Latest / Bad Decisions Podcast / 84% of Humans Have Never Used AI
Transcript
- 0:01Do what? Happy Monday, everybody.
- 0:02I guess you were surprised. That's how you wanted the intro?
- 0:06Is that real? Yeah.
- 0:07Did we just start? Yes, we.
- 0:08Did OK, I'll write you my surprise.
- 0:10I didn't know because usually we do a three to 1 and then.
- 0:12You 32. Happy Monday, everybody.
- 0:15We are back again with another show where we break down
- 0:18everything related to tech, AI and business and we are coming
- 0:21back to you every Monday, Wednesday and Friday.
- 0:24So what did you guys do over the weekend?
- 0:27What did you do over the weekend?
- 0:29I cannot remember. Oh, we did a live stream.
- 0:31Yes, yes, we did a live stream for our community.
- 0:33It was great. I remember I had the great
- 0:36Argentinian sandwich. OK, that's pretty good.
- 0:39Why don't you don't want to talk about the reading, the working,
- 0:43You just talk about eating. That's the first thing that
- 0:45comes to my mind. That's the first thing that
- 0:47comes to my mind. What about you guys?
- 0:48What were you guys up to during the weekend?
- 0:50Let us know in the comments. So today's going to be a banger
- 0:54episode. We're going to be talking about
- 0:57an article that went viral on X about global AI adoption rates
- 1:02and why most people are wrong about the numbers, which is
- 1:05quite fascinating because we ended up being wrong too.
- 1:07Then we're going to be talking about clod codes, no?
- 1:09No, no, wait a minute. We were wrong too.
- 1:12Yes, you too. Last episode, last episode,
- 1:15everybody, we knew that I got the facts right.
- 1:18This episode, I'm back with amazing facts and figures.
- 1:21Jensen didn't attend AI Summit. We all remember that.
- 1:25We all remember that you got. 1. Fact, right?
- 1:28Let's see. Let's see.
- 1:29We are entering this episode. We're going to get a lot of
- 1:31facts on that cloud. Code with a new feature that
- 1:34completely shook the market. Yes, exactly.
- 1:36And according to a request from one of you guys, we're going to
- 1:40be taking a look. We add Microsoft Trellis, which
- 1:43is a 3D asset generation model. Looking at the results and
- 1:48seeing it, comparing it to other models like Hanyon, which is a
- 1:51Chinese model, and seeing how good or bad they are.
- 1:55Shall we start with the article? Let's go.
- 1:57This first article, which went viral on X, has about a million
- 2:02views by Noah Epstein. Whoa, what a name.
- 2:07Noah Epstein. Shout out to Noah.
- 2:09Shout out to Noah, not Epstein. I'm pretty sure the last name is
- 2:14causing controversial right now. But thank you for this article
- 2:18actually. How does he introduce himself
- 2:20Epstein? No more.
- 2:21I don't think so. I think he just goes with Noah
- 2:23right now at this point. This article recommend every
- 2:28single one of you guys to read. What is it about?
- 2:31It's about global adoption rates of AI and how most people are
- 2:36false, including us and Farhat. And I'll explain why.
- 2:40Pride. Can we go all the way down to
- 2:42the the image? So before I start talking about
- 2:47this image, I just want to tell you what's written in the
- 2:49beginning of the article in a summarized manner.
- 2:52It talks about how you are still early.
- 2:56Your social media feed convinces you that you're not, that AI is
- 3:00taking over the world, and that you're being left behind.
- 3:03That AI is saturated. When we wake up, we believe that
- 3:07the world has lapped us. New models, new agent
- 3:10frameworks, new updates. Open Claw this, open AI that,
- 3:14Meta this. Someone automated their entire
- 3:18company overnight. Artists are cooked.
- 3:21Coders are cooked. A 19 year old just made a $10
- 3:25million startup from their dorm. All of that sounds like you are
- 3:30being left behind, but when you actually look at the results,
- 3:34the reports, the data, we are actually wrong.
- 3:38So this photo here, if you pay close attention, the world
- 3:44population is 8.1 billion humans.
- 3:47I'm not considering AI agents. There's 2500 dots on this image,
- 3:54and each dot represents 3.2 million people.
- 3:57So the entire dotted graph is the entire population of the
- 4:02world. Now if you look at the grey
- 4:05dots, which makes up most of that square grid, that's the
- 4:10number of people that do not use AI, which is 6.8 billion people.
- 4:15That's a. Huge population.
- 4:17That's 84%, 84% of the population of planet haven't
- 4:24used AI. So when you open your X, when
- 4:27you open your Instagram and you are in a bubble of seeing other
- 4:30people posting about all of the things, you are not late.
- 4:34Yes. And we are looking actually at
- 4:36the bottom of the graph. So what do we see there?
- 4:38The green dots that you see are the people who use AI, but they
- 4:43are your free chatbot users. Shout out to our mom and dad.
- 4:47Exactly like this is your parents that are using ChatGPT
- 4:50right now. That makes up for 1.3 billion of
- 4:54the entire world's population, which is 16%.
- 4:57Now when we're talking about a professional use, which is all
- 5:01of us watching and listening right now, this is not the same
- 5:04category. So all of those dots that you
- 5:06see, the green ones, you know what they asked?
- 5:09They're like, how can I make macaroni and cheese for my
- 5:12girlfriend tonight? Or what should I respond to this
- 5:15girl on Bumble? Do you understand what I'm
- 5:17saying? That's your green dots.
- 5:19Where do we sit? Pay close attention to the
- 5:23bottom right? So right, right here.
- 5:25Your yellow dots are the people who pay about $20.00 a month for
- 5:30any AI subscription. That's 15 to 25 million people
- 5:35in the world, which makes up 0.3% of the world's population.
- 5:42That's tiny. That's tiny.
- 5:44So all of the stuff we see online.
- 5:46By the way, we are not in that category.
- 5:48We pay way more than $20. We just pay $200 or.
- 5:51More $20 or more right I. Consider myself in the last.
- 5:55So that one little. Square, we are paying at least
- 5:57$500 a month for AI at least. So that one last square is your
- 6:02red dot and that is your advanced users of coding, your
- 6:07people who are paying for Opus 4.6 codecs, 5.3.
- 6:12They're the people who are developing or building tools
- 6:15using these softwares or AI agents.
- 6:18And that's 2.5 million people only, which makes up 0.04% of
- 6:27the entire world's population. So when you're going on X and
- 6:31Instagram and you're seeing all of the AI craze, you're looking
- 6:35at 0.04% of the world's population.
- 6:39That's a crazy observation and when you look at it in the
- 6:43graph, in this graph manner, you will understand how much
- 6:47opportunity is actually out there.
- 6:49Imagine the entire section that has never used AI or only using
- 6:54free chat bots are looking for AI, consultant looking for AI,
- 7:00architect, looking for people to come and help their business
- 7:04with AI. Whether that help could be
- 7:06automating certain workflow, that help could be helping them
- 7:10with their data database, helping them with their
- 7:12day-to-day business. I think this is going to be
- 7:16bigger than the Internet boom that we had during 2000s.
- 7:22And something very similar happened in the 2000s.
- 7:24Mark Cuban was talking about this, where he would go and
- 7:27knock up on every office in the area and say everyone, you guys
- 7:32need to use machines, computers, you need to set up websites for
- 7:37your business. And at the time, everyone was
- 7:39making fun of him, all the entrepreneurs and the businesses
- 7:42and the CE OS were like, what the fuck are you talking about?
- 7:46I have Lucy and Sally and Adam who are doing the work here on
- 7:50the counter. We don't need websites.
- 7:51Customers walk in. We don't need a digital website,
- 7:54a digital footprint. Well, it ended up that they were
- 7:57all wrong, right? Every business started making
- 8:00their websites. And now the same thing happened
- 8:03with social media. Gary Vee was the first one or
- 8:06one of the first ones to go really big on social media
- 8:09platforms. Every account of businesses
- 8:13should be on Facebook. Every business should be on
- 8:16Facebook. Every business should be on
- 8:17Instagram. Why?
- 8:18Because that's where everyone's spending their time.
- 8:20That's where the attention is. And so right now, what this
- 8:24article is talking about is how if 84% of humanity hasn't even
- 8:29touched AI, the opportunity that we have as the early birds is
- 8:35just so freaking high because it allows us to go ahead and serve
- 8:40that 84%. And that 84% are the people who
- 8:43probably won't even believe you if you show them AI.
- 8:46Let's not even talk about the other percentage at the bottom.
- 8:49The free users. Right.
- 8:50The guys who are using AI for free, they're probably just
- 8:53asking ChatGPT basic questions. Yeah, right.
- 8:56Maybe they have some medical questions.
- 8:57They're not actually knowing how to use this for their business.
- 9:01That's where we can come in. And we still haven't started
- 9:04seeing ads on ChatGPT that would dynamic like the moment we
- 9:08introduced ads to the Internet like Google ads, YouTube ads,
- 9:12Meta ads. The whole economy changed.
- 9:15The moment ChatGPT has announced that they are going to bring ads
- 9:18to their free tier, we can have enormous opportunities in the
- 9:23entire ecosystem. So the reason why we wanted to
- 9:27show you this article is because for most of us, we believe that
- 9:33the AI space or the technology space is too crowded for
- 9:38artists, for coders to come in and be able to provide value, to
- 9:43come in and actually do something new.
- 9:45But the truth is the contrary. So that's why we wanted to show
- 9:49you this image. There is another image.
- 9:51Father, before we go to the business, I'll have you go
- 9:53through that pride. Go down in the article.
- 9:57We're going to go down right there.
- 9:58Yes, look at this, US businesses using AI with any function of
- 10:05AI, right. So for any, sorry, let me
- 10:07rephrase that. For any of the functions in the
- 10:09business, if businesses are using AI, they'll be included in
- 10:14that report by Census Bureau, which is the government's report
- 10:17in the US and the number turns out to be 18%.
- 10:21So only 18% of businesses in the US are using AI for one or more
- 10:27of their functions. That's a tiny percentage.
- 10:31By the way, any function that could be literally rewriting
- 10:34emails. No, no, no.
- 10:36Even worse than that, that could be as basic as rewriting emails.
- 10:40If a business is using ChatGPT to rewrite their emails, they'll
- 10:44be included in that 80 percent, 18%.
- 10:48So what we're talking about 82% of businesses not even touching
- 10:55AI right now in the US, US is the first world country.
- 10:59We're not talking about all the other countries out there.
- 11:02So just think about the opportunities that are
- 11:04available. You wanted to talk about the
- 11:07business side? Yeah, at the end of the article,
- 11:10it mentioned the businesses that you can build from this gap.
- 11:13And I think it's very interesting.
- 11:15And just like we discussed, it reminds me of the Internet and
- 11:19the social media era because immediately when social media
- 11:22became a thing, we had social media agency, then we had
- 11:25influencers, then we have social media ads, then we have agency
- 11:29who manage your ads. There would be a lot of
- 11:31businesses starting out because of this technology.
- 11:35I'm going to go through the list and I think few of them, what
- 11:37would be very interesting for the audience 1 is AI automation
- 11:42agency. This is the one that we talked
- 11:43about that you go to a company and you will look at their
- 11:46process and you will see how you can help them automate these
- 11:50processes using systems like anything Zapier or so on and so
- 11:53forth. Another one is AI implementation
- 11:56services. Imagine you go to a company and
- 11:58say, hey, let me set up your open cloud for you.
- 12:01Let me set up your cloud desktop so you can create apps.
- 12:05That would be, I think a very good business idea just to.
- 12:09Add on to that, I think what he's trying to say there is just
- 12:11creating a chatbot that responded responds to the
- 12:15questions that customers may have for the business.
- 12:18So using open claw and clod to ingest all of the data of the
- 12:23company and then create a simple chat bot where either the
- 12:27business owner himself, the teams, or the clients can use
- 12:32two requests for data. So information that they may
- 12:36need, because we're talking about some of the businesses
- 12:38that are traditional that have thousands of documents.
- 12:42Over like you go to over legal. Decades.
- 12:44Yeah. Over.
- 12:45Decades. You get all of those
- 12:47information. Yes.
- 12:48So over decades of data that is stored in the worst possible
- 12:54ways, you just feed all of that dirty data into your open clause
- 12:59setup or the clause setup and just turn that into a chatbot.
- 13:02And you can just talk to the data, which is why people love
- 13:05Notebook LM. Notebook LM is the simplest way
- 13:09you can imagine how that works. It's a Google product where you
- 13:12can feed it APDF and it turns APDF into a podcast, and you can
- 13:17ask it questions so that you can easily learn the material at
- 13:21hand. Now, the same thing is true
- 13:23here, but this is a much larger scale.
- 13:25Can you create that notebook LM internally?
- 13:28Yeah, internally that's the whole thing that we're talking
- 13:31about #2. The other, the other one is
- 13:34enterprise AI training and consulting.
- 13:37This is going to be huge. I think this would be one of the
- 13:40biggest pillars, especially consulting.
- 13:42A lot of business comes from the a lot of businesses come from a
- 13:46brick and mortar background. So they are very traditional.
- 13:49They people could go and do consulting, either provide the
- 13:53services themselves or they can connect them to the service
- 13:56provider for a fee. So there are a lot of money I
- 13:59think to be made here as well. Then we will have chatbot, we
- 14:03talked about it and they also talked about AI powered content
- 14:07operation. And when we talk about content,
- 14:10it doesn't have to necessarily be social media, video content.
- 14:13A lot of companies, they have to write articles, press releases,
- 14:17newsletters, that they have a team that handles all of this or
- 14:21they outsources to an agency, write an article for for, I
- 14:25don't know, for our annual dinner or for our annual product
- 14:28launch. They can bring everything in
- 14:31house and they can use all this system to help them.
- 14:34By the way, another one that falls under the AI powered
- 14:38content operations, perhaps we're not talking about the
- 14:42traditional companies here, I'm talking more about the creative
- 14:45companies. If you look at how a lot of the
- 14:48commercials are made now on Instagram specifically, you'll
- 14:52see that the behind the scenes has like an entire list of names
- 14:56of the director, the gaffers, the 3D artists, the 2D artists.
- 15:01And you can slowly see there's now a slot for AI artists.
- 15:05And the AI artists are going in there.
- 15:07And people, these directors are actually looking for AI artists
- 15:11who are able to come in and be that person, the team that can
- 15:15educate everyone else on how they can use AI.
- 15:18Is it on the creative side or? Are we talking on on the
- 15:20creative side? It can be on anything, of
- 15:22course. But since I've been experiencing
- 15:25the creative world, releasing new content with that list that
- 15:30usually just had 3D artist, 2D artist, writer, director,
- 15:34associate producer. These names have been there
- 15:37since a long time ago, but now you actually see one or two AI
- 15:40artist added. So if you go ahead and upskill
- 15:43yourself as a 3D artist, as a 2D artist, whatever kind of artist
- 15:47you are, or maybe you're just a color gradist, whatever your
- 15:51skill set is in the creative world, if you just add AI tools
- 15:55to your workflow, you can go ahead and help a lot of these
- 15:57companies learn how to use AI cuz this is more towards the non
- 16:01creative side these. Right.
- 16:03You can absolutely become an AI artist.
- 16:06Do the previous or not actually be in the actual production at
- 16:09the end of it and they will use some of the shots could be used
- 16:12exactly. Yeah, exactly.
- 16:14Or it doesn't have to be video all the time.
- 16:16It could be audio, it could be a lot of other aspects.
- 16:19Yes, I'm sure we're gonna see that more often in the future.
- 16:25And the last two are data readiness, data organization and
- 16:28AI readiness and workflow audit. So I would say this would go
- 16:32much more to companies that they have tons of spreadsheets and
- 16:36documents and data. Again, I always think about
- 16:39accounting firm, law firms that they have years and they have
- 16:43hundreds of thousands of customers.
- 16:45All these data are in a folder in in their storage room.
- 16:50So they multiple folders of folders.
- 16:53So they could, they could definitely, it was, it was a
- 16:55very interesting article and I think he was really spot on.
- 17:00Especially I love the way he showed how early we are at this
- 17:05stage. Let's go a little bit lower in
- 17:07the article just for people who are interested and want to know
- 17:11how to implement this. I'm not going to read the entire
- 17:14thing, I'm just going to read the first line of everything.
- 17:16How to actually start? Stop overthinking it.
- 17:18Here's the playbook. Pick one.
- 17:20Vertical law, dental, real estate, insurance.
- 17:23Learn their pain points. Learn their weaknesses, the
- 17:27problems, problems that they have.
- 17:29Sit with them and understand them.
- 17:32Build one thing that saves them a couple hours per week.
- 17:36Make the value visible. Build a dashboard, show them the
- 17:39numbers and then get a testimonial from that one
- 17:42client. Then go in and show that perhaps
- 17:45on social media. Shout about it, market yourself,
- 17:49and then using that same workflow, you can repeat it for
- 17:53other clients. This is perfect for anyone who
- 17:56is also just starting out, especially if you're younger and
- 17:59you're looking at something that you could pick up.
- 18:01You have the extra time. Maybe you're in university.
- 18:04I would have easily done this because you know, when you're
- 18:06younger, you're just so hungry. You'll do anything, and this for
- 18:10me is the golden opportunity. There's nothing better.
- 18:12There is no barrier to entry. You don't have to learn coding
- 18:17or you don't have to spend hours and hours learning a new skill.
- 18:21Basically, you are just going to study the industry that you want
- 18:24to get into. For example, you want to go into
- 18:27medical centers or clinics. You just have to go and see how
- 18:31they do their appointment booking.
- 18:32I think that's something that they're going to spend a lot of
- 18:34time on it. How we can automate it using an
- 18:37agent or using using a workflow and then there are thousands or
- 18:41like, yeah, hundreds of thousand clinic across the US that you
- 18:46can go to. And you don't even have to be
- 18:47smart about it. You can just ask them.
- 18:50If you're smart, you'll try to analyze what their problems are.
- 18:53But if you don't even want to spend that time, just go and ask
- 18:56them. Hey, what is one thing you wish
- 18:58that you could have solved in your business that you can't
- 19:01right now? What is one thing you wish was
- 19:03faster in your business? I can guarantee if you ask us
- 19:05that question, Farah can easily answer what the problems in the
- 19:09next are. So just go ahead and ask and you
- 19:12can figure it out. Very interesting article.
- 19:14Shout out to you Noah. We're going to move on to the
- 19:18next. Topic Yeah, let's talk about the
- 19:20new feature that was released on Friday.
- 19:23Clod Code Security, which is a new feature by Clod, can now
- 19:28scan code bases for vulnerabilities and suggest
- 19:32targeted software patches for human review.
- 19:35So this in a process of software development, if you have ever
- 19:39done it before, there is a part at the very end which we call it
- 19:42code review. We go through the code, we go
- 19:44through the structure, we look at what could be fix or what are
- 19:49the potential bugs. So this is something very
- 19:51interesting that Clod released and I think it's only available
- 19:55for team and enterprise. So we haven't been able to test
- 19:59it. But I'm really curious to test
- 20:01it on all the programs that we have created because all of the
- 20:04apps that we created internally in our studio is created by
- 20:07either codecs or Clod. So if you do the scan, let me
- 20:11see if they can find bugs from themselves.
- 20:13That's right. And what did it actually do to
- 20:15the market? That's very interesting.
- 20:18On Friday, the moment they announced it, Pratt, can you
- 20:20show the yes, all of the cybersecurity stock started
- 20:25depleting. We they wiped out $15 billion
- 20:30from the market and this is, I want to tie it back to what we
- 20:34have discussed initially. This is only market reaction.
- 20:38The funny thing is a company like Crowdstrike or Cloudflare
- 20:43the their services is not even similar to a feature of this
- 20:47software they are doing here. I have the detail here.
- 20:50So Cloudflare and Crowdstrike have no direct relation to Cloud
- 20:55Code security. Cloud Code security is pre
- 20:58deployment dev tool for static code scanning.
- 21:01Cloudflare is a runtime web security for things like DDoS or
- 21:06API protection on live traffic. Crowd Strike is for runtime
- 21:11endpoint detection and response, threat hunting and incident
- 21:15response on live systems. So basically what is happening
- 21:19now is all these AI companies are releasing features and then
- 21:23all the traders will think that this feature in the future
- 21:27because they are trading the stock for the future, right?
- 21:29They expect for the price to go up or to go down.
- 21:32They would think that this particular feature will kill
- 21:35this company and they're going to go sell that stock.
- 21:39That's why you would see that all the cybersecurity stocks are
- 21:43decreased, have decreased on Friday.
- 21:45Probably they're going to go back today, Monday, start of the
- 21:47week, the weekend was closed. But it's very funny, the
- 21:50psychology of people when we have all this news that are
- 21:53coming out and when you see it on stock, this is something that
- 21:57you can measure for us when you're looking at phone, of
- 22:00course we don't have a stock, we cannot measure it.
- 22:02But you would feel the same thing when you see motion
- 22:04designer artists are being replaced, artists are going to
- 22:07be replaced, Hollywood is cooked, all those things again
- 22:10and again. And I think the important
- 22:13message here is when you hear something, when a feature is
- 22:16released, wherever a company is announcing something, it takes
- 22:20time for the adaptability. It will take a lot of people to
- 22:23go through it and try it and test it.
- 22:25The impact wouldn't be immediate.
- 22:29So here's what I actually found out.
- 22:31I've been talking to Grok about this piece of news.
- 22:34See what Grok thinks about it, because Grok seems to be the
- 22:36least biased. And this is what I found.
- 22:40He called this the SAS Pocalypse, SAS Pocalypse, SAS
- 22:46Software Service, but these are software as a service.
- 22:49SAS apocalypse, play on words with apocalypse and the AI
- 22:53disruption panic in software. This is not brand new.
- 22:57Grog said that we've had this pattern long ago with things
- 23:02like GitHub Copilot when it was announced, when it helped you
- 23:05with the review of your code. But of course, the level and
- 23:08intensity and the speed is very new.
- 23:11This is a very 2020 five 2026 thing, which is what we covered
- 23:16in our previous episode, how fast AI is moving.
- 23:19So with investors, of course they sell first and then they
- 23:24ask the questions later. That's what's happened here.
- 23:27And I asked Grok if the sell off was accurate.
- 23:31Grok said no, the traders knee jerk was an overreaction to the
- 23:34headlines and most analysts called it incongruent.
- 23:38Then I asked Claude. I said OK cool.
- 23:40So for Cloud code security, this is for static code pre
- 23:43deployment and for Cloudflare in crowd strike that is runtime.
- 23:48That's when a system is live with, you know, thousands,
- 23:51potentially millions of users using the system.
- 23:54It's protecting it with all of the behavioral data in real
- 23:58time, very different things. But then I said OK, but can't
- 24:03Cloud Code release a runtime security tool sometime in the
- 24:07future? Said technically yes,
- 24:10realistically no chance in the next few months or even the next
- 24:13year or two because it's a massive long shot. 5% chance.
- 24:17That's what it gave. Because the amount of data you
- 24:21need to process in real time and the trust you need to have in
- 24:24the AI system for it to deal with real time systems needs to
- 24:28be so high that you're like OK cool, I trust one AI tool to do
- 24:33this. Yeah, and it's a totally
- 24:34different industry. It's it's something that I think
- 24:37everyone has to understand. With traders, they try to
- 24:39predict 5, 10 years into the future and of course in news
- 24:43like this always scares them, so they have to quickly move the
- 24:45money in the economy around so that they can essentially adapt
- 24:51to the situation. Now for you, what do you think
- 24:55will happen? Five years, I mean five years is
- 24:57even. No, five years is too.
- 24:59Actually, I really hate the five years look out anymore.
- 25:03It's not even. It doesn't work.
- 25:04Yeah, it doesn't work. It doesn't work.
- 25:06Anymore. But do you see a potential where
- 25:08claw code could release a runtime?
- 25:11But why would they? Because I think their business
- 25:14is very different. Their business is very
- 25:15different. They don't even have it in the
- 25:17road map. They don't.
- 25:18It's not. But couldn't, couldn't an AI
- 25:20agent or AI system be developed to handle of course eventually
- 25:25what crowd strike can of course, but you can say that for every
- 25:28single stock. Even not in cyber security,
- 25:31right? No, no, no.
- 25:32We talked about this. For things like Uber, you need
- 25:35the infrastructure of the cars. Crowdstrike.
- 25:37Also you need an infrastructure of the database and all because
- 25:40it takes a huge. But that's digital, so they
- 25:44can't. The AI can eventually take the
- 25:45database. No, you get my point.
- 25:48Yes, like with Uber, a business like.
- 25:50Uber, you need the infrastructure of the.
- 25:52Cars, yeah, And the people. So you've got all of that.
- 25:55You cannot. We talked about this.
- 25:57You cannot vibe code in Uber and say I just made Uber.
- 26:00You know you can, but there is no one to get the car from.
- 26:03Yes. You just connect with all the
- 26:04governments, go through all the laws.
- 26:07The same is true here, by the way, with Crowdstrike and
- 26:09Cloudflare, the licenses, the security certificates, all those
- 26:13things. But I think because it's purely
- 26:16digital, it is something that potentially in the future can be
- 26:23taken over by AII. Think if I'm not I have to check
- 26:25but I think cloud they have a lot of physical servers so it's
- 26:29not fully digital because they are routing.
- 26:31I mean they can just have the same for AI right?
- 26:34Yeah. You mean like AI to the servers?
- 26:36Yes, potentially. Why am I saying that?
- 26:38I think if we're thinking about threat detection security, you
- 26:44could have an AI in the future manage that and look over that
- 26:48oversight. But the trust has to be so I
- 26:53that's the only thing that how do you trust them versus AI?
- 26:58That's the question. I don't have my knowledge caps
- 27:01right there. So I don't know exactly what is
- 27:04the trust system in Cloudflare and Crowdstrike versus AI.
- 27:08Will there be a day where you want to actually trust an AI
- 27:12model or system to look over your runtime versus Cloudflare?
- 27:18Don't know if that's going to be the case.
- 27:19That's where a developer in the comments has to let us know.
- 27:22That's where my knowledge caps right there follows yes.
- 27:26What are we talking about next? This one is going to be
- 27:29interesting. This is requested by one of you.
- 27:32As we promised you guys, we look at the comments and we read
- 27:35them. This one specifically was about
- 27:37Microsoft Trellis two point O. This is state-of-the-art 3D
- 27:43asset generation. You give it an image and it
- 27:46turns it into a 3D model. This is useful in many different
- 27:51industries, many different cases and there are different
- 27:54competitors in the space, namely Hongyuan by Tencent, that is a
- 27:59Chinese model. Trellis 2 is by Microsoft and
- 28:04they released this a couple of weeks ago, so it's not entirely
- 28:07new. You have access to the model on
- 28:10hogging face, which you can easily sign up for, pay for
- 28:14credits, and use the model there.
- 28:16Alternatively, you can run the model locally on your computer,
- 28:20Pratt, I believe through Linux. Right through Linux.
- 28:24We've been trying to do that, but we've had some difficulty we
- 28:26have. Done but with one of the models
- 28:28previously, right? Was it Hunyan that you run it?
- 28:30On yeah, but that was the older model.
- 28:32That was the older model. And it's much easier to run
- 28:34these models on your computer now because you can use
- 28:37something like Open Claw to create all of the dependencies
- 28:41for you and run the system. Doing it manually was a pain in
- 28:44the ass, especially if you're not a developer.
- 28:47So right now you can do it using open claw.
- 28:49However, with Trellis 2, specifically Pride, you said you
- 28:54have to get certified by the group before you can get it on
- 28:57Linux. It's not like I cannot just get
- 28:59it in and start using. It it was the case with Honeyon.
- 29:02Yeah, that was the case with Honeyon.
- 29:04So there is no UI interface that I can.
- 29:05No, no, it's not that. You need to get approved by
- 29:08their team to use their models locally.
- 29:10Microsoft. Yeah, that's right.
- 29:12Yeah. So that's the only caveat.
- 29:13Microsoft, we are in the process of getting approved.
- 29:16OK, Yeah. Anybody from Microsoft?
- 29:18No, but we have. It from hugging face.
- 29:20On hugging face you can use it. No, but I wanna use it locally.
- 29:23We have all these Dell towers here, we have to use them. 96
- 29:26gigabytes of VRAM, no, on each of them, so 1-2.
- 29:30And then we have these two. Yeah, Microsoft approved that
- 29:33thing. We have over 200 gigabytes of
- 29:36VRAM just from our Dell computers, so we have enough to
- 29:39run this locally and run it high.
- 29:41High. This is the challenge.
- 29:43So on Hugging Face, you want to run it high, it's going to cost
- 29:45you a bomb. You prefer to use it, you want
- 29:48to use it locally. So Pratt, we have trellis
- 29:51running on Hugging Face. Can we go ahead and give it a
- 29:53try and show people what it's like on Hugging Face?
- 29:57So what do we have here, Pratt? So basically for the input image
- 30:01I'm going to use Kratos from God of War just as a benchmark this.
- 30:05Is from who RAF yes, RAL or RAF RAF RAF credits to you.
- 30:10This is a gorgeous Kratos model and we see it on art station.
- 30:15So if you guys want to check him out, this is his profile.
- 30:18So we just wanted to do a quick test and see how well does
- 30:21Trellis work with characters, if you give it characters that are
- 30:25incredibly intricate in their design.
- 30:28We have the soft body of the human, of course, and we have
- 30:31hard surfaces like the metal armor and the axe.
- 30:35One of the other things that is really interesting is previously
- 30:38when you generated 3D models, you had to be very careful
- 30:41because they did not look at things like specularity or
- 30:45roughness. You just have everything as
- 30:48super rough. But with some of the recent
- 30:50updates, these models have introduced shininess,
- 30:52specularity, and roughness. So do we actually get to see
- 30:55that here? So this is Trellis.
- 30:58This is Trellis, Yes, Pratt. What are the settings that we're
- 31:01using? The current ones. 300,000
- 31:04decimation rate for trellis and for honey on is 300,000 face
- 31:08number. Also we are testing, we are
- 31:10comparing this. Kind of on an equal benchmark.
- 31:13OK, so Trellis has a lot more options.
- 31:16So if you open up advanced at the bottom, you can go and
- 31:19choose at every phase of the creation what are the number of
- 31:23samples and. Wow, it's ready.
- 31:25Yes, it is ready. So just one thing before we
- 31:27start, you can crank up the samples to the highest and get a
- 31:30much better quality. If you're doing it locally and
- 31:32you don't want to pay, we don't have that access right now, so
- 31:36we're going to go ahead. That was a preview.
- 31:38Use a single image, right? A single image, both of them,
- 31:40Both of them. OK, what we just saw was a
- 31:44preview, am I right? That was a preview, yeah.
- 31:46The preview allows you to confirm the model before you go
- 31:48out and generate it, so I'm really curious to see what the
- 31:53textures are gonna look like. We have fur on the code, right?
- 31:57We have fur. Yeah, we have.
- 31:58Fur on the code, yes. Yeah, we got some fur.
- 32:01It actually has a lot of detail. Yeah, we have the axe.
- 32:05We've got the belt, we've got the ABS.
- 32:07I'm thinking where does Microsoft gets all their 3D
- 32:10data? Of course, Tencent Hunion is for
- 32:12Tencent, right? So all those games that they
- 32:14have developed, definitely. On G mobile.
- 32:17Yeah, they have a shit ton of library, Microsoft.
- 32:20Does Microsoft have access to the 3D files of Xbox games that
- 32:26they can scan? Let's find out.
- 32:29That's a very good question. Right Where?
- 32:31Do you? Is it done?
- 32:33Do you? That's who you are, that's
- 32:36Trellis. Or who you are?
- 32:36That's Trellis. Yeah, Trellis.
- 32:39Can you zoom in a little bit more?
- 32:43That's not bad. OK.
- 32:46And again for of course not for your hero asset, but if you need
- 32:50him to be in your background protecting the castle.
- 32:54That could be like a figurine as a toy and the side of the room
- 32:58depending on what you're using for.
- 33:00This is one of the hardest models you can.
- 33:01Give on you guys gave it one of the hardest thing to do.
- 33:04Which is good, we want to try to break things.
- 33:07OK. And this one isn't done as well.
- 33:09Honey on is also done. Bro, you're so.
- 33:12Fast. Can you zoom in a little bit
- 33:14more? We would.
- 33:15Generate image with the speed last year.
- 33:18But can you zoom into the whole window?
- 33:20Not this window, I mean control outside.
- 33:23Yes, yes, yes, thank you. And then let's zoom in a little
- 33:26bit on the clothes. I want to see the clothes, the
- 33:28detail of the clothes. More, more, more.
- 33:31Zoom in, zoom in. Oh, it has shininess.
- 33:33Do you see that? It recognized the metallic
- 33:36areas. This is not possible.
- 33:39Just a couple of options. Trellis did a better job in the
- 33:41face part. Trellis.
- 33:43Overall it did a better job because if you look over here,
- 33:45it's pretty much flat. Can we look at Trellis?
- 33:48Trellis you have? Oh yes, 100.
- 33:51Percent. Oh yeah.
- 33:52Trellis is 100 times. I just looked at it from the
- 33:54top. Look at the fur.
- 33:55Yeah, go to the back. Look at the fur, zoom in.
- 34:00Wow, no Trellis for the win. Within less than a minute.
- 34:06What is the cost of using Trellis online If you want to
- 34:09use it I. Mean I just paid like $9 for
- 34:11hugging space so I can use like any model.
- 34:14Up here, what did you do? Pay $9.
- 34:17That goes to our $20.00 a month. So add $9 to that $20.
- 34:21You're one of the top users of our top users.
- 34:23We definitely go to that box. Pratt, I believe you have this
- 34:26in Blender as well. Can we pull it up there, see
- 34:28what it looks like in the 3D software?
- 34:31I just want to see Pratt has made it cinematic for us.
- 34:36Bro, it looks, yeah, it looks totally fine.
- 34:40So can we say Trellis is better than Hunion?
- 34:43Yeah, it is. It is from this one test.
- 34:45But look at the axe, Pratt, look at the axe.
- 34:47Can you just Yeah, look at the axe.
- 34:51The shininess. I would not be able to use this
- 34:55as the main character. Of course that is impossible.
- 34:57But that's not the promise right at this stage.
- 35:00No, but this is a promising improvement over the past
- 35:05versions. Let's go close to the face and
- 35:07then I want to check out the clothes as well.
- 35:10Because you're telling me there is no 3D software that
- 35:12integrates a 3D model generation within the software.
- 35:16Not yet. They have ethical issues.
- 35:19Why they need to? Microsoft doesn't have ethical
- 35:22issues. No, they do, but they need to
- 35:25prove it. I really think this year we are
- 35:28going to have mark my words. This year 2026, one of the 3D
- 35:32softwares will have built in, maybe integrated maybe with API
- 35:373D model generation, at least for the background assets, Mark
- 35:41my words. I think you're right.
- 35:43I mean, we did talk about a unity thing right there, right?
- 35:45But. No, but Unity is for code, it's
- 35:46not 3D model generation. No, it's 3D model generation.
- 35:48No, no, it was for code. 3D model.
- 35:50Generation also for 3D models. When you generate worlds, it
- 35:53comes with 3D models it gave. Me fact like Jensen attending
- 35:55the AI Summit, I. Have to check double check it.
- 35:58Pratt. What do you think about this as
- 35:59a 3D artist? To be honest, it's really
- 36:02impressive. Like I hate making background
- 36:05assets that I know I wouldn't even see on the camera.
- 36:08And if I can use this, Oh my God.
- 36:11Yeah, no, I agree with you. I if you go closer to his
- 36:15shoulders, I just want to pay attention to this part.
- 36:17Like look at this, it has holes on the neck.
- 36:21Check the neck. So it does mess up.
- 36:23I mean, it's not a perfect. Model like guys, you give a very
- 36:26difficult character for it to try.
- 36:28Imagine your background asset, our table chair cups, mugs, the
- 36:33things that you use in the background, right?
- 36:34You are not going to use this in the background asset.
- 36:36But even he did a great job for this too.
- 36:38Look at the biceps. Yeah, look at the shoulders and
- 36:41the biceps and the triceps. The fact that this God of War
- 36:44creators model was created by AI in less than a minute is insane.
- 36:51How far as you're right, yes. So it does come with yes, yes.
- 36:54Unity is also going to integrate.
- 36:56OK, that's good. OK, so I checked this out.
- 37:04Trellis, developed by Microsoft Research in collaboration with a
- 37:08university, is a large scale AI model for generating high
- 37:12quality 3D assets. Based on available details, it
- 37:15was trained on the Trellis 500,000 data set, a curated
- 37:19collection of about 500,000 diverse 3D objects sourced from
- 37:23public repositories like Objectiverse XL, Amazon Berkeley
- 37:29Objects 3D Future, HSSD, and Toys 4K, with filtering for
- 37:34aesthetic quality to ensure versatile high fidelity outputs.
- 37:38I have not heard of a single one of those libraries.
- 37:41Look at that wood, ceramic, fur, plastic hair.
- 37:45Yeah. So this is the promise and we
- 37:47got. It the really cool thing that
- 37:49I'm really impressed about is it can do glass.
- 37:53Or it can actually no, no, no, go back, go back, go back to
- 37:57that the yeah. Wow.
- 38:01This is created by AI. Man, that's interesting.
- 38:03Which one is glass? Yeah, this one.
- 38:04Look at the water. Yeah, yeah, yeah.
- 38:09See what? Are you serious?
- 38:15This is my Microsoft trellis. I don't think any AI has done
- 38:20Glass before. OK can I be honest though?
- 38:23Go closer. How come the quality?
- 38:26Oh, we should try this locally I'm afraid, because.
- 38:29Our number is happening. You know, yeah, because look at,
- 38:32look at the edges of the cart. It's perfect.
- 38:37This is perfect. You can have you should get the
- 38:39foreground asset. We should get the approval, run
- 38:42it locally, get it some of the assets that we already built and
- 38:45see how it. Works.
- 38:46Yeah. And if you guys want us to do a
- 38:48comparison of honey on Trellis and other 3D generation assets,
- 38:53let us know and we'll do it on the show.
- 38:55We'll compare them live. We'll compare them live.
- 38:58OK. These are freaking amazing.
- 39:00Of course, the number of samples were high on this, but look at
- 39:02the material. Can you zoom in or no?
- 39:05I don't fry. No, I don't.
- 39:06Think I can? No.
- 39:08That's sad. That's very interesting.
- 39:11Let me see. Pineapple.
- 39:13Yeah. No.
- 39:14Imagine there's 3D assets. At the back?
- 39:16Oh, at the back, yes, that is detailed actually.
- 39:20Let me tell you one of the best ways you can use this.
- 39:23You're running an LED volume for a show, TV series, or even a
- 39:30film, and you need a lot of background assets.
- 39:32Traditionally, you have to either model them, buy them, or
- 39:36go outside and do photogrammetry.
- 39:38Imagine you're sitting behind the computer because we've been
- 39:40to an LED volume here in Dubai, right?
- 39:43You usually have about 3 to 4 people in a medium sized LED
- 39:46volume who are constantly updating the assets depending on
- 39:50the director's decisions. Imagine he's like, I need three
- 39:53more pineapples in the corner. Boom, dude.
- 39:56Imagine you have one of these connected and you just write
- 39:59pineapple. Or you give it a Google image
- 40:01and the pineapple's created. You drop it in the.
- 40:03Background. That is the promise of AI,
- 40:04right? Make you more creative so you
- 40:07can focus on the creative work. That's insane.
- 40:10OK, Very cool. Amazing.
- 40:11Very, very, very cool. Far as I think we are reaching
- 40:15the end of this episode. That's right.
- 40:17We talked about some interesting stuff.
- 40:20I really like the first article. I do recommend that you guys go
- 40:23ahead and check it out for yourself.
- 40:25And yes, it is a Monday. That means we'll be coming back
- 40:28to you guys on Wednesday for another episode.
- 40:31Guys, don't forget to subscribe on YouTube and if you are
- 40:34watching us on X or Spotify, make sure to follow us there as
- 40:37well. If there's anything you'd like
- 40:39for us to cover in the Wednesday episode that you'd find
- 40:42interesting, let us know and we'll take a look at it.
- 40:44Thank you so much for watching and listening to this episode.
- 40:48We'll see you guys on the next one.
- 40:50Until then, Ciao.