Latest / Bad Decisions Podcast / The Most Powerful AI MODEL Leaked 👀
Transcript
- 0:00Anthropic accidentally leaked information about their most
- 0:04powerful AI model. Researchers just released a
- 0:07digital human brain model that can accurately predict how we
- 0:12respond to what we see and what we hear.
- 0:15We now finally have full control over generating AI videos and
- 0:21images. Let's talk about it.
- 0:23What is our beautiful people? Welcome back everybody to
- 0:26another episode of Bad Decision Podcast.
- 0:28Where we bring you guys the latest and the greatest and the
- 0:31hottest in the world of science, technology, AI and business.
- 0:36I realized far as I think every single time we scream into the
- 0:39microphone when we. Start the show but some people
- 0:41are complaining about the show not being loud enough so I'm
- 0:44trying to shout on the top of my.
- 0:46We just shout in the 1st 5 seconds and then the rest of you
- 0:49are not shouting. I am relatively loud though, as
- 0:51a human being for me, sitting beside you the entire.
- 0:54Time my gain is high by default, so I'll try.
- 0:58To. Whisper for the rest of the
- 1:00show. Welcome back everybody, it's
- 1:01Friday, end of the week. And what a what a beautiful.
- 1:05Day it is, Yes, and. It's right before the weekend.
- 1:09Yes, it's time to chill. Time to relax, play some video
- 1:12games. And again you.
- 1:14And your weekend? Yeah.
- 1:15What is your weekend plans? I'll tell you what we're gonna
- 1:18do. We're gonna play some video
- 1:19games. I haven't played video games for
- 1:21the longest time and I just started again 2 days ago.
- 1:24What games are playing? I'm playing on my phone, I'm
- 1:26playing wild drift, which is. League of Legends, roast them.
- 1:28League of. Legends for the phone.
- 1:30I used to play league. Don't get me wrong, it's just I
- 1:33do not have enough time right now to get back on the computer.
- 1:35Those games are long. You're really scared of people
- 1:38roasting you, right? No, I you are just explaining
- 1:40yourself. I could just say I love the
- 1:43game. And I console players hate PC
- 1:45players, PC players hate console players, and both of those
- 1:48groups hate mobile players. So you are the 1, you are the
- 1:51mobile. I had all of them and I'm not
- 1:54taking any sides. I'm playing mobile only because.
- 1:57Why are you turning red? Why are you turning red?
- 2:00Am I turning red? Why you pulled up the
- 2:02saturation? I see what you're doing on OBS.
- 2:05But OK today. We've got we've got some
- 2:07exciting stuff today. We have, we have and we have a
- 2:09live demo at the end of the topics that we are covering.
- 2:12One of the topics that we are covering, it's going to have a
- 2:15live demo. So stay until the end.
- 2:17And if you guys are listening, make sure towards the very end,
- 2:21the last topic that we're covering, try to watch it
- 2:24because the live demo will demonstrate a very new update
- 2:27that can solve the control problem we've had with AI
- 2:31content generation. Definitely recommend you guys
- 2:34watch that aspect at least. Shall we begin with the first
- 2:38topic of the day? Let's go Anthropic just leaked
- 2:42information about a new AI model that they're working on that is
- 2:45so powerful that they have to hold it back due to security
- 2:49concern. Actually, this news got out
- 2:52because a university researcher from Cambridge University found
- 2:56more than 3000 unpublished Anthropic asset just sitting in
- 3:01a database. Imagine they are building the
- 3:04most powerful AI model in the world and their press releases
- 3:08are in a database that everybody can access.
- 3:10Let's look at the article together.
- 3:13This is Claude Mythos. That's the code.
- 3:17Name. That's the code name for the new
- 3:18project, so I'm going to read it a little just to.
- 3:21Give everyone an idea. We are now using Haiku, Sonnet
- 3:25and Opus, right? These are the three versions.
- 3:27Of is the most powerful 14.6. This is supposed to be the.
- 3:30Replacement. No, this is going to be even way
- 3:32more powerful. So when usually they release new
- 3:35models, they give you the options to play around with
- 3:38them. Yeah, so this is going to be a
- 3:39new edition. They mentioned we have finished
- 3:42training a new AI model, Claude Mythos.
- 3:46It's by far the most powerful AI model we've ever developed.
- 3:51Mythos is a new name for a new tier of model, larger and more
- 3:56intelligent than our Opus model, the one that you were talking
- 3:59about. And Opus models are by far the
- 4:02most intelligent I have ever tried.
- 4:04We've been using Opus for all of our coding tasks.
- 4:07It is so intelligent that we just tell it what we want to
- 4:11build, leave the studio, come back and it is done.
- 4:15Pratt, can you, can you just just testify to what I just
- 4:19said? Yeah, that's true.
- 4:21I didn't pay you, neither did Entropic.
- 4:24I'm just letting you know. They said they chose the name to
- 4:27evoke the deep connective tissue that links together knowledge
- 4:31and ideas. First of all, I have a thought
- 4:36that this might be an intentional PR move.
- 4:40This could be right. They would release some press
- 4:43releases in a database. People will find it.
- 4:45Everyone's like, oh, this is a secret that is leaked out and
- 4:47now everybody's talking about it.
- 4:50Because the press release, if you go down, if you look at it,
- 4:52it's very complete. And they talk about it's so
- 4:56powerful that they have to only give it to the cybersecurity
- 4:59experts first so they can test it.
- 5:02Then they have to come back to Anthropic and tell them that
- 5:04whether they can release it to the public or no, so they're
- 5:07going to slowly release it to people.
- 5:10I am a typical employee of Anthropic.
- 5:14I'm about to release these files.
- 5:16Wouldn't I give it to Claude and be like, can you verify if what
- 5:20I've written is complete and not include any information about
- 5:23the new models? Like do they not use Claude
- 5:27themselves to verify what they're about to upload to the
- 5:30public? That is why I do believe, just
- 5:33like you said, this could very well be APR moment.
- 5:37But so be it. So be it it.
- 5:39Is a smart idea that you can use for any company.
- 5:44If you're trying to market a new release for a product, whether
- 5:47you're a fashion company, a drink company, an apparel
- 5:51company, whatever it is, you can use a tactic like this to get
- 5:55people talking about your product.
- 5:57So there is no problem with that.
- 5:58I'm more interested in finding out if Methos is actually real.
- 6:02If it's, it's Israel. Fortune.com, which is one of the
- 6:05first agencies that release this news, got the confirmation from
- 6:09Anthropic that they are working on this model.
- 6:12This model exists and they are doing the safety measures first,
- 6:16so they're not going to release it.
- 6:17So the news is real. So it's so powerful that they
- 6:20had to hold it back and test it with a small group of people,
- 6:25make sure it is safe for everybody else to use it, and
- 6:28then release it. Why?
- 6:29I think what they were they were afraid of is people start using
- 6:32it immediately to hack banking system, to hack emails, to hack,
- 6:36you know, different networks. So they need to set the guard
- 6:40rails around their model exactly exactly.
- 6:42This is this is very interesting.
- 6:44I truly we're getting to that point.
- 6:46Yeah, I truly cannot wait to even test it for our own apps.
- 6:49Imagine if Office can make an app so well, if this is way
- 6:53better, way better. That is not a different version,
- 6:55it's an entirely different model.
- 6:57What it can do and imagine now integrate it to code, work cloud
- 7:01code, dispatch code review, all the different features that they
- 7:05have. But it's going to be very
- 7:07expensive. They mentioned that it cost way
- 7:09more than Opus, so I think you should.
- 7:11Corpus already cost the bomb, we're paying 300. 50 No, but we
- 7:16are we are at the highest you can get it you're.
- 7:18Telling me your but you know the compute for this will be more
- 7:21expensive than Opus unless they say it's more efficient because
- 7:27you can have newer models be increasingly more efficient too.
- 7:31If it is not more efficient, then you're going to pay more
- 7:34for compute. No, they what they mentioned is
- 7:37Mythis is also a large compute intensive model.
- 7:40It's very expensive for us to serve and will be very expensive
- 7:44for our customers to use. We are working to make the model
- 7:46much more efficient before any general release.
- 7:49So no, I can tell you that they're not going to release it
- 7:51to the meat tier. Brad, we need to go start
- 7:53washing dishes to be able to afford the next version of
- 7:56cloud. No, bro.
- 7:57And then what you need to do, I think you shouldn't do
- 8:00everything with the most expensive model.
- 8:02So if you have Opus now 4.6 if you're asking basic question, if
- 8:07you're asking them to create documents.
- 8:08What should I text my girlfriend?
- 8:10Yeah, exactly. Don't use the most.
- 8:11Expensive model to do? No.
- 8:13But what if the most expensive model mythos can give you the
- 8:16right response so that your globe won't break?
- 8:20You need to look at the outcome. What is the outcome that you
- 8:22want from that? Message I want my girl.
- 8:24What if she's using AI to fight with you?
- 8:27So if I'm using Mythos, I'll have a much better response.
- 8:31Join planet for just having an argument.
- 8:33Just argue that's why. Ram costs more right now
- 8:36arguing. Yourself.
- 8:37Actually, I'm very good at arguing.
- 8:38You can get my help. I'll charge you.
- 8:39I know first hand multiple people in our closed group of
- 8:45friends. First hand.
- 8:46First hand, I know multiple group of people in our closed
- 8:49group that use ChatGPT and other LLMS to respond to women and I
- 8:55know women that use LLMS to respond to men.
- 8:58Therefore this is already being done.
- 9:01I'm just saying you can use a smarter model for a better
- 9:03response. What if you can get her out on
- 9:05that date with Mythos? Bro this is life changing
- 9:10technology for you to build app and software to solve human and
- 9:14mankind problems. Not to go on a date, you can go
- 9:18on a date yourself. But that person who's going to
- 9:20go to the moon is going. To the moon.
- 9:22Anybody from Antioch be watching this?
- 9:24Ultimately, they're not going to give you access.
- 9:26Ultimately speaking, I want to apply.
- 9:27For the man who became the astronaut decided to study
- 9:31because he wanted to have a higher status in the hierarchy
- 9:34of men and through that higher status, he can then mate with
- 9:37higher status of women. And so if you understand, do you
- 9:41guys not understand where I'm going, bro?
- 9:42You're totally off, I think. Doesn't make any.
- 9:45Sense. No, it kind of does.
- 9:47No, I know where all of our tokens are being spent.
- 9:50OK jokes aside, Pharos is part of that group.
- 9:53He's using all of our tokens from Opus and from ChatGPT to
- 9:56just message girls. This is embarrassing.
- 9:58Listen, jokes aside, what do you think about Mythos?
- 10:05We talked about this briefly during the last podcast episode
- 10:09that Claude is already overtaking ChatGPT in terms of
- 10:12intelligence and user base. They've done such a great job
- 10:16that we feel like it is more human to speak to Claude now
- 10:21that they're saying they have a model substantially stronger
- 10:24than Opus. I cannot even comprehend what
- 10:26that is. Is this closer to the AGIASI
- 10:31conversation we're talking about?
- 10:33I mean, we're. Definitely it's a step closer,
- 10:35but I don't think we are gonna reach that fast today.
- 10:37My office was quite dumb to be very honest, but they can turn
- 10:40it down. Yeah, that's what I'm saying.
- 10:42Maybe they're using the same compute to train mythos so.
- 10:45The problem is also the problem is when you're saying the
- 10:48compute is gonna be expensive. Elon has talked about this quite
- 10:55detailedly in a detailed way that compute is going to be a
- 11:00problem on Earth, so he's going to space.
- 11:03Sam Altman has said, Nope, we have plans.
- 11:07We know exactly what we're doing.
- 11:09We're going to have enough data centers and enough power to
- 11:12supply all of the users of Chachi BT.
- 11:16Now that you're saying they have new models coming out with a lot
- 11:20more computationally intensive architecture, and that they are
- 11:28gathering more users, how are we going to supply the resources
- 11:32necessary so that Quad Mythos can work?
- 11:36No, by limiting access, they're only going to serve a certain
- 11:39group of customer. With these models, not everybody
- 11:41have access to. It no, but you can pay for it.
- 11:43Not everybody will pay for it. It's so expensive.
- 11:46Is $1000 how many people will pay for it?
- 11:49Not that many I tell you. Not many people are on.
- 11:52Actually I want to ask people how many of them are above pro
- 11:55accounts. So pro account is $20 Anything
- 11:57above that is like 100 hundred $5200.
- 11:59I don't think a lot of people unless you are coding or unless
- 12:02you are a big company, I don't think you are part of that
- 12:05ChatGPT I wrote. I read this that the other time
- 12:08most of the users are free. Users like our mom and dad are
- 12:12free. Users.
- 12:13I agree with that. So you don't have to serve
- 12:15everybody just. Like how Google initially
- 12:19everyone on Drive were free users, and then over time I
- 12:22acquire so much data that I have to get the pro plan.
- 12:26But a lot of people still use Google Drive for free.
- 12:28Yeah, I get it for I'm a paid price.
- 12:31Yeah, I'm a free user. There you go.
- 12:33Yep. What I wanna say?
- 12:35Limiting the customers. I do wanna ask this question as
- 12:38you did. Please let us know in the
- 12:39comments how many of you guys, you that are watching this
- 12:43specifically are paying for a subscription of an AI model,
- 12:48whether it's ChatGPT or Grok or Claude or Gemini.
- 12:52Let us know if you're paying for a model or any tier, if you want
- 12:56to be specific. Or are you using it for free?
- 12:58Because I think it's important to understand the percentage of
- 13:02the population. I think our audience would be
- 13:04mostly paid at least because the base, the base you.
- 13:06Run out of credits and you cannot use the smartest models.
- 13:11But then. But I do want to say one thing.
- 13:13This is the craziest conversation.
- 13:17Intelligence is going to come at a price.
- 13:21Previously, you could say it is also true.
- 13:24If you have a lot of money, you can send your children to the
- 13:27most elite universities and colleges.
- 13:30But you can still access through the Internet the courses online
- 13:36or books online. You can still train yourself to
- 13:38be highly intelligent with AI. Now.
- 13:42What I'm afraid of is a future where you have to really just
- 13:45pay for intelligence. The more money you have, the
- 13:47more intelligent you'll become. And if you do not have money,
- 13:50you'll just be retarded compared to the elites because they have
- 13:55access to the highest levels of intelligence.
- 13:57The level of intelligence gap will just be expanded to such a
- 14:01level that it is impossible to compete with the richest people.
- 14:06I think then that's why open source models also exist.
- 14:09And that's my counter argument. Did you counter yourself?
- 14:12I had both sides. What happened to We Don't need
- 14:15you anymore? Did you just counter yourself?
- 14:18What was that? Argument.
- 14:19I believe both sides could be possible.
- 14:22So which side you're on? You're no side, you're in the
- 14:26middle. I don't know which side is more
- 14:28possible, I just know both sides exist.
- 14:30What was your counter? Tell me.
- 14:32You just countered my. Counter, you said open source.
- 14:34Yeah, OK. Do you know why?
- 14:37What I mean, do you know why now we are getting Quen models which
- 14:40are free 0.8 billion parameters compared to 92 billion.
- 14:45And for the task that you're doing, day-to-day task for
- 14:48average user, they're performing the same.
- 14:50Again, if you are making a very detailed enterprise solution,
- 14:57sure you need Mythos. If not, then I think you're good
- 15:00to go with Quinn. You don't need the highest level
- 15:03of intelligence forever. Just like right now you don't
- 15:05need all the books in the world. You just need a few books that
- 15:08just teaches you the skill that you want to.
- 15:10True, but if you're competing with a few other companies for
- 15:15the same pitch, whoever has the best pitch wins, right?
- 15:19If you have the most incredibly intelligent Claude Claude
- 15:23mythos, it can device a much better plan and a storyboard and
- 15:28script and so on and so forth for you.
- 15:30That's one argument. But to add to yours, and I agree
- 15:34with you on that, we were talking about this with Pratt
- 15:36the other day. They are now distilled models.
- 15:40What distilled models are, For instance, you get a Quen 0.8
- 15:44billion parameter distilled Opus 4.6.
- 15:48So you take a 0.8 billion parameter model that is
- 15:50available for free online and then you train it with the
- 15:56responses of opus 4.6 so that it responds more like 4.6 opus.
- 16:03Much more formidable option. Specific domain or how do you
- 16:06train it like? Is it for like coding or is it
- 16:08for? Content exactly.
- 16:10You decide, you decide. You can fine tune it to be more
- 16:13like Opus 4.6 but you are running it with no cost for free
- 16:20on your computer or even on your phone. 0.8 billion parameter
- 16:24quen model can run locally on your device smartphones and so
- 16:28if it responds more like 4.6, that's how you can get it
- 16:32closer. This is what Pratt is showing
- 16:33you right now. The reasoning is close to a
- 16:38cloud 4.6. You can do this with a 27
- 16:40billion parameter model. You can do it with 9 billion
- 16:43parameter model. That's up to you.
- 16:45But the distillation is what makes them even smarter.
- 16:49And this is where I would add to your argument and say, yes,
- 16:52you're right. Even with open source models, we
- 16:54can actually keep getting them closer to the paid models.
- 16:57But the paid models so far have always been slightly ahead.
- 17:00Or perhaps this is where we don't have enough knowledge.
- 17:03Maybe not just slightly ahead, maybe a lot more.
- 17:06We just don't get them publicly available.
- 17:09Of course, there are a few versions that are ahead of
- 17:12whatever is available to us. But again, for really depends on
- 17:16your use case. If you're not using it for
- 17:18research, for whatever we do, project delivery, app
- 17:21development, content creation, even a complex solution,
- 17:26software solution, I think the current models are really,
- 17:29really good. And the differentiator between
- 17:32two competitors is not which AI model they use.
- 17:35I think is how they are orchestrated these different
- 17:38models together. I partially disagree with your
- 17:41statement. Good is subjective.
- 17:44When you say good, what do you define as good?
- 17:47And I've seen you, Pratt and myself cry on certain days where
- 17:51we have to go back and forth and shout at these models to get
- 17:54both of the job. Yeah, but what if he gets it
- 17:56right in the first try? It knows it's like going to a
- 17:59designer. Remember the opening I first
- 18:02Super Bowl ad they did with the most incredible motion design
- 18:07that awed everybody. Imagine you go to Claude right
- 18:10now, Claude Mitos as an example, and be like design me a motion
- 18:14design ad for my brand. Don't tell it anything else.
- 18:18Give it your website and it creates something like that, but
- 18:20it is original as in it doesn't look like anybody else's work.
- 18:24It feels original as it did on the Super.
- 18:27Bowl no. But your example is even worse
- 18:29because you make a subjective outcome out of AI.
- 18:32So something that is good to you, it might not be good to me.
- 18:36Let's talk about a software. The open AI ad everyone loved
- 18:39What? Are you talking about?
- 18:40I'm sure there are people who are criticizing that.
- 18:42Ad no, just like they do with Chris Nolan movies.
- 18:44But Chris Nolan movies are generally agreed to be great.
- 18:48You're saying no Chris Nolan has haters of course, but Chris
- 18:51Nolan is known for incredible films.
- 18:53By the majority. By the majority.
- 18:56By the majority. I'm arguing on that side.
- 18:59What I'm saying is dude, you can look at 2 websites and say
- 19:02what's good and what's not good. Yeah, yeah, yeah.
- 19:04I again, I would think that if everybody has access to all
- 19:08these models at the same time, the competition is not who has
- 19:12the best model, it's who can orchestrate these models better,
- 19:15who can use better models, right?
- 19:16Yeah, sure. You're talking about the moment
- 19:19that we have a genie, everybody has a genie.
- 19:21Give me the best of everything. Yeah, but who has the stronger
- 19:23genie, who's faster and more capable pays more?
- 19:27No, that's what I'm arguing. You're thinking that at the end
- 19:30of the day, if all of them can design websites, then you can
- 19:33keep talking to it and get the website that you want.
- 19:36I'm saying, what if you are not the one telling it?
- 19:38What if I don't know how to design websites?
- 19:40You know how to design websites so you can tell it what you
- 19:43want. What if I don't know?
- 19:45I'm a guy in a brick and mortar business.
- 19:48I know nothing of websites. This is telling me oh bro you
- 19:50need a freaking website and it should look like this and when
- 19:53it's done it actually works. Marvelously, the better model
- 19:56you have, it's like the better Genie I have done.
- 19:59That's what I'm arguing for. I'm saying I believe there's
- 20:02always going to be a price to pay for intelligence in the
- 20:05world of AI for sure. But I also agree with you, you
- 20:08can. If no, you remember you're in
- 20:10argument with yourself. Put me aside.
- 20:12You are just debating yourself here.
- 20:14I agree with both of them. Should we do the team?
- 20:17Farrah and team Farrah? No, it's just Team Farrah's one
- 20:20and Team Farrah's 2. Pharaoh's one.
- 20:21And I didn't even start this argument, but we do a nice
- 20:24thing. What do you guys think?
- 20:25Should we go and have a fight about it outside?
- 20:27Let us know in the comments. What do you guys think about
- 20:30this whole paying for intelligence?
- 20:32This is not a new concept, but it's definitely being
- 20:36exaggerated in this world of AI as people get access to tools
- 20:41that genuinely do make you more intelligent and almost instantly
- 20:45right the moment you subscribe to Clot, suddenly you now know
- 20:49how to code, You now know how to design websites.
- 20:52And this is only going to get stronger and better over time.
- 20:58What are you talking about Pratt?
- 21:03Do you have any opinions on this?
- 21:04No man, you can. You can keep fighting with
- 21:06yourself. All right, shall we move on to
- 21:10Let's go the next one? Anything else on Mithos or no,
- 21:13we're good or not? We're good.
- 21:14OK All right, next one. As you all know, my favorite
- 21:19topic in this podcast has to do with brains and neuroscience,
- 21:25and we luckily have something to show you almost every single
- 21:28week. Meta just released a digital
- 21:32human brain model that can accurately predict how our brain
- 21:38responds to everything that we see or hear.
- 21:42This digital twin of the human brain acts as a visual
- 21:46representation of how we think and feel.
- 21:50When you watch films or listen to podcasts, you show the
- 21:55digital human brain a video of this podcasts and different
- 21:59parts of it start to light up, showing you exactly what the
- 22:03brain is feeling in real time. Show it a video of a romantic
- 22:08couple and the brain responds to it with a feeling of love.
- 22:13Let it hear an intense argument about AI and it will response,
- 22:18potentially with anxiousness. So what is this model?
- 22:23It's a Research Foundation model called Tribe V2 Trimodal Brain
- 22:28Encoder. It is a DID.
- 22:30Thank you. It is a.
- 22:34Apologies, it is a digital. Twin trained to predict human
- 22:39neural activity. It was trained on over 500 hours
- 22:44of F MRI data from roughly 700 volunteers.
- 22:49Those volunteers were exposed to a wide variety of images,
- 22:53podcasts, videos and text. And for people who are unaware
- 22:59of fMRI, I'm sure you might have heard of MRI, the method by
- 23:04which we can scan the organs and tissues of the body. fMRI is
- 23:08more specifically tuned for the brain, and it looks at blood
- 23:12flow and oxygen activity to then see which parts of the brain are
- 23:16activated. So the higher the level of
- 23:18oxygen in a certain part of your brain, the more activated it's
- 23:21becoming. And using that and all of the
- 23:24training, they're trying to figure out how the human brain
- 23:28thinks and feels when it's exposed to certain sounds and
- 23:32visuals. And So what makes this model
- 23:36even more interesting is that it can accurately predict new brain
- 23:40responses for entirely new people that it never scanned for
- 23:46and for entirely new content that it was never trained for.
- 23:51And it's also open source and free to use for research.
- 23:55You can get this model right now, give it to Claude, and then
- 24:00design for yourself a user interface that you can use to
- 24:03test the brain on content. No, I think somebody already
- 24:07done that. Pratt, can you go to the
- 24:08comments? I've seen someone immediately
- 24:11created a Scroll down. Yes, look at this open the
- 24:15images, he says. So I built an app that uses this
- 24:19code. You can just paste the URL for a
- 24:22YouTube video or a TikTok video and it will tell you the
- 24:25emotions. So somebody already done that.
- 24:29You see, with all these AI tools, you can immediately.
- 24:31You have an idea, build it you. Build it immediately.
- 24:35Yeah, that's yes, these are the ones.
- 24:36What did he say? Can we read what he just said?
- 24:38Pasted any? So I built an app with it.
- 24:41Paste any YouTube, TikTok or IG link and get predicted attention
- 24:44curves over your entire video. Emotional intensity mapping.
- 24:48Cognitive load analysis. A live brain visualization that
- 24:52lights up as your video plays. Exact timestamps where your
- 24:56viewers lose focus. It's basically Vid IQ on
- 25:00steroids. Vid IQ is the add on you can
- 25:03have on YouTube and creators have been using it for ages.
- 25:07It tells you how your video is performing.
- 25:10But this is. No, let me tell you what do I?
- 25:13Think vid IQ. In neuroscience, the concept and
- 25:17the research looks very promising and very interesting.
- 25:22Just like every other episode, I find our brain a very complex
- 25:27structure. And really?
- 25:30Mine mine is even more complex. How can you get all of this data
- 25:36from 700 people with 500 hours of content?
- 25:40This is this means less than one hour video per person.
- 25:45It means they watch a podcast, a bad decision podcast.
- 25:48And oh OK, we figured out every single thing.
- 25:51First of all, probably all the 700 were from a specific region
- 25:55or location. All 700 from states.
- 25:57People from states react differently from people from
- 26:00Japan react differently from people from Malaysia to the
- 26:05exact same video. Something that is funny or evoke
- 26:08an emotion to you might not evoke the emotion into an
- 26:11Italian guy or a Japanese guy. So I find it very very hard for
- 26:15this to be even accurate in a global level.
- 26:18Just since you brought that up, that whole Trump video with the
- 26:21Prime Minister of Japan talking about Hiroshima, yeah, yeah.
- 26:24If you show a video of Hiroshima to a Japanese person and an
- 26:27American person, they will have slightly different reactions.
- 26:29Not slightly. Show it to someone in.
- 26:31Afghanistan, they will have a completely different reaction.
- 26:34That's why I'm telling you 700 people, 700 plus let's say 799,
- 26:39which is 805 hundred hours of content.
- 26:42If we talk about the different types of content, we talk about
- 26:45horror, sci-fi, action, thriller, funny, comedy,
- 26:48whatever. There's so much content for
- 26:50someone to consume. And let's put content aside the
- 26:53same content with different music.
- 26:56So if you play a fast music, a slow music, a sad music on the
- 27:00same. The same content slowed down in
- 27:03slow motion will evoke a different emotion.
- 27:06So I find it very, very, very, very, very difficult to get to
- 27:13the result that they claim they have achieved with this amount
- 27:16of data, unless this was an improvement from the previous
- 27:20model. And in the previous model they
- 27:21have this around 1,000,000 people.
- 27:22OK, sure. But I find it very skeptical.
- 27:25We should give it a try and see what are the results.
- 27:29We can actually give one of our videos and see and then compare
- 27:31it with our YouTube channel. Yes 100% I love that I agree
- 27:37with your concern. I do believe, considering how
- 27:40complex the structure of the human mind is not yours, you
- 27:45cannot use my joke against. Me, I hate it when someone does
- 27:48that. Can do it.
- 27:48Come up with your own original. If you're you have 86 billion
- 27:52neurons in your brain, can you not use one of those neurons to
- 27:54come up with a new joke? Use one.
- 28:01Just confirm, Farhad. Has less than one neuron in his
- 28:04brain, but. 85,000,000 five 900 How much really left?
- 28:11But the point I wanted to make is this.
- 28:13We have 86 billion neurons. We as in me and Pratt and.
- 28:18Together, right? So these are like 43 only.
- 28:22We have 86 billion neurons. We just compared this to the
- 28:25fruit flight the other day when we were talking about the fruit
- 28:28flies. They have about 120,000 neurons,
- 28:31less than a million. No, no, no less than 1,000,000,
- 28:34less than 1,000,000 neurons. And in that specific episode, we
- 28:39talked about the scientists and the researchers concerned about
- 28:43how complex it will be to scale their tests from a fruit fly all
- 28:47the way to a human brain. They have to go through a mouse
- 28:50brain 1st and then go to a human brain.
- 28:52And it's going to take them years.
- 28:54They are sure they're going to achieve it, but mapping out a
- 28:56human brain is going to be incredibly complex.
- 28:59Not just that, but the amount of content available on the
- 29:04Internet, I mean, just scrolling on TikTok and Instagram, you can
- 29:07consume hours of content within a single day.
- 29:11And podcasts, when they're exposing these people to
- 29:13podcasts, which podcasts? Are they funny podcasts?
- 29:16Are they sad podcasts? Crime mystery.
- 29:20You have to expose a much larger group of people to a wider
- 29:24variety of content over a longer period of time to be able to
- 29:29accurately map out the brain. I agree with that concern, but
- 29:33that doesn't mean this technology is not fascinating.
- 29:36The fact that this is open source and free to use for
- 29:40researchers, it is already fucking insane.
- 29:43Can you tell me someone that can go right now and get 400 people
- 29:47or 700 people and do an fMRI scan on them and then analyze
- 29:51that to create this, train an AI model and create this?
- 29:55How much money does it cost? How can you raise that money to
- 29:58do this? Meta did it and release it
- 30:00publicly. That is something I.
- 30:03That is already fishy. Meta did it and released it
- 30:05publicly. Listen.
- 30:07Sure. No.
- 30:08OK, the why would they release it publicly?
- 30:11I'll tell you why. The whole concept of open source
- 30:14is that when you release this publicly, other researchers from
- 30:18other parts of the world with fresh new revolutionary ideas
- 30:22will work on the models and develop new features and
- 30:26products. After which you can go and hunt
- 30:29them and hire them to be part of your team and they can then
- 30:33design the next generation and iteration of your product.
- 30:37That is one reason. There are other reasons I'm
- 30:39unsure of. I'm not part of the Meta team,
- 30:42but what I want to say is, generally speaking, the concept
- 30:45of free and open source has always helped innovation and new
- 30:49technologies appear on the surface on the planet.
- 30:54I've come up with two strategies and two theories of how I think
- 30:59this model can and may be used in the future.
- 31:03Both of which can be scary, yet exciting.
- 31:07Are we ready? I mean, one of them is obvious,
- 31:09right? You want me to say it or you
- 31:11will say it. I think we think about the same
- 31:13thing. Go ahead.
- 31:14Well, go ahead. I will.
- 31:15You can say it. I will confirm whether you can
- 31:17say it. No, no.
- 31:18Let me think about it. Yes, I know you're thinking
- 31:20about the same. Thing are you reading my screen
- 31:23yeah he's like let me think about it I said peek and my.
- 31:27I didn't scream. It's obvious, right?
- 31:29The first one. You're still peeking now that
- 31:32you said you are a horrible cheater.
- 31:35I cheated. You cheated like this.
- 31:36I cheated my entire elementary school.
- 31:39Oh my God. With precision, this level of
- 31:43precision, we didn't have laptops, so we had to use paper.
- 31:46Yeah, but you still peek the same way nonetheless.
- 31:48Let me just say it. You can tell me and confirm if
- 31:50you already know what you read off my screen One.
- 31:54Brands like Apple and Adidas can test hundreds of commercials and
- 32:00ads, every photo, every frame of the video, and see which one
- 32:06triggers the most emotional response while keeping the brain
- 32:12engaged. Which ones create the highest
- 32:15levels of desire from consumers so that the next time they have
- 32:20a Super Bowl ad, it has been tested and verified on digital
- 32:25human brain models to ensure that people will love the ad?
- 32:31Isn't that fucking insane? How does original content then
- 32:36work? If it's still making the
- 32:39original content, like we try dozens of versions and see which
- 32:43one works. Previously this was called a
- 32:46focus group. You'll get 20 people.
- 32:49I'll give you the example. You get 20 people for the new
- 32:52Domino's Pizza flavor. And then you show them the
- 32:55Domino's Pizza flavor and you're like, are you going to be
- 32:57excited to eat this pizza? And they're going to say yes or
- 33:00no. And based on that focus group,
- 33:02you then design your new flavor and taste.
- 33:04Sometimes you even feed them the flavor, but you can you.
- 33:06Better feed me the flavor here. Just show me the.
- 33:08Photo OK, fair enough, but you get my point.
- 33:12Now instead of a random focus group or a trial group which can
- 33:16cost a lot of money and also may not be 100% accurate here, you
- 33:23can just sit behind your computer and test it.
- 33:25And you can test it unlimited. You can sit and test it for
- 33:29hours on hours on hours, for days and days and weeks and
- 33:33weeks. And why you need to test it for
- 33:35weeks and weeks if you can just spawn the brains and show them
- 33:39your content. No, because you're generating
- 33:41new content and you're testing. The new content and new content
- 33:43to. See how it works, right?
- 33:45New audio, new images, new videos.
- 33:48And imagine you didn't hook this up into your AI generation.
- 33:52That's content Machine. Non-stop AI, AI generated video
- 33:56to AI generated brain and then AI model to decide whether this
- 34:00content is good or no. Then give it to humans bro.
- 34:05Listen, I'm not creating this. I'm telling you a theory of what
- 34:11could happen in the future. As always, I'm just telling you
- 34:15what could happen. Why do you hate on me?
- 34:18I don't want to live in your future.
- 34:20It's not my future, it's the future that you are drawing.
- 34:25Before it used to be a guessing game.
- 34:27Now it's I still scientific, I still think with precision.
- 34:33Again, let's actually try one of our Instagram.
- 34:35Let me give you the other theory that I've got.
- 34:38OK, this is a research done by Meta.
- 34:42Meta is the owner of Facebook and Instagram.
- 34:47Imagine Instagram creating the Super algorithm knowing exactly
- 34:53how our brain responds to every single post that is uploaded to
- 34:59the platform. What kind of emotional response
- 35:02does bad decisions, new video invoke?
- 35:05How long can it actually keep the brain engaged?
- 35:08Then create the ultimate feed for every user on the platform.
- 35:14Again, you might say no, but this is horrible.
- 35:17They're predicting how they're already doing this.
- 35:20They're already. Doing this.
- 35:21They already know which age group you belong to which.
- 35:25Ethnicity. You belong to, they know exactly
- 35:29what you like and what you dislike, and they know that
- 35:32because you tell them by staying on the screen or perhaps even
- 35:36putting your finger on the screen to hold on to a video and
- 35:39pause it for a little while. So that three seconds that you
- 35:42just pause the video for shows your level of intent and desire
- 35:46for that video. Furthermore, I watch another
- 35:48video today, the iPhone here, the face unlock, which was I
- 35:55believe an iPhone first technology, but every phone has
- 35:58it now, is taking 3 images per second or three images every few
- 36:05seconds. I'm unsure of.
- 36:06But a lot of images that are being captured the moment your
- 36:10phone's on the table and you don't see that because it's an
- 36:13infrared camera. And that's why it's so smart at
- 36:17unlocking your phone, because it doesn't unlock it all the time.
- 36:19It unlocks it when you're looking at it, right?
- 36:22It's not always unlocked. Your phone can be on the table,
- 36:25but it only gets unlocked when you're looking at it from the
- 36:28side, and they know that because they're capturing the pupils and
- 36:33the eyes of every person. Meta could potentially use that
- 36:38same technology to know where you are looking in the video.
- 36:42Are you looking at the eyes? Are you looking at the clothes?
- 36:45Are you looking at the top or the bottom?
- 36:47They have full detailed analysis of every user, Which helps them
- 36:52serve the user better. Again, this sounds very dodgy
- 36:57and very scary, but when you think about how it's primarily
- 37:02being used, it's not that evil. Secondarily, yes, we'll talk
- 37:05about secondary, but primarily used to serve you better
- 37:09content. The reason why we're so addicted
- 37:12to our social media platforms is because they show us what we
- 37:15like and to a certain level that's great.
- 37:18For instance, when I like to watch more Unreal Engine videos,
- 37:22the moment I hold on to an Unreal Engine video just
- 37:25slightly longer, the next day I get incredible tutorials on my
- 37:29Instagram and those tutorials then help me figure out what I
- 37:33want to learn next. I then create videos about those
- 37:35features on the YouTube platform, which generates more
- 37:38views for the YouTube channel which I like.
- 37:40This simple formula is being used for every human and you can
- 37:45now manually dictate what you see on your feed.
- 37:49That's a new feature they added, which makes it less evil, so you
- 37:51can tell them you do not want to see politics on Instagram and it
- 37:54actually removes political content.
- 37:56Have you tried that? I've done that and it works.
- 37:59I've done that and I added neuroscience and reading and
- 38:03communication, all these new tags, and it is actually showing
- 38:07me those pieces of content. So primarily used for good, but
- 38:12secondarily then selling your information to advertisers
- 38:16without your permission and so on and so forth.
- 38:18That's when it can get crazy. But I think this technology will
- 38:25eventually be used to see how we respond to content before the
- 38:30content is shown to us. I believe that can be the
- 38:33future. Although this is just a research
- 38:35model, it may not be fully accurate yet.
- 38:37This is probably, yeah, definitely at the very.
- 38:39But it could go to that level. Yeah, in the early.
- 38:41Stages, right? But like, imagine the
- 38:43possibilities you can. For example, if I want to look
- 38:46good, I can take a picture of myself and I can run tests on
- 38:50it. On what clothes should I wear?
- 38:52What hairstyle should I have? Or how does it evoke emotion on
- 38:56other people? Right.
- 38:57Yeah, negativity. I never thought about.
- 39:00That maybe I can take a picture of my house.
- 39:03And how does the colors impact your brain?
- 39:06Am I more creative in that space?
- 39:08Wow, that's actually a genius idea.
- 39:10Imagine you just open up a new coffee shop.
- 39:12You're thinking about the colors on the walls.
- 39:15Which colors and which theme invokes a level of excitement
- 39:19for kids who are going to walk into the coffee shop or not?
- 39:22The coffee shop, the candy store, coffee shop, whatever.
- 39:26How do you want kids in? Your no, I saw in your eyes.
- 39:31Why do you want I saw in your eyes, Mr. Epstein, I saw in your
- 39:34eyes the way you looked at me. I just take kids to a coffee
- 39:37shop. Why do you want kids will have.
- 39:40Candies. And that's why I said candy
- 39:42store. You just said coffee shop.
- 39:43Why you? I just changed it.
- 39:45Half sentence, but what I wanted to say is because I remember
- 39:48Andy. Remember Andy?
- 39:50Yeah, he's not a kid. He doesn't want kids.
- 39:52And it's coffee shop. No, but he did design coffee
- 39:55cups in a specific way with colors and themes.
- 39:57Yes. Yes.
- 39:58What Pratt is alluding to here is that you can.
- 40:02Don't bring Pratt in your team, Pratt was saying.
- 40:05I will take a photo of my clothes.
- 40:07I will see if I look good or no. You want to have take a photo of
- 40:10your. Clothes and show it to who?
- 40:11Kids. I just brought him into the
- 40:15argument. What I'm saying is very simple.
- 40:17Yeah, you understand? Please don't make me go to
- 40:19prison. I'm just.
- 40:21I'm having a good time on the show.
- 40:23What he said is actually genius for design.
- 40:26What you said was absolutely stupid.
- 40:28Yeah, I should probably retract my statement.
- 40:31No, but it's true. But again, we are giving away so
- 40:35much control to all these algorithms.
- 40:38Imagine if you want to actually open the coffee shop.
- 40:41As the owner of coffee shop, don't you want to have a say and
- 40:44dictate how you wanted to? Look like you are giving OK,
- 40:47you're you're acting too pessimistic.
- 40:49No, I'm acting on with you. I'm.
- 40:51Turning on my recruitment, My recruitment let me.
- 40:54Let me explain. What pleases you?
- 40:55Do what pleases you. You are deciding between orange
- 40:59and purple, and then you give the brain both colors and the
- 41:04brain tells you hey if you go with orange it can invoke a more
- 41:08exciting don't. You have a brain yourself, why
- 41:10don't you use your own brain? Why the fuck do you use ChatGPT?
- 41:13And no what I take this one I see him talking to Claude like
- 41:19should I say yes or should I say no?
- 41:21I say that I never say should I say fucking chat right now is
- 41:25shared whatever you they will see yours.
- 41:28I never say should I say yes or should I say this?
- 41:30If you're going to act all cool, make sure your brother's not on
- 41:33the show. No, Tell everybody the truth.
- 41:36More that shouldn't. Be there if you're going to
- 41:38speak like this. If you're going to speak like
- 41:40this, you better be alone, man. If you want to open your own
- 41:44coffee shop or if you want to have your own place, don't you
- 41:48want to design it based on your own?
- 41:51Taste you are, but you still ask advice from your friends.
- 41:54You still ask for advice from ChatGPT and Claude.
- 41:58That's what people are doing on a daily basis.
- 42:00You're doing the same, I'm doing the same, he's doing the same.
- 42:02You guys are probably doing the same.
- 42:04You still verify your analysis with a smart, intelligent model.
- 42:11You're doing that with your app. You're like, hey, is my.
- 42:13Website. I don't know how to make that.
- 42:14That's entirely no, no, no, no, no.
- 42:16I don't know how to make No, I don't know how to code.
- 42:19When I designed a website for our program, I talked to ChatGPT
- 42:23and I said is this flow good? Do you think this flow is easy
- 42:26to read, easy to understand? Bro, you've done that?
- 42:30No? No, I don't do that.
- 42:31No, I don't do the things that I can do.
- 42:34Give example. Ladies and gentlemen, this is
- 42:38the at this moment, I decide to remain silent and call my lawyer
- 42:42to come and help me out in this argument.
- 42:44But the point I wanted to make is you're not You're not.
- 42:47Please do. Please do.
- 42:48Team Pharaoh. Team Pharaohs right now.
- 42:50Let me know, and if you're trying to be on different teams
- 42:54for different arguments, just give your reasoning.
- 42:56One team Pharaoh's topic. Two team pharaoh.
- 42:58This topic is I, I think what we are discussing, I'm saying for
- 43:02the things that you can make decision.
- 43:04For example, my wall should be blue or red.
- 43:07I don't need to ask another intelligent model whether people
- 43:11like better blue or red, better blue, what is the better people
- 43:16like it better if it's blue or red, right?
- 43:19I'm saying that you don't need to.
- 43:20Ask. You're telling me that no human
- 43:22in the history of mankind has asked for advice from other?
- 43:26Humans, no, but again it depends on if you have shit ton of
- 43:30money. Yes, do whatever you want.
- 43:31But if I'm a broke guy, this is my strategy.
- 43:34Like I'm gonna take pictures, what color should I take?
- 43:37OK, Team Pratt. I'm a team.
- 43:40Pratt, wait, no. But first of all, I kind of
- 43:43agree with him, but it has nothing to do with being broke.
- 43:45You broke. Broke.
- 43:46People, you are an isolated team.
- 43:49Broke and rich people ask for advice the same.
- 43:52You're designing your new coffee shop, you're buying new clothes.
- 43:56You're like, hey bro, what do you think about these clothes?
- 43:58On where they look good all for it, but I think when I have are
- 44:01implying that for every single thing you have to ask the
- 44:04intelligence you have to see that Instagram video that in 20.
- 44:08How should I? How should I respond?
- 44:09They're in front. Of they're in front of each
- 44:12other, yes, the two. People, that was a big debate.
- 44:16I will just conclude my statement by saying I'm not
- 44:20suggesting that one should use this AI digital twin of the
- 44:24human brain for every single question they have.
- 44:29However, it is incredibly helpful for important and
- 44:37critical questions in your life to refer to an AI human brain to
- 44:42have accurate responses of how the general public may reflect
- 44:48your opinions, may respond to your opinions.
- 44:51So as Pratt discussed with us earlier, if you're opening up a
- 44:55new coffee shop and need to design the walls and you're
- 44:58unsure between the colors because you just don't know
- 45:01which colors would be the most exciting, you can use this maybe
- 45:05for every fucking clothes that you wear every day.
- 45:08Don't do it right? But perhaps, but maybe, but
- 45:15perhaps for some of our friends on Tinder and Bumble, right?
- 45:20You're setting up your profile for good for 2000 women to see
- 45:23or for women for 2000 men to see.
- 45:26You want your photos to respond positively.
- 45:30You can go ahead and use the AI human brain.
- 45:33You know what I'm saying? So.
- 45:34Bro one behind this research, never thought about all this
- 45:38cool. That's why they open sourced.
- 45:40It for you to go and use it precisely.
- 45:43So go and do it. I, I will, I get, let us, let us
- 45:47tell you this. We're going to use cloud, we're
- 45:49going to try to make this a usable application on our
- 45:52devices and maybe we keep and then tell you guys what we think
- 45:55about it. And if we do and if it works
- 45:57well, we'll send you guys the link and you guys can test it.
- 46:00This is why it's amazing. This is why I love it, and
- 46:04that's why. We move to the next topic.
- 46:06Yes, we had another topic I just realized.
- 46:08OK, yes, we promised you guys a live demo.
- 46:13And this is for all the people who love content creation with
- 46:20generative AI or have been looking forward to learning
- 46:24about content creation with generative AI.
- 46:26The reason why this topic is important is because as 3D
- 46:30artists, we have been creating content using tools such as
- 46:35Premiere Pro, After Effects, Da Vinci, Resolve for 2D content
- 46:40and Blender, Houdini, Unreal Engine, Cinema 4D for 3D
- 46:45content. Primarily.
- 46:46These 3D softwares and 2D softwares help us create
- 46:49content, whether it's for social media, for big screens, for
- 46:53movies, films, commercials, ads, games.
- 46:57They allow us to create content. And the content I'm referring to
- 47:00is not the I pick up my phone and I record myself talking
- 47:03about something. It's designed content, an ad for
- 47:07a shoe, a commercial for open AI shot for a film, a commercial
- 47:12for a hotel. All of these different examples
- 47:15need content, so I'm referring to that kind of content
- 47:18specifically. We as 3D artists have been
- 47:22playing around and experimenting with AI generated content since
- 47:27its launch, since the early days of Mid Journey, the initial
- 47:31launch of Dolly, which is open AI image generation model.
- 47:34We played around with free pick runway, Higgs field, cling
- 47:42models for video generation, Nano banana for image
- 47:45generation. We've tested every single tool
- 47:48out there and we always try to tell you guys on the show
- 47:51whether a tool is genuinely practical and useful or whether
- 47:55it's just pure hype. Just earlier today, we saw a
- 48:00post by Freepic, and Freepic, for anybody who's unaware, is a
- 48:06hosting platform of generative AI models.
- 48:10They have all the image generation models and video
- 48:13generation models in one place. So you pay a subscription fee to
- 48:18be able to use all of the models in one place.
- 48:21And the reason why that's useful is because usually if you want
- 48:24to use cling video generation model, which is a Chinese model,
- 48:27which is amazing, you have to go to cling and pay them.
- 48:30But when you pay them, you cannot then take those videos
- 48:33and then edit them further with other models or do anything else
- 48:37on Cling's website. The reason why Higgs Field and
- 48:39Freepic are increasingly more popular platforms is because
- 48:43these platforms have all the models in one place.
- 48:47You pay them a subscription, you get to use nada, bananas,
- 48:50Google, you get to use cling video and everything else.
- 48:55Freepic has one extra advantage over Higgs field, which is their
- 48:59spaces. They have a node system and for
- 49:013D artists, we know nodes give us so much more control and
- 49:05flexibility over the content that we create.
- 49:08But very recently, this post that they just uploaded by
- 49:12Freepic has one thing that has amazed us.
- 49:16AI content has always lacked control, more specifically 3D
- 49:21control. Say you have a product, this
- 49:26coconut water, and you need to take a photo of a very precise
- 49:32angle for an ad that you're giving to your client.
- 49:35Prompting that has been incredibly difficult.
- 49:39How can you tell it to be exactly at a specific angle?
- 49:42And then if you didn't like that, how can you move that ever
- 49:45so slightly and then move the bottle also by 23°?
- 49:50You see, in 2D and 3D software, as we've always had control, but
- 49:54with AI, you always lacked control.
- 49:56You always had speed, but you lacked control.
- 50:00This new update is promising to solve that.
- 50:04Do you get 3D control with AI or do you not?
- 50:08We're about to find out with this live demo.
- 50:10What is it called? 3D scenes.
- 50:13I'm calling it 3D scenes. But they call it 3D scenes.
- 50:15They're calling it 3D scenes. Damn it, I was about to coin
- 50:18that term. All right.
- 50:22What you're seeing here, what you just witnessed was a
- 50:25Gaussian splat. When we've talked about Gaussian
- 50:28splats on the Internet a long time ago.
- 50:30Gaussian splats are an incredibly efficient way to
- 50:34display and view 3D scenes. They're so lightweight and so
- 50:38photorealistic that you can view them on the Internet, on a web
- 50:41browser like we are doing right now, and go around an
- 50:45environment that looks pretty damn real.
- 50:47And so the Gaussian splat was all of those little spots that
- 50:51you just saw drop down to build this environment, which is
- 50:54inside a house. What you can do then in this 3D
- 50:58space, as Platt is moving for you, you can bring in a 3D
- 51:02object of your product. So you can have your own 3D
- 51:05objects or you can use whatever is there in three pic or you can
- 51:09generate it right. That's where it gets really cool
- 51:11because a 3D object you will have as a 3D artist or perhaps
- 51:16you're working with BMW or Adidas and they give you the 3D
- 51:20model in advance. But even if you don't have the
- 51:223D model, as Farad said, you can then use an image and use a 3D
- 51:28model generation such as trellis or triple, which are both
- 51:33available on Freepik, and turn your image of the product, say I
- 51:36have an image of this to a 3D model and then continue working
- 51:41in the space. So all of the possibilities are
- 51:43available for you here. Pratt has the environment and
- 51:48the sneakers. Pratt, we can get rid of the
- 51:50back and just place the sneakers using the sliders on the floor
- 51:57just like we would in real life. What did you want to say you
- 52:01want? To say.
- 52:02And then if you can make these sneakers a little smaller,
- 52:04because that's not realistic yet, we have to adjust it so
- 52:09that it matches the real world scale.
- 52:12Yes, smaller, smaller. Yes, that's that's that's too
- 52:15small. Baby stinkers.
- 52:16Yeah, OK, that's good. And this is where it gets really
- 52:20exciting. I get to have full control over
- 52:23where I am in the scene, choose the aspect ratio and the camera.
- 52:27Can we go to 50 mil and then change the aperture to 2.8 or
- 52:31one? This means we're going to have a
- 52:33shallow depth of field, so we're going to get blurring.
- 52:35All of this had to be prompted, but now we also get a live
- 52:38preview in the bottom right. If you extend that, it's showing
- 52:42us what this image would look like if it were properly
- 52:46generated with AI or rendered in a 3D scene in Blender.
- 52:50So it adds the lighting, it adds the texture and the materials
- 52:55that you want. But the best part is the actual
- 52:58precise movement of the camera. This is exactly what you cannot
- 53:01achieve with prompting a text message.
- 53:04This is exactly for anybody who doesn't work work with 3D
- 53:07software. This gives you the exact same
- 53:09powers and controls, yes. And I love that, as you said.
- 53:13So what you can do additionally, which makes it even cooler, is
- 53:16you have that prompt bar. You can then go ahead and add or
- 53:19remove details from your image. What did you just write, Brad?
- 53:23Put fire on the shoes. Did you see that in the live
- 53:25preview? Can we capture it?
- 53:28Want to capture it? Then it goes ahead and creates a
- 53:31full resolution image for you. Now is this a production ready
- 53:36image? Not yet.
- 53:38Perhaps that's why you can go ahead and use the image, take it
- 53:41to the node system spaces and then modify it even further.
- 53:45Perhaps you can turn it into a video that you can use for your
- 53:48ad and commercial. That is what Pratt is doing
- 53:52right now. What gives it so much power is
- 53:55the entire ecosystem like you mentioned, right?
- 53:57The fact that you can generate the image, turn it into video.
- 54:00But for image generation, maybe you want to use Pro, but for the
- 54:03video generation, maybe you want to use Cling or one way or one
- 54:06of our other models that is available out there.
- 54:10What makes this platform so powerful is now they're bringing
- 54:15the control from the 3D softwares, the AI models that
- 54:18are really superior to the other ones.
- 54:20And as a creator, you can go in, use different models, use
- 54:24different methods to create the output that you desire.
- 54:27I absolutely agree with you and the thing I want to see honestly
- 54:31for 3D artists is this level of control of rendering inside a 3D
- 54:37software. So they took what we had in 3D
- 54:39softwares which is the 3D camera control and Gaussian splats and
- 54:43brought it into a website for AI generated content.
- 54:46Unreal Engine, Blender a Unity need to bring the prompting and
- 54:50the live render to our 3D softwares you.
- 54:55Already have everything right. We have the control, we have the
- 54:57cameras, we have the lights, we have 10,000 more features inside
- 55:01the 3D engine. The part that is missing is set
- 55:04this shoe on fire or set this, I don't know make this shoe muddy.
- 55:08Instead of me going making the material and textures and adding
- 55:11all those elements, especially for creating commercials or
- 55:15short films, that would be more than enough.
- 55:18Let's just talk about pre production alone.
- 55:20Yes, just showing this to your client so that you can get your
- 55:24budget to get paid so that you can start on the project is
- 55:27already 1. Yes, this is fucking insane.
- 55:30By the way guys, this part is not sponsored.
- 55:33No, it's not. We just saw it and we are
- 55:35talking about it. Can we try it with the cool?
- 55:42It's not work. OK, so now you know.
- 55:46Actually, the part that is that if you don't step on it looks
- 55:49quite. Good.
- 55:49No, it does if you prompt it properly.
- 55:51Pratt just prompted it with one line.
- 55:53You need to prompt it with like an entire paragraph.
- 55:56So you need to do negative prompt.
- 55:57What did you do? Cling.
- 55:58Make the yeah. Cling.
- 56:00What is, can we? This is the first step.
- 56:03Pratt, can you while we talk about it, please do the cooler
- 56:07test we just did before. So put the cooler in the.
- 56:10Same room, just show the cooler. Yeah, yeah.
- 56:13Pull it up if the cooler is there.
- 56:16What were you saying? No, what I was saying is that
- 56:20they brought every single thing from the 3D software here.
- 56:23They listened to the consumer because the biggest problem with
- 56:27video generation with AI was the control.
- 56:30Yes, and platforms like Freepik, they already did that, they
- 56:33brought it here. I just don't understand why the
- 56:35other side doesn't do it. Yeah.
- 56:37You know, I mean, we're seeing some movement with.
- 56:40Unity and. Cinema 4D did announce that
- 56:44they're bringing in 3D model generation, I believe.
- 56:47Don't do it here. This is such an ugly place.
- 56:49Yeah, because I feel like that Gaussian splat was AI generated.
- 56:53Just do it in the same. Just do it in the same
- 56:55environment. Just put the cooler there.
- 56:58You can do it here. Sure.
- 56:59Do it here. Doesn't matter, we'll bring in
- 57:02the cooler house. We did this test earlier on and
- 57:05it worked really well. So that is a cooler that you
- 57:08would take with yourself when you're going hiking.
- 57:10You put some drinks in there and imagine you're doing an ad and
- 57:15you want to visualize this for a client and you need it to be
- 57:18precise. This is where you would come
- 57:20into Freepic 3D scene and quickly do a render of what they
- 57:24could have. Same camera Pratt, Let's go for
- 57:2750. This angle is great, actually.
- 57:31Tango is perfect. We can just say add
- 57:34condensation. Oh, my God.
- 57:37OK, so it was already there, the prompt.
- 57:39OK, can you remove that ice? What is that?
- 57:42There's an ice on the left side. So we just said add condensation
- 57:45on the box and it rendered it as if it's There you go.
- 57:49Yeah. Do you see that?
- 57:50Yeah. Now can you turn it into a
- 57:51video? Capture it?
- 57:55So imagine how fast we just created an idea to show to a
- 57:59client who would have wanted an ad for their cooling box.
- 58:03This is fucking amazing. Actually, this can put you 100
- 58:06times ahead than your competitors, whether they are 3D
- 58:11artists or whether they are AI artists.
- 58:13Whoever is not using tools like this is going to fall behind
- 58:17this. I would 100% use for pre
- 58:20production. I'm still unsure if it's
- 58:22production ready, maybe it's not.
- 58:24Yet definitely Shopify video for.
- 58:27Shopify and all the yes. That's production.
- 58:29Not every production has to be Hollywood level production for
- 58:32Shopify. The amount of videos.
- 58:35I still see a lot of garbage commercials so this could
- 58:38definitely replace it. While Pratt loads that up, I do
- 58:42want to have one thing mentioned every single video.
- 58:46You guys are asking us how I can learn about these tools because
- 58:51we try to show it as much as possible here.
- 58:53You're doing live demo. Yeah, we try to show it through
- 58:56a live demo, but these are not real tutorials because we're
- 58:59only generating one image and not a full story.
- 59:02We're not showing you the entire pipeline.
- 59:04How we would go ahead and use these images with the note
- 59:06system to actually create a full one minute long ad.
- 59:10All of that we're actually taking into our new education
- 59:13platform specifically designed for new tools like these.
- 59:18Those will require literal chapters dedicated to teaching
- 59:23you guys this, and it comes at the back of the demand that you
- 59:26guys have sent us either on Instagram, on Spotify, YouTube
- 59:30X, and all the other platforms. If you want to be notified for
- 59:35when we launch this program, make sure you guys go ahead and
- 59:38sign up at AI dot batdecisions dot studio.
- 59:42AI dot batdecisions dot studio. There's going to be a name field
- 59:47and an e-mail field. Just put in your name and your
- 59:50e-mail and then we'll send you a confirmation e-mail after which
- 59:53you are part of this list of early birds who will be notified
- 59:57when we launch and who will also be coming in at a huge savings
- 1:00:02compared to anybody else who comes later.
- 1:00:05That is going to be perhaps the most exciting project after this
- 1:00:09show that we've done in 2026. We've already started developing
- 1:00:13it and it is going to be one of our biggest projects ever.
- 1:00:18It will be something you guys love, I guarantee it.
- 1:00:20So make sure you guys go ahead and sign up.
- 1:00:21We already have over 100 or 200 people who.
- 1:00:24Signed up from the previous episode.
- 1:00:25Yes, and those people are part of the early birds.
- 1:00:28Pratt, do we have? The video is ready for the cool
- 1:00:31box. Yes, this is what you get with
- 1:00:38AI content sometimes. No.
- 1:00:39What was your prompt? Is it a one line?
- 1:00:41Yeah, it's a one line. Prompts never work, by the way.
- 1:00:44I just want to clarify that. So you have to enhance your
- 1:00:46prompt. All of this, by the way, is part
- 1:00:49of the program we just talked about.
- 1:00:52These are the things that we're going to be teaching in there,
- 1:00:54how to actually get videos that you can use in production.
- 1:00:57Because what Pratt just did is how most people use AI.
- 1:01:01They go in there, they describe one line of what they want, and
- 1:01:06the AI has to guess everything else.
- 1:01:08For instance, the AI doesn't actually know yet how to open
- 1:01:12the box. So here we used a larger prompt,
- 1:01:15as you can see on the side, and it did a much better job at
- 1:01:18opening the box because the AI sometimes does not know how to
- 1:01:22open the box. It will just open it from the
- 1:01:24front, which is not the right way mechanically speaking.
- 1:01:27But you know what's really interesting, as we're talking
- 1:01:30about stronger models, as Farah did with Claude Mythos, the
- 1:01:34reason why these models aren't able to recognize your prompt in
- 1:01:38a smart way. Pratt hold on to this image for
- 1:01:40a moment is because they are running not on Clod Mythos.
- 1:01:45When you're talking to Cling, it's using Cling's understanding
- 1:01:48of your video and Cling is great at video generation but not
- 1:01:52understanding a prompt as much as Clod Mythos is.
- 1:01:56But you have to understand Google owns Nano Banana, which
- 1:02:00is why Nano Banana runs on Gemini in the back end, being
- 1:02:05one of the smartest LLMS in the world, thereby producing the
- 1:02:09smartest, most incredibly accurate understanding of your
- 1:02:16image. That's why everyone uses Nana
- 1:02:18Banana. I would assume one to two years
- 1:02:21from now, Cling or whatever model that is generating videos
- 1:02:25is connected to the smartest models and therefore you can
- 1:02:28actually give it one line eventually and it will guess
- 1:02:31everything it has to guess correctly.
- 1:02:34But nonetheless, in 2026. March, I mean this is the first
- 1:02:38step and I think it's looking good, gives you enough control
- 1:02:40especially for pre production and even commercials.
- 1:02:43What is this image product that you generated?
- 1:02:45I just gave it a prompt, change the background to a jungle and
- 1:02:50the shoes are on the pre. And the angle of the shot is the
- 1:02:54same, right? There you go.
- 1:02:55This is why 3D control is important.
- 1:02:58Never change the texture. Did you say change the texture?
- 1:03:00Change it, yeah. I think this is promising.
- 1:03:04We have to see. Probably Higgs field is going to
- 1:03:07bring this new feature. Funny, Higgs field doesn't have
- 1:03:09nodes yet. Maybe they will add it on.
- 1:03:13But Higgs field is known for its simplicity.
- 1:03:15Yeah. So this would, yeah, this would
- 1:03:17make it a bit complicated. This will make it more
- 1:03:18complicated, but I know they market it having a 3D control
- 1:03:25but I never used it so I don't know if it is like this or not.
- 1:03:29But you guys can give it a try. I think it's available for
- 1:03:32everybody. Are we on a specific?
- 1:03:34No, we're paying. We are paying.
- 1:03:36We are finding it should be available for everybody.
- 1:03:38Whoever is paying, go ahead and try it out.
- 1:03:40As we always mentioned, it is important to experiment.
- 1:03:44It's important to try different tools to be prepared for what is
- 1:03:48coming in the future. So make sure you guys give this
- 1:03:51a try. I do want to say thank you to
- 1:03:54every single single one of you watching and listening,
- 1:03:58especially if you've stayed up until this moment.
- 1:04:00It's been about an hour and 8 minutes that you're watching and
- 1:04:03listening to us. We know in this hour and 8
- 1:04:05minutes you could have done anything else.
- 1:04:07You could have been playing games or just talking to random
- 1:04:10people or wasting time. I don't know what you would have
- 1:04:15done instead, but the fact that you decided to watch this show
- 1:04:18and be here with us to learn more together means truly the
- 1:04:22world to us, every single one of you guys.
- 1:04:24That is why I want to say thank you for being here every single
- 1:04:28episode. And for the ones, the rare ones
- 1:04:32who comment and tell us what they think about the episode and
- 1:04:36what they think about the topics, even more power to you.
- 1:04:40I appreciate every single one of you guys.
- 1:04:42You guys are the best really. And I'd like to end it with
- 1:04:45something Andrew Huberman always says.
- 1:04:47Thank you for your interest in science and take time.
- 1:04:51See you guys on Monday. Ciao.
- 1:04:53Ciao.