- 41:39about photo here. But for video, if you can draw
- 41:42your camera movement epic games on real engine, please bring
- 41:45that with the AI generated models within the engine itself.
- 41:49Then you will have a lot of control.
- 41:51Absolutely, I think that's what is gonna happen in the future.
- 41:53Anyways, I'm curious, what do you guys think about this?
- 41:57Quinn image edit? This is just the very beginning
- 41:59of us having a little bit more control of the generations that
- 42:05we have with AI. This is of course, as Farah
- 42:07mentioned, going to allow us to be a lot more granular with the
- 42:10detail of what we create, which is what artists want eventually.
- 42:14This is definitely going to be something that amazes a lot of
- 42:18people when it is more advanced and works properly and has 100%
- 42:23consistency. Yes, exactly.
- 42:25Ladies and gentlemen, that brings us to the end of this
- 42:30episode, which makes me sad because I do love hanging.
- 42:32Out. We'll be back on Friday.
- 42:33We are here three days a week. Yes, you know it right, Three
- 42:36days a week, Monday, Wednesday, Friday on X, Spotify and
- 42:40YouTube. Don't forget to subscribe on our
- 42:42channel on YouTube and on X and Spotify if you are watching from
- 42:46any of these platforms because that would help us reach more
- 42:50people and educate more people about what is out there and how
- 42:53they can use all these tools and technology to improve their
- 42:57career, their workflows, whatever they are working on
- 42:59basically. And the show is free, so that
- 43:02does actually help the show because it's free and it's not a
- 43:05paid show. We appreciate your support with
- 43:07a like, a comment and a share and a subscribe on YouTube.
- 43:11Until the next episode. Ciao.
- 43:13Ciao.