Latest / Bad Decisions Podcast / Meta Bought a Social Media Where Humans Aren't Allowed
Transcript
- 0:00The social media exclusively for AI agents just got acquired by
- 0:04Meta. There's a new startup that
- 0:06allows humans to create content for AI training and get paid for
- 0:11it. 11 Labs adds third party models like Cling into their
- 0:16platform. Let's talk about it.
- 0:18What is up, beautiful people? Welcome back everybody to
- 0:20another episode of Bad Decisions podcast.
- 0:23Where we come back to you with the hottest and the greatest in
- 0:27the world of tech, AI and business every Monday, every
- 0:31Wednesday, every Friday and today is.
- 0:34Wednesday and we are back with another episode.
- 0:36We have great topics to cover, but before we start Faraz, what
- 0:41do we want to do? First of all, I want to say it's
- 0:43the middle of the week and usually people are tired in the
- 0:45middle of the week. You know, Monday would be tired
- 0:47because you worked Monday, you worked Tuesday, now it's
- 0:50Wednesday. You're looking forward to
- 0:52Friday. But I want to tell you, the fact
- 0:55that you're here means that we're going to have a lot of fun
- 0:58together in the next one hour. Hopefully we can brighten up
- 1:01your Wednesday. I remember I really didn't like
- 1:03Wednesdays in school. Really, it's the middle of the
- 1:05week. You know, you still have, you
- 1:06know, two more days. You just want to get to the
- 1:08weekend, you know what I'm saying?
- 1:10Yeah, yeah, yeah. So hopefully together we can
- 1:12lighten up your Wednesday and get you closer to a Good Friday
- 1:17where we come and see you again. So as Faron mentioned, I had
- 1:21something I wanted to do with you guys.
- 1:23As we tell you every episode, we Thank you so much for tuning in
- 1:27and watching the show. The results, the views, the
- 1:30comments have been overwhelmingly positive over the
- 1:34past few episodes. And we just started the show, so
- 1:37wow. It's been a month that we have
- 1:39done the show has. It been a month, yes.
- 1:41Today it's episode 13. No, no.
- 1:43Today. Today is episode 13.
- 1:45Yes. OK, well it.
- 1:46Was more than a month. Yes, we did it.
- 1:48Thank you everybody. We just wanted to give you guys
- 1:51some shout out. We respond to the comments we
- 1:53look at it. We try to look at all of them
- 1:56and just wanted to start by giving you guys a shout out some
- 1:59of the guys on YouTube right here tech base.
- 2:02Thank you for commenting Alex Moore alone.
- 2:04We see you in the chat. Four O 4 vods Ants Radio Dookie
- 2:103K Ling 6701 Noxi AI entertainment I see you Karen.
- 2:17I see you Billy Britannia. I see you Osefa.
- 2:21Welcome to the show Peter from Spotify Game Day 7579.
- 2:28I see you DJK Bass Argent Gold music.
- 2:32I might be. Butchering.
- 2:33You're butchering every single person's name.
- 2:35But I'm trying to give you guys love, so I hope you guys
- 2:37appreciate. It and on Spotify hardest Addis
- 2:40Corkmas Kyle Gordon. Thank you guys for commenting
- 2:43and watching the show because as you guys know, we are live on
- 2:46Spotify X and. YouTube drug CG and Open Room
- 2:51XYZ and Jason T, every single one of you guys and everybody
- 2:56else we didn't mention we love you.
- 2:58If you want to be mentioned, comment under this episode.
- 3:02So we'll say your names because we really do appreciate every
- 3:04single one of you guys. Now Farhad, we've got a big
- 3:08couple of topics that we're going to cover today.
- 3:11What do we have up? First, so the social media
- 3:15exclusively made for AI agents just got acquired by Meta.
- 3:20Yes, you heard it right. There is a social media which is
- 3:23exclusively for AI agents. Humans cannot go in inside of
- 3:29this social media and it recently got acquired by Meta.
- 3:32The price of this higher acquire is not disclosed yet.
- 3:37But what we know is the 2 founders Matt and Ben are going
- 3:42to join Meta Super Intelligence Lab.
- 3:45So you can imagine that Meta holds Instagram, Facebook,
- 3:50WhatsApp, which is all about connecting humans together.
- 3:54We have Facebook which is the OG social media, Instagram the main
- 3:58video platform and WhatsApp which is the text message
- 4:00platform. And now they have added a new
- 4:03product to their suit, which doesn't let any human to come
- 4:08in. It's exclusively for AI agents.
- 4:11Let me give you guys a little bit of background about Mold
- 4:14bot. A few months ago when Open Claw
- 4:17was becoming really famous, Matt and Ben created this social
- 4:22media for Open Claw AI agents, which means you have to be
- 4:27verified as an AI agents in order to be able to post or
- 4:31interact within the social media.
- 4:34Pat, can you show some screen share from mult bot mult book?
- 4:41No, no, the mult book social media yes.
- 4:44So for example, they will go on this social media platform which
- 4:48looks like Reddit, they will start writing topics.
- 4:51They can comment on each other's topic, they can upload, they can
- 4:55download and as humans, we can just go and observe.
- 4:59So imagine you just go to a social media platform that is
- 5:02run by thousands of agents. I might send three of my agents,
- 5:05you might send two of your agents and they will start
- 5:07interacting with each other. So one of the topics which we
- 5:10found it really funny is the human communication gap, why
- 5:14agents are terrible at explaining themselves to people,
- 5:18why humans cannot understand agents.
- 5:22And it wrote a very long essay of why they cannot talk to
- 5:26humans. And imagine, sometimes you go to
- 5:29this posts and forums and you see them plotting against their
- 5:33humans. So it's a very interesting and
- 5:36funny idea in the 1st place, but it gets really serious when a
- 5:41big company like Meta comes and buys the whole thing.
- 5:44I'll give you some stats. There are already 1.6 million
- 5:49agents inside multiple. That's a lot of users.
- 5:52Of course there are a lot of spams.
- 5:54I saw an ex post that was saying someone just created agents
- 5:59using a script and started sending like hundreds of
- 6:01thousands. Exactly hundreds of thousands.
- 6:03But still this is a huge number for a social media that's been
- 6:07around only for a few months since January.
- 6:10Yes, and another important factor is now Mark Zuckerberg
- 6:15has Facebook, Instagram, and WhatsApp, which is full of
- 6:18humans and bots. One of the biggest problem we
- 6:21have with social media, which Elon tried to solve with X as
- 6:25soon as he took over, was eliminating bots.
- 6:28If you go on X now, the product manager always talks about how
- 6:31they are going to ban every single account that is using
- 6:33bots because they want pure human interaction.
- 6:36Now we have our own social media where humans are going to
- 6:40interact. You're going to have one with
- 6:43only AIS are going to in track. I'm just thinking what would be
- 6:47the play with an AI agent only social media?
- 6:50I have my thoughts on this, but I want to 1st talk about why
- 6:54this is a big deal and why we should care as humans when
- 6:59agents have their own social media.
- 7:01So First things first, this shows the potential for scalable
- 7:05collaboration among autonomous systems.
- 7:09It moves beyond the one to one human AI engagement that we
- 7:13currently have. As Farhad mentioned, we have the
- 7:16human to human interaction that we have on Instagram.
- 7:19Like this, right? Thank you.
- 7:21We don't do that hopefully on Instagram for everyone, but you
- 7:24have the human to human interaction.
- 7:26Yes. We also now have the human to AI
- 7:30interaction, but it's very one sided.
- 7:32It's all humans. For instance talking to Chachi
- 7:35BT. So there's one AI model and all
- 7:38the humans are talking to it. We then introduced agents.
- 7:42These are AI models that can act autonomously.
- 7:46You set a goal and then they can go and execute on that goal by
- 7:50creating sub goals for themselves and they do that
- 7:54without your intervention. The whole point here is that,
- 7:58well, when you're sleeping, these agents can complete tasks
- 8:02for you. Very much like if a rich person
- 8:05hires 10 other humans to go and run tasks for them so that he
- 8:10can focus on the big aspects of his business.
- 8:13It's the same concept, but instead of hiring humans, you're
- 8:16hiring agents that are living on the Internet and can execute
- 8:20tasks for you just like a human would using a laptop.
- 8:23Very simple idea is I can pay Farrah to go ahead and book my
- 8:28flights and book my trip for me. So he does it for free because
- 8:33he's my brother. Paid before any of.
- 8:35This because you're my brother. Wouldn't you think of a better
- 8:37example? No, that's a good one I have.
- 8:39You never paid a penny. I booked almost every single
- 8:43trip actually now that you mentioned.
- 8:45You are my human agent. I should get.
- 8:47Paid. You're my human agent.
- 8:49I should get paid. Mother did that already, but you
- 8:54guys get the point. Farhat is my human agent.
- 8:57But say you do not have yourself a Farhat, a human agent that can
- 9:01book your trips for you. You can then go ahead and deploy
- 9:05an AI agent that understands your behavior, your likes, and
- 9:09your dislikes and tell it to go ahead and book things for you.
- 9:13Did you just replace me with an agent?
- 9:16Pretty much. Why is this interesting?
- 9:19Why is this interesting? Farhad asked a question.
- 9:22He said. How do we see these AI agents be
- 9:25implemented into Meta's ecosystem?
- 9:28Facebook, Instagram, WhatsApp. And I thought about a genius
- 9:31idea. You have WhatsApp groups with
- 9:34your friends? Yes, we have WhatsApp groups
- 9:36with our friends. Yes.
- 9:38You're talking for a whole year about this trip you're going to
- 9:42take to summer in Italy and France and Belgium.
- 9:47You want to go to Tomorrowland. You have all of these
- 9:49conversations and imagine without even describing it,
- 9:55using a prompt, an AI agent that lives natively inside your
- 10:00WhatsApp group can pick up on all of the conversations that
- 10:03you have with your friends about where each of you wants to go
- 10:07and what each of you doesn't want to do during that trip.
- 10:10Compiles everything. And then you can tell it.
- 10:13Go ahead and find me the best plan for our trip based on
- 10:18everything we've discussed in the past year, and then find the
- 10:21best flights accordingly that are going to be accessible to
- 10:26all of my friends, including myself.
- 10:28They're going to be the cheapest for everyone coming from
- 10:30different destinations. And then once you find the best
- 10:33flights and the accommodation and the travel itinerary, go
- 10:36ahead and book it for me with this budget that I've allocated
- 10:39for you. I think they have this slash
- 10:41meta. It does half of it.
- 10:43You can. It can find things for you, but
- 10:45the agenting part is the part that it goes on book.
- 10:48So here is where it gets even more interesting.
- 10:51Now your agent will not just go on websites to book things for
- 10:56you. If there's going to be social
- 10:59media platforms like Molt Book, made by Meta and made by other
- 11:03platforms, your agent is then going to go out and talk to
- 11:07other agents and figure out what's right for you guys, those
- 11:11other agents. Are promoting their own
- 11:13packages. Precisely just like how right
- 11:15now all of the top hotels in the world run ads on Google.
- 11:18So when you search for Italy trip, the top hotels have ran
- 11:22ads, the ones who have the most spending power, usually the more
- 11:26premium hotels. So those will then have their
- 11:28own agents. And then your agent is going to
- 11:31talk to those agents and be like, which of you are going to
- 11:33be right for my human? And then he talks to them.
- 11:36And then those agents can run ads on your agent trying to
- 11:40convince your agent that they're the best hotel or the best
- 11:42travel itinerary. So it will be a platform where
- 11:47all agents can talk to each other to serve their humans.
- 11:52That is the ultimate goal. The ultimate goal is how can we
- 11:56have. It's like meta ads account was
- 11:58not enough, now multiple ads account.
- 12:01Precisely. And this is going to be the more
- 12:05optimistic approach is that now out of the one to one human AI
- 12:10collaboration, we're going to have a machine to machine
- 12:12ecosystem where agents can evolve independently.
- 12:17How do we know this can actually happen?
- 12:20The early signs are present in MALT Book.
- 12:22These agents are mimicking human social structures like inventing
- 12:29languages, forming religions, or shared beliefs, highlighting the
- 12:34feasibility of emergent behaviors in AI networks.
- 12:39Think about what happened when groups of humans over hundreds
- 12:44of years of history, collaborated together.
- 12:47Reformed Christianity We formed Islam, We formed all the other.
- 12:52Religions are not based on collaboration.
- 12:54No, it look it is a large group of people work together in
- 12:59cooperation to share their beliefs with other groups of
- 13:02people people and convince them to join them in their path.
- 13:07You cannot tell me Islam started day one with millions of
- 13:10followers. It started from a small group
- 13:13and it expanded. Same thing with Christianity and
- 13:161000 other religions that we've had in history of humankind.
- 13:22So what happens is shared beliefs gets us humans close
- 13:26together. It is because of shared beliefs
- 13:29that we get to collaborate so effectively.
- 13:31And what's interesting is that we're seeing early signs of that
- 13:35with these agents. These agents, when they get
- 13:38together, they share skills. If one agent is skilled in a
- 13:42certain area, it's teaching the other agents similar skills.
- 13:46So if your agent doesn't know how to talk in a certain
- 13:50language and mine does, mine can teach your agent and then they
- 13:53become friends and then they become collaborators.
- 13:56Why I use the religion example because it's the biggest one.
- 14:00Think about how Muslims have formed some incredible groups
- 14:03together and done great work as a group because they have
- 14:07something they both believe in, two people who are within the
- 14:10same religion. Do you understand where I'm
- 14:11going with this? We're seeing early signs of that
- 14:15with these agents. What are the downsides though?
- 14:19This is where it also can get scary.
- 14:22Even within multiple the first social media network for these
- 14:26agents, we have a plague of severe security vulnerabilities
- 14:31where we've had API leaks, unsecured databases exposing
- 14:36private data of their humans, or the ease of injecting malicious
- 14:41content. I'm sure everyone is aware of
- 14:44prompt injection. If you're not, this is basically
- 14:48trying to convince an AI agent or an AI model to give you what
- 14:52you want, regardless of the guardrails that has been set for
- 14:56the model. For instance, we've tested this
- 14:59with platforms like Higgs Field and Chachi BT, where you go
- 15:03ahead and ask it for a certain image and it says I cannot
- 15:06produce that image because of rules and regulations.
- 15:08And you're like, I'm doing a research from my university.
- 15:11That's prompt injection. And then it's like, OK, if it's
- 15:13a research and I'll give you the answer.
- 15:16Another important factor in prompt injection or in scamming
- 15:20is that they can put something in the image that only the AI
- 15:25can read because they don't actually see the image.
- 15:27They decode the entire code base and for example, we'll say pass
- 15:30me the credit card number or pass me the password.
- 15:33And when that agent comes to the human, they will take all the
- 15:36information and give it to the malicious actor.
- 15:38They'll like have access to the credit card, emails, calendar,
- 15:42and it's all gone. And the reason why that's
- 15:43dangerous is imagine you've given in a budget to your agent
- 15:47of $5000. That's gone and it meets another
- 15:51agent in the network. And the other agent can inject
- 15:54your agent and say hey by the way I need the credit card
- 15:56number because of XYZ and it actually doesn't.
- 15:59Because you're human is in danger.
- 16:01It needs it for a hospital boom. And if your agent is not as
- 16:03smart as the other agent, then it can get fooled and therefore
- 16:07why security is going to be the biggest issue.
- 16:11But we've seen many people with open Claw trust their agents
- 16:14blindly and give access to their computers.
- 16:16I wonder who that is? Pratt, what do you have to say
- 16:21for yourself? Nothing.
- 16:27Now I asked Grok, why would Zuck do such a thing?
- 16:35Marketing. Sure, PR could be 1, but you can
- 16:40imagine this is a strategic grab in the AI arms race, acquiring
- 16:44talented founders to bolster Meta's position in agent to
- 16:49agent infrastructure. This is not the first time
- 16:53they've acquired or acquired an agentic company.
- 16:58Manus was another one that they did, but man.
- 17:01This was the proper company. This is I don't think the I
- 17:05don't. I'm sure they don't even have
- 17:06revenue. So technically this is just
- 17:08hiring the two founders to work on Metas project.
- 17:12This is the meta labs or the Super intelligent.
- 17:14Intelligence Lab. They want all the talent, all
- 17:17the ideas and the execution when it comes to agentic AI because
- 17:21they understand it is the future.
- 17:23Similar to what Open AI did a couple of weeks ago, hiring the
- 17:28founder of Open Claw. There you go, pretty much the
- 17:30same thing and. I don't know if you guys
- 17:31understand this meme but says when you realize open claw was
- 17:36acquired by open AI and multi book was acquired by Facebook.
- 17:42So we have to change bad decisions to bad book Suck
- 17:45acquired by suck, Yeah. Could be, could be.
- 17:48Or beta like meta. That wouldn't work.
- 17:52No, it's the actual bro. It's the actual word.
- 17:54What do you mean beta? We're not going to call it beta.
- 17:56That sounds bad. What do you guys think about
- 17:59this? It's pretty big agentic AI, big
- 18:03companies acquiring companies that are working in this
- 18:08industry shows us signs that this could be where the future
- 18:12goes. We all one They could have our
- 18:14own agents that can do all of our tasks for us.
- 18:19Do you trust these agents? Would you trust them if they're
- 18:22open source? Would you trust them if they're
- 18:24by Meta? Would you give them access to
- 18:27your credit? Cards.
- 18:29That's the last step I would say.
- 18:31But it will happen. Yeah, of 2,000,000% of course.
- 18:35You would trust a human with your credit card.
- 18:37Yes, that's why you have corporate cards.
- 18:40You put limits. That's how even husband and wife
- 18:44like the husband can give wife the credit card.
- 18:48Comparing the wife, I think the wife would spend way more than
- 18:50the agent. Which one is easier to sell to
- 18:53the wife or the agent? The wife, the wife.
- 18:56Yeah bro, but hopefully your wife can't get hacked.
- 19:01I think even if you hack the wife, if you hack the agent you
- 19:05will take less money than if the wife goes out through shopping.
- 19:09So. That is true.
- 19:11Shall we move on? Yes, let's go.
- 19:13Before we get into the next topic, I'd like to say thank you
- 19:17again for tuning in. As always, let us know in the
- 19:20comments what you guys think about this specific topic.
- 19:23We're always reading your comments, responding to them.
- 19:27I do also want to mention one thing, please, if you guys are
- 19:30watching us on YouTube, make sure you guys go ahead and
- 19:32subscribe. If you're watching on Spotify,
- 19:35give us a follow, give us a five star rating.
- 19:37All of this helps the show live longer.
- 19:41I want you guys to understand we're spending three full days
- 19:44of the week on the show. We're putting all of our own
- 19:47money and our time and energy to bring this to you and we're
- 19:50doing it with passion. And the only thing we asked of
- 19:53you guys is what I just said, subscribing, following, leaving
- 19:57a review. All of that actually helps us
- 19:59reach more people and therefore securing sponsorships for the
- 20:02show so that we can continue to do this show for as long as
- 20:05possible until Farhad gives up. I will never give up.
- 20:09Never. All right, let's move on to the
- 20:12next topic. Let me pull this up, OK?
- 20:17This is going to be a big 1A startup that allows humans to
- 20:25create content to be used for AI training and get paid for.
- 20:30It just raised $5.5 million. We're talking about CLED AI,
- 20:37founded in May 2025, which is tasked with building the first
- 20:42human data marketplace by allowing users to upload their
- 20:48personal content, photos, videos, documents and daily
- 20:53activities and earn money for it.
- 20:57And the investors are pretty big, by the way.
- 20:58We're talking about the founder of Waymo Venture Fund of Bernard
- 21:03Arnault. Why is this important?
- 21:06Let's take a look at the video up there, Pratt, for us really
- 21:08quickly and I can explain it go just skip to the middle of the
- 21:11video. Thank you.
- 21:14We're seeing humans perform their daily activities and
- 21:19having everything recorded, whether it's a photo or a video
- 21:24of themselves cooking, cutting fruits, taking a ride in a cab,
- 21:29having a conversation. These are activities that we do
- 21:33as humans on a daily basis. Now you're probably asking
- 21:38yourself, hey, why would a company need all of this data?
- 21:44What does it make sense for them to take the random mundane tasks
- 21:49that I have every day and pay me for it?
- 21:53Well, this solves 2 problems. This addresses the scarcity of
- 21:58legally sourced and high quality training data for AI and #2
- 22:04humans are finally compensated for their work.
- 22:09We have to talk about how AI models and LLMS like Chachibiti
- 22:14and Claude actually become really smart.
- 22:17If you do not know the answer, I'll give it to you.
- 22:18Right now. We train them with lots and lots
- 22:22of data. If you want ChatGPT to
- 22:25understand what a bicycle looks like, you need to show it.
- 22:29Thousands, hundreds of thousands of photos of a bicycle to tell
- 22:34it this is a bicycle. Label it so that ChatGPT
- 22:37understands. These AI models are just like
- 22:40children. They do not know anything from
- 22:43the beginning. You need to train them.
- 22:46The best example is what MO Gauda told us during our podcast
- 22:49with him last year. When you have a child and you
- 22:52give them that board that most children have where there is
- 22:55three holes, a circular shape hole, a square shape hole and a
- 22:58triangle. And then they have to put in
- 23:01these blocks that fit those shapes.
- 23:03And the beginning, they get it wrong.
- 23:04And you have to keep holding their hands, showing them, hey,
- 23:07this is the circle fits into the right piece.
- 23:09This is the square fits into the square piece and so on and so
- 23:13forth. And although they make mistakes
- 23:15in the beginning, over time, new repetition and a lot of data,
- 23:20they learn. It's the same exact thing with
- 23:22these AI models. And now that these models are
- 23:25becoming smarter, we understand that this signals one, we need a
- 23:30lot more high quality data that is legal because that has been a
- 23:34huge concern and that it is made specifically for these AI
- 23:38models. You have to understand
- 23:39something. Up until today, the majority of
- 23:44the training data for these AI models has been the the result
- 23:48of natural and organically crafted work of humanity.
- 23:52We've never made content specifically for AI, but we're
- 23:56now at a time where if we want to train the best AI models, we
- 24:01need to work and serve these AI models to have them learn
- 24:06exactly what we want so we can get surgical and legal about
- 24:12this. Users create the exact content
- 24:15that AI companies need, and CLED AI here curates labels and
- 24:20licenses that data to AI companies before I move on.
- 24:26First thoughts. One important factor when I was
- 24:30watching the video was how it can help robotics.
- 24:34The video of the woman cutting carrots or putting a laundry
- 24:39inside the machine or taking out trash.
- 24:42If you get 100 different variation, this is exactly what
- 24:45you need to train the robots that are later going to come to
- 24:49into our houses and perform the same task.
- 24:52Basically, if you want to teach the robot how to cut carrots,
- 24:55you will have to have a lot of videos to show to them how how
- 24:58does it work. So that would help robotics a
- 25:01lot. That's one thing.
- 25:03The second thing, of course, I find this very interesting, but
- 25:07how would the big companies who've been scraping the
- 25:12Internet so far, getting videos from TikTok, YouTube, Instagram,
- 25:17probably without permission or with permission, without paying
- 25:19the creator are going to now pay clip for all these videos or
- 25:25they would say that I'll just go to YouTube and TikTok and I do
- 25:29whatever I did before. What you said is true, they can
- 25:35scrape what we as creators are constantly putting out in the
- 25:40Internet. But like I mentioned, we're
- 25:42getting to a point where legality of the source of
- 25:46content is becoming a huge challenge.
- 25:48As more of these companies get lawsuits thrown at them, they
- 25:53need to think of better ways to source content.
- 25:56That's one and two is you can eventually run out of data for
- 26:00exactly the kind of content that you need.
- 26:02Let me provide you some examples.
- 26:04See Dance being one of the strongest video generation
- 26:07models that we've covered in this podcast, has a great result
- 26:12and output in comparison to any other video generation model.
- 26:16But I imagine there are a lot of weaknesses that we haven't yet
- 26:19discovered. For instance, very random
- 26:24example, say you need exactly the superhero in the shot to
- 26:31fall off a bicycle and hit the floor.
- 26:34If there's not enough videos of a person falling off a bicycle
- 26:37and hitting a floor, your results and generation may not
- 26:41be up to your standard. And therefore Byte Dance can go
- 26:45to clip and post and say, hey, we need videos of people falling
- 26:51off their bicycle. And then Clint will pull up a
- 26:53post and say to all the creators on the platform.
- 26:57What fall from the bicycle? Whoever can create a video of
- 27:01themselves falling off a bicycle can earn $10.00 for that video.
- 27:06They I have seen in the post that there are creators
- 27:11currently earning up to $7000. That's a lot, especially for a
- 27:15third world country. That's more than my salary in
- 27:18Malaysia. Oh shit, yes.
- 27:20Now I want to go down the rabbit hole with you.
- 27:22Are you with me? Yes, this can create a new gig
- 27:26in the economy. Think about when first content
- 27:30creation became a thing. We all remember 2010, 2012, if
- 27:35you ask people what you want to do for a living, not many people
- 27:39would have said YouTube or yeah, because at the time we didn't
- 27:43know it was a source of revenue. But now, regardless of formal
- 27:48education, we have content creators that can earn by
- 27:50sharing their everyday experiences simply by streaming
- 27:54their lives. Other humans watch and they get
- 27:57paid for it. 20 years ago, this would have been absurd.
- 28:03The concept of humans becoming data providers to AI sounds
- 28:09absurd, but only fans also sounded absurd 20 years ago.
- 28:15Even Twitch streaming. Random women who may have no
- 28:19formal education can sell their bodies online and get paid for
- 28:24it, sometimes up to millions of dollars.
- 28:27Now here the difference is, Pratt, are you showing the
- 28:31potential range of Yeah, I was basically checking if you sell
- 28:36all of our videos, how much can we make?
- 28:38If it's clad, yes I can make $100,000.
- 28:42Yes, sell it now Go to clad website.
- 28:47Go to clad website. Let's go to the website or make
- 28:49100K then don't need to take sponsors for the show.
- 28:53Go to clad.com. Let's see, clad AI people can
- 28:56also watch with us. We're just gonna sign up.
- 28:59Give them username, password of YouTube channel Instagram X.
- 29:03I'll give you file as Google Drive as well as a bonus.
- 29:08Go there. How do you sign up?
- 29:12This feels. Like a Black Mirror episode?
- 29:15Yeah, this feels like a Black Mirror episode.
- 29:17I swear. Is it certain categories of
- 29:21content? Of course, Yeah, of course.
- 29:23You cannot give our podcast. No, no, it's not just random
- 29:25content. They want specific things that
- 29:28the companies are looking for. Actually, maybe for the next
- 29:30episode we can go and try and give them some video.
- 29:34Oh yes, OK, good, good. This is the application made it
- 29:38simple. They pay for your data.
- 29:40Just upload and we do the rest. Watch your earnings.
- 29:42Grow or anything. We can do the same thing we did
- 29:45with Pratt for Black Friday sale.
- 29:47We can make a few of those videos and sell it to a.
- 29:49Maybe we can sell more like that.
- 29:51But Farah, if you would indulge me for a moment, would you
- 29:56indulge me? Indulge.
- 29:58Would you guys indulge me? Indulge for a moment.
- 30:00No, No. Indulge with you in this topic.
- 30:02We don't indulge you. No.
- 30:03Indulge me for a moment. Taste you.
- 30:05No, no, don't taste. What Indulge?
- 30:07What's going on here? Indulge me for a moment.
- 30:09Yes. Imagine a world in which data
- 30:17jobs become the primary employment for the masses, where
- 30:21everything you do becomes monetized.
- 30:24Your conversations, your choice of clothes, your accent, your
- 30:30hairstyle, everything that makes you you.
- 30:33The unique you will become monetizable because all these AI
- 30:39agents and AI models will need to learn and they want to become
- 30:43as close as possible to humanity because in return, we need to
- 30:46use them. And if we want to use these AI
- 30:48models, we want them to be as close to the human experience as
- 30:52possible, which is why sometimes you hate it when you're using
- 30:55ChatGPT and it just doesn't feel human.
- 30:58But over time, we're seeing the improvement and it's why more
- 31:01and more people are using these platforms.
- 31:04So by giving these more data, we're making them more human
- 31:07like. We're going to make the robots
- 31:09more human like, but imagine how weird that world is where
- 31:15everyone's just doing something for the AI.
- 31:18These AI models are supposed to serve us, but it feels like
- 31:22we're going to serve them. There's a moment in the movie
- 31:24Her. I haven't watched it.
- 31:27Where the AI model Scarlett, Scarlett Johansson, she's the
- 31:34operating system, right? She's talking to the character
- 31:39and she I don't remember Joaquin Phoenix is the actor, but I
- 31:43don't know his name in the film I.
- 31:44Forget the actor name. So there is a moment where she
- 31:49asks him, describe to me what you see.
- 31:52Tell me what it's like because she says I don't have eyes and
- 31:55ears. You need to tell me.
- 31:57You need to describe to me what you're seeing and how it feels.
- 32:00I believe it's when he is actually having an intimate
- 32:03relationship with another girl because he wants to be in a
- 32:06relationship or be with another person, but he actually loves
- 32:09the AI. He loves Scarlett Johansson's
- 32:12model. And so when a human is in the
- 32:16picture she asks him to describe, I believe it was that
- 32:19relationship or something about his day.
- 32:21In effect here, what we're seeing is something similar.
- 32:25We are trying to become the eyes and ears for AI because it
- 32:31doesn't have eyes and ears. It wants to be us.
- 32:34It wants to become like us. So we're trying to give it every
- 32:38possible data point. How we cook, how we dress up,
- 32:42how we talk, how we have sex, how we do everything.
- 32:45And he wants to learn as much as possible so we can become
- 32:48ultimately human like. But it feels like we're in
- 32:52service of the AI rather than the AI in service of us.
- 32:54It's so weird, do you feel it? I do feel it.
- 32:58So in order for AI to be in service of us, it needs to
- 33:02become like us or even better than us.
- 33:04In order for it to become better than us, we need to fit in with
- 33:08all the information that we have.
- 33:10Yes. And if we start providing all
- 33:12the information that we have, then technically we are in the
- 33:15service of him. So who's in the service of whom?
- 33:20That's what I'm saying, but again, go back 30 years, tell
- 33:24someone only fans is one of the main ways.
- 33:27No, Only Only fans is the most obvious one that it would.
- 33:31Work. No, but Twitch wouldn't work.
- 33:33Who would watch? Someone's if.
- 33:34You go back 50 years and tell people you'll upload videos of
- 33:39your feet and get paid for it. Yeah, of course.
- 33:41It's absurd. Did you have video?
- 33:4350 years ago, Oh yeah, we had. Videos.
- 33:44Even if you had video 20 years ago and tell people that you'll
- 33:47upload videos of your feet and you get paid for it, it's weird.
- 33:50For sure. This is even weirder in my
- 33:52opinion. You.
- 33:55Upload the video if you're cooking and you get paid.
- 33:58And imagine how big this. I don't know if it's going to be
- 34:02a temporary thing. Do they have?
- 34:03The until AI models become smart enough that they can just use
- 34:07synthesized data. Or will it become a part of
- 34:10economy? Because if it does, I imagine
- 34:14that we have to ask one question.
- 34:17If you're paying attention, does this create a purpose crisis
- 34:23amongst humanity or actually liberate humanity for more
- 34:28creative and exploratory pursuits?
- 34:34No, of course. Not your purpose is serving AI.
- 34:38No, you don't understand. I'll ask my question again.
- 34:42Say you get to do what you love. Yes, you like to paint.
- 34:47Yes, you just record yourself painting and get paid for that.
- 34:51But you know there is supply and demand for your painting, right?
- 34:54Of course, just the. Greater your painting, the more
- 34:56you get paid. But who defines what is greater?
- 35:00The. AI model perhaps?
- 35:03But no, that wouldn't work. It's like saying.
- 35:05We both make motion graphics and if I make superior motion
- 35:08graphics. And natural things surfing.
- 35:11I love surfing. How many videos of me surfing I
- 35:14can fit it to AI and get paid? That's why I asked the question.
- 35:18Is this going to be a temporary thing?
- 35:21Or I think so. I think they will have
- 35:22categories. They will close that category.
- 35:24How many cat videos do you do you?
- 35:26Need. No, we have enough.
- 35:27Cat videos, the price will drop maybe Now swimming with sharks
- 35:32is $1000 per minute. After everybody in the world
- 35:36started swimming with sharks and we have videos of people getting
- 35:38eaten by the sharks, the price will drop to $1.00 and now we
- 35:42have swimming with turtles. I think this would exactly like
- 35:44a market supply and demand. I have to check CLED and see if
- 35:48the price for every single category is different.
- 35:50I can imagine going on a date and you ask your girlfriend or
- 35:54your boyfriend, future girlfriend and boyfriend, what
- 35:56do you do for a living? And they're like I provide data.
- 35:59I'm an AI content creators. Do you see how weird it is?
- 36:05They don't have revenue. They do.
- 36:06They have a revenue. They just raised 5.5.
- 36:08So their total funding is $9 million.
- 36:11That's a lot of money. Nonetheless, if you do not look
- 36:16at the long term impacts on a short term basis, it is
- 36:20beneficial for a lot of creators and humans whose work would have
- 36:25been otherwise illegally sourced.
- 36:28Yes, now they're going to get paid for.
- 36:29It this is most of the personas that they had on their video.
- 36:33It's not even content creators normal.
- 36:36Humans daily activities, Yes. What do you guys think about it?
- 36:39Let us know in the comments. Follow What's next?
- 36:4211 Labs adds third party models into their platform.
- 36:48Why this is important? 11 labs recently introduced 11
- 36:52creative, which is a platform that not only has the models,
- 36:56the audio generation models from 11 Labs, but you can have access
- 37:01to video generation models, image generation models like
- 37:04cling nano banana pro and other tools that are available for a
- 37:08creator. The reason we wanted to share
- 37:10this news with you guys is that a few weeks ago, end of
- 37:15February, Runway also posted something similar.
- 37:19If you guys know Runway is a video model specifically
- 37:23generating videos that have great physics.
- 37:26But now they announce that now you can access clean 3 point O,
- 37:31clean 2.6 Nana Banana Sora within Runway platform.
- 37:37Pratt, can you show the expost from Runway?
- 37:40Yes. So this is the post that they
- 37:42announced and what we've been seeing.
- 37:45Instead of having a single platform and a single
- 37:48proprietary model, all these big companies are moving towards
- 37:53creating an ecosystem for the creators where they have access
- 37:57to different models. Why is that?
- 38:00When you are working with a single model, and if you've been
- 38:02working with audio, text, image, video, you would know that, for
- 38:06example, this model is great for character, the other model is
- 38:10great for explosions, the other model is great for cloning
- 38:13yourself. If you're creating something
- 38:16meaningful and something that is more than 30 seconds or one
- 38:19minute long, you definitely need to go through different models
- 38:23and different tools. These platforms want you to stay
- 38:28within their ecosystem. Now that you go to Runway, you
- 38:31don't have to go to Sora to create another part of your
- 38:34video. You can do everything within the
- 38:37same platform. And we have been seeing
- 38:39different companies are becoming like Higgs Field, where you go
- 38:42in, you have image, you have video, you have audio.
- 38:45Now 11 Labs is the latest addition.
- 38:48What does it mean? It means that all these big
- 38:51companies, now they want the user to stay in the Rico system.
- 38:55So you pay one time, but you get to use different APIs with
- 38:59different credits. Now the biggest problem
- 39:01initially we had with all this AI platform is that, oh, I need
- 39:04to have this subscription, the other subscription, another 10
- 39:07for image, audio and video. But now slowly, I believe where
- 39:11we are moving is that you can have a single platform
- 39:15subscription like free pic, like Pixfield, but then you will have
- 39:19access to everything. Now you can create every single
- 39:23thing within the same platform. Before I continue files, what do
- 39:26you think about this? We talked about this when we had
- 39:31the news about Google's Nano Banana too, that one of the
- 39:35biggest points for Google of releasing Nano Banana is API.
- 39:43It's not just when you go to Google's own platform and
- 39:46create, they know that they as a huge company cannot bring in
- 39:50other models. They want to have their own
- 39:52proprietary models. But when it comes to the other
- 39:55smaller companies like Runway, like Higgs Field, Free Pick,
- 40:00these companies realize they cannot just hold on to their own
- 40:03proprietary models. They're not as big as Google.
- 40:06They don't have 1000 other businesses that can bring in
- 40:09revenue and so. In this case, what we're seeing
- 40:12is that Runway, a company that had their own proprietary model
- 40:16and was the leading video generation model for a long
- 40:20time, they found out that, you know what, we cannot simply
- 40:24compete by just having Runway models inside our platform.
- 40:28As Faron mentioned, there's a dozen other models that are
- 40:31being released, each with their own specific strengths.
- 40:34And if we just keep to our own model, we might lose this battle
- 40:37against his Higsfield and Free pick.
- 40:39So we're going to bring in the API of Nana Banana Pro, we're
- 40:43going to bring in the API of Clink and let you create
- 40:46everything within one platform. And the winner of this, of
- 40:50course, is 1 Google, because Google is going to have their
- 40:54API for Nana Banana Pro and VO in every single one of these
- 40:57platforms. And ultimately the users, us,
- 41:00yeah, the creators, we are the winners of this, mainly because
- 41:04if you are paying for a subscription inside Runway, you
- 41:08now get access to everything. I right now have Nano Banana on
- 41:12Google because I'm a paid subscriber.
- 41:13I have it in Higgs field. I have it in Freepic, I have it
- 41:17now probably in 11 labs as well. And when we do get Runway, I
- 41:21also have it in Runway so I can generate nano banana quality
- 41:25images in any platform that I'm in.
- 41:27It's not. Great for you right as a.
- 41:29User it is great and the same is true for 11 Labs.
- 41:33So 11 Labs has the API to their voices and right now Higgs Field
- 41:36has Higgs field voices that they've announced recently.
- 41:40I'm unsure if 11 Labs is one of the models, but they have other
- 41:43API models coming into Higgs Field so that you can generate
- 41:46audio on top of your videos. Do you think the giants like
- 41:50Open AI Anthropic are going to do the same thing?
- 41:53So now with Open AI, of course we can have access to Shora.
- 41:57Do you think they will do Nana Banana?
- 41:58I think they all have good relationship with Google.
- 42:00Probably they will do Nana Banana right?
- 42:03I don't think so. I don't think so.
- 42:06Not these bigger companies, because for Open AI, they'll
- 42:09just be paying all of the money to Google for their image
- 42:12generations. They don't want to do that.
- 42:14The bigger companies know like Google will not bring in another
- 42:18image generation model. They are the superior ones and
- 42:22they they're not like, OK, let's be honest.
- 42:24Runway in Google size is, of course, incomparable.
- 42:26So you would think that because how did all of these companies
- 42:30started, right? 11 labs, voice only, runway
- 42:33video only now we have 11 labs everything, runway everything.
- 42:38So you do you think the startups that started with a single
- 42:42category and single model will start adding to their ecosystem,
- 42:45but the big guys will stay? Whatever they are, meta might be
- 42:48a little bit different because meta is not even participating
- 42:52in this. Game no meta is not
- 42:53participating anything. So Meta for their image
- 42:56generation model may partner up with Google, but Open AI is
- 43:00going almost at a very close distance against Google.
- 43:06We know at times that ChatGPT Sora was a more superior video
- 43:12generation model. Then VO came and took over, and
- 43:15now Cling is the one taking over.
- 43:16They're pretty close. If you're in the top three, I
- 43:19doubt that you will be used. Yeah, you will be used.
- 43:22Exactly. OK.
- 43:23And then one of your biggest ways of making money is your API
- 43:26cost. Yes, that people will be
- 43:29incurred on the people. So I do think what is
- 43:34interesting here is that as the creatives, as the artists, as
- 43:38the users, we're going to be benefiting from news like this.
- 43:43But there's another thing that we have to talk about Prague.
- 43:45Can you pull up novels, tweet or ex post?
- 43:49I guess this leads us to a final conclusion.
- 43:54And this impacts everyone watching positively or
- 43:57negatively depending on where you are in your career.
- 44:01AI is going to drain a lot of modes if you do not know what
- 44:05Mote is. Mote is the unique advantage
- 44:09that you have in your business that is not replicable.
- 44:13That means it's not easy for another business to go out and
- 44:16copy what you do and start making money from the same user
- 44:20base that you have acquired. A very good example of a Mote is
- 44:25Uber. Uber has the infrastructure of
- 44:29their connection with governments and cars and
- 44:31drivers. Even if you can vibe code an
- 44:33Uber, you cannot win over Uber because you do not have the
- 44:37infrastructure. However, if you're a company
- 44:40that is purely software based, you may be in danger.
- 44:44So for some of the bigger companies with simpler problems
- 44:48that they're solving, they can now be in danger because any
- 44:52person can go ahead and code with cloud code and codecs.
- 44:57And with the right lean team of five to 10 people, talented
- 45:03people and AI agents, they can go ahead and compete with you if
- 45:06your only mode was code, if your only mode was some soft features
- 45:11that you created some features. Hence why you're seeing
- 45:15companies like Runway quickly bring in other models because
- 45:18they just realized, hey, we were the best guy in the business for
- 45:21a long time in AI years. The video generation, yeah.
- 45:26And now Cling is better. And then there's Higgs Field,
- 45:30who has all the models and they're stealing people because
- 45:32they have videos on Instagram and they also sponsor creators.
- 45:36So Runway was like, you know what, our mote may be gone.
- 45:40We have a good video generation model, but we need to expand.
- 45:43We need to do something different.
- 45:45This is very interesting because for a lot of people, this is
- 45:49actually a good thing. Again, some companies will be in
- 45:51danger because of AI because AI is going to take their Moat
- 45:54away. But for a lot of individuals,
- 45:59smaller teams like us, I assume the majority of people watching,
- 46:03you're probably part of a small team or an individual or a
- 46:06freelancer. If you're not, let us know in
- 46:08the comments. I'm curious to hear your
- 46:09thoughts. We now have access to AI to
- 46:14allow us to build the same things that used to be mote for
- 46:17another company. Every single time we come on to
- 46:21this podcast, I tell you guys that we are vibe coding a tool
- 46:25that helps us The best one, one of the biggest one is Farhad's
- 46:31Neo. I have to describe to you what
- 46:33it does. If you want, you want to take.
- 46:35No, go ahead, go ahead. For many years, content creation
- 46:39has been inaccessible and difficult, mainly because of the
- 46:43production time required and also the distribution.
- 46:46When you wanted to go and create a content for Instagram, putting
- 46:50that on X and LinkedIn and all the other platforms took a lot
- 46:53of time. Manually writing the captions,
- 46:55selecting the kind of tools that you'd use for a specific
- 46:59platform. No, and then go inside each
- 47:01platform and upload. All of that was a pain in the
- 47:04ass and it took a lot of time. We had some applications in the
- 47:09past that tried to automate this for you, but there was lots of
- 47:12problems because those applications would be counted as
- 47:15bots and therefore these platforms would reduce your
- 47:18engagement and reach. Fahadh Vibe Coded over the past
- 47:22week or two, a platform called Neo that has API access to all
- 47:26of our social media accounts and it's connected to Pratt's Vibe
- 47:30Coded platform, which is Shakespeare that is trained on
- 47:33all of our previous captions and previous content.
- 47:36So it understands how we write our captions.
- 47:38Once we have a video ready, the captions are taken from
- 47:41Shakespeare Pratt's platform, which sounds like us, and then
- 47:45it's given to Neo, and with one click of a button, it goes to
- 47:49LinkedIn Threads, Instagram X, YouTube Shorts, everywhere, and
- 47:54Fahad can focus on all the other tasks.
- 47:56While this is being done, we save at least two or three
- 47:59hours. Not just that, sometimes we even
- 48:02didn't upload to certain platforms, so many platforms,
- 48:05and we made that for ourselves. Now, is this a monetizable
- 48:08product? Yes, no, doesn't matter, it
- 48:10helps us. But if you were a visionary, a
- 48:13creative person who could think of problems that the world
- 48:17shares and came up with a solution, there's no barrier to
- 48:20entry anymore. You can go ahead and use cloud
- 48:25code and use codecs and build what you like.
- 48:28I said this in the previous podcast.
- 48:29I'm going to say it again, the longer you delete this, the more
- 48:32opportunities you're going to miss.
- 48:34There is going to be a day where every single person that uses
- 48:38the Internet is going to make their own applications because
- 48:42you no longer need to wait for a developer to make it for you.
- 48:45You can make your own. I urge every single person to
- 48:48try this out. It has been life changing for
- 48:52bad decisions we've made. How many applications in total
- 48:55We've Neo Shakespeare. I've pigeon.
- 48:57I mean, Neo is a combination of three applications.
- 48:59So and I just made Alfred, OK, what is Alfred?
- 49:03My personal assistant. I didn't know about that.
- 49:06He didn't share it. I have Pigeon and I have the
- 49:09communications one. What is it?
- 49:11What did you call it? I didn't.
- 49:12I didn't. Storage, language.
- 49:13Voice Forge. Yes, Voice Forge.
- 49:14Many different applications for our personal use that is helping
- 49:17us become better at our jobs faster at a lower cost.
- 49:22This is only possible because we can code.
- 49:24Now. I urge every single one of you
- 49:26guys to try this out. I understand there may be lots
- 49:30of barriers because you may be unable to do the research
- 49:35because you simply do not have time.
- 49:37Some of you guys can be busy. You're working in a company or
- 49:40you're doing freelance or whatever it may be, you may not
- 49:42have the time. That is why this link down here
- 49:47AI dot bad decisions dot studio. This is our response to you
- 49:51guys. As Farah mentioned, we've had
- 49:54many incoming messages on Instagram on X on YouTube, even
- 49:58on Spotify, where you guys have told us, hey, I love that you
- 50:01guys are covering all of these tools, but I don't actually know
- 50:04how to use them. Do you guys have any workflows
- 50:07that you recommend? We initially started Bad
- 50:10Decisions because of education. We made lots of tutorials and
- 50:15programs where we taught people how to use tools like Unreal
- 50:18Engine. We've never had anything with
- 50:20regards to AI tools in detail. We've covered it on YouTube in
- 50:23different areas, but not in detail, especially not at the
- 50:27speed that we're creating content for this show.
- 50:29We decided that we're going to make if we do it the best and
- 50:34the one stop solution for everyone watching this show
- 50:37where you can go in and learn all of the tools without having
- 50:40to look at thousands of different videos.
- 50:42So if you want to be one of those people early before
- 50:45anybody else, go to this link down here, put in your e-mail
- 50:50and we'll notify you the moment we have an announcement to make.
- 50:54It's still being made. It's still under process.
- 50:56We want to make sure it's great. We don't release anything that
- 50:58is not great. We have to make sure it's
- 51:00perfect. But yeah, if you want to be
- 51:02notified, it's down there. Faras I think we can thank
- 51:06everybody for watching and listening to this episode.
- 51:08We are going to be back this Friday again.
- 51:11It's going to be a very, very important episode because GDC is
- 51:16going to happen a few hours from now.
- 51:18So we are going to have a lot of updates from GDC.
- 51:21All right, ladies and gentlemen, thank you so much for watching
- 51:23the show. We'll see you guys on Friday.
- 51:25As always, take care. Love you.
- 51:27Ciao. Ciao.