The following video transcript is generated automatically by a computer algorithm that learns and gets better on a daily basis. Please accept our apologies if some content below doesn’t make sense:

[00:00:00] So moving on to the next topic, something which I came across the other day on Twitter that was absolutely blew me away. I’ll be honest. So there is this artificial intelligence language model which has been doing the rounds in recent months called g.P T-3. I’m not sure what GPE Stansel to be on this. But in essence, it’s created by an organisation called Open Eye, which is a non a non-profit organisation doing research into AI and machine learning and things like that. And essentially, it was a giant language corpus created from a huge amount of stuff on the Web. So basically they pointed LOEs and in those sort of machine learning computing power at a whole load of content that they scraped from the web. And they basically set it the job of sort of learning English or learning language by using the input of all of this sort of written word that they found on the Web. But not just written word, just like how gob of stuff that they found from the Web. And the reason why this was originally interesting was because they use 10 times as much input data as they had done on previous models, with the idea being that the more data you throw at something, the more in machine learning, the more interesting stuff kind of like floats to the top. So this has been around for a while now and it’s it’s been used in some interesting ways. So you may well have seen on line there are a few these sites which do I storytelling. And basically you can go along to these sites and you can just type in the start of a story. And then this language model, this GPE three language model, will hallucinate the next section of the story, which is something that’s been around in computer games. [00:02:03][123.4]

[00:02:04] It’s like the very early days of tech space computer games. I’m not sure if people are old enough to remember any of that be used against this type in go left, go right, pick up the thing, do whatever. And it was used to really frustrate you because the computer always come back saying update on stand up and it’s that don’t stand there. Whereas these ones just completely hallucinate this very compelling sort of scenario that it feels like you’re talking to another human or that there’s an author on the other end which is producing this output anyway. That’s all very interesting. But what was really interesting on Twitter the other day, we’ll put the link to it on the show next. But a guy basically took that giant. Model. And on top of it. Created a little bit of a UI and the question he also it was basically what did it learn from the Web? Because we they chucked a whole load of data through the stuff that they just pulled off the web. So with a little bit of extra tweaking. What else does it inherently know? Just by the fact that it’s crunch through all this kind of like information and. What he sort of pieced together was that he pieced together this amazing little demonstration where essentially you you have a little textbooks and you type into the textbooks, your description of a Web page or of you are a user interface. So you and he’s got a great little video and is on his tweet showing and doing this. And it’s he’s doing things like typing in create a page with a giant red headline that says X, Y, Z and a blue subscribe button. And this language model basically creates code in a in a language called J.S. X, which is kind of like a layout code, similar in style to shemale or something like that. And then just creates this output code, which then actually just shows you, you know, this code actually is well-formed, is not buggy or broken in any way. And it actually shows like what this. Language model has created just from a single textual description of what it is that you want to display. And he sits there and he just types in various things. And pretty quickly, on the order of five to 10 seconds. This thing is output’s new versions of a UI just entirely based on just a few was of description. [00:04:19][135.2]

[00:04:20] Now, the reason why this sort of jumped out at me was I that was like, wow, wow. Because it was never taught to do that. [00:04:28][7.8]

[00:04:29] It’s just something it picks up. Bye bye. Getting all this information off of the web. But the second thing that was, well, you know, this is just some guy who has just tried a little experiment now with the bit of extra elbow grease and a little bit of extra effort, you could really see like a whole variety of different business opportunities falling out. So there’s already quite a big movement in sort of programming or app creation in general called the no code movement, which essentially is people coming out with a whole suite of different tools and apps which allow our non programmers to create sort of fairly sophisticated apps. So you can put together things like, so, for example, these things are Air Table, which is kind of a cross between a spreadsheet and a database, which allows you to sort of create and plumb together some quite complicated sort of database type things without having to know a line of ASCII well or do any sort of database programming of any sort. There are plenty of other sort of companies things out there doing that. And this feels like the next step on from that, which is like. You have this sort of little II co-pilot who’s sitting there in the cockpit next year while you are maybe you’re a designer and trying to so like design an interface for an app or something about that. And you could be there and, you know, in figure or sketch, your Adobe illustrator or any of these programmes, you can have this little sort of I co-pilot that kids or plug in, which would essentially hallucinate whole well-formed, well-thought-out out kind of like you eyes from very little input from someone who doesn’t know I can’t design worth a fake. So if I can go to an AI co-pilot and say, hey. I want this thing and describe it in very sort of general terms, and this thing pumps out version one of this thing. You can then yourself like tweak it, you know, just move things around, change the colours a bit and away you go. Then that just speeds up a whole section of sort of asset creation. You know, you could imagine it doing the same thing for like Facebook ads. So a classic thing. With online advertising is you want to try lots of lots of different creative to see what sticks. And I’ll be tested against each other. And what are the bottlenecks there apart from the cash to do? It is actually coming up with all these variations in the first place. So having something like this applied to it, you could say, OK, here’s the raw materials. Here are a few photos. Here’s a few bits, a copy of whatever. And this thing like Kelly Sunnites, thousand different variations of have created from nothing. So I think there’s loads and loads of options here. And it’s the first time we’re really where something sort of A.I. has sort of, you know, demonstrated any form of actual intelligence. Excuse me. As far as I could tell, I. Now, you’ve looked at this briefly, Marseillais. What what did you think? [00:07:31][182.5]

[00:07:32] Yeah, what’s chicken, actually. So our recent comments. So the counterargument will be, from my technical point of view, that they just threw a lot of computer computing power to that two to a problem or to that project. They argue, let’s say on here, VentureBeat, that the computer they team up with Microsoft on the computer to process this kind of information, which had one hundred and seventy five billion parameters. So is Uruguay a system with a word through kind of 50 joyed of memory and untaught which cost? Or I mean, millions, tens of millions. [00:08:19][46.9]

[00:08:20] And they say 12 million in compute credits, which these guys have opened. I can do with Microsoft what they said is done. So the underlying technology is the same. So a neural nets that were used before, even for lightning, probably Google Translate or other platforms, they just did it bigger. So it’s more impressive, the results, but it’s not moving the world forward. They need you in the research of how to. I would say probably there is a race towards general A.I. or jow artificial intelligence. So on. And so there was a leap forward with then your net deep mind. We’d go out to go on, son. [00:09:13][53.3]

[00:09:13] So but since then there have been all these incremental advances. And I think I, I agree with you and you probably you feel it more closely because you are a programmer, so it will make it easier for you. And also give give gifts Europe. [00:09:35][21.6]

[00:09:36] It’ll put me out of a job. [00:09:36][0.8]

[00:09:38] Yeah. So by your intrapreneur. So you may try to find a way to get into that lather. But yeah, ultimately if it builds tools and what they’re trying to achieve now is with the eye as well, these two to help code that codes for itself basically. So. So that programming could be outdated, which makes sense anyway. If it’s Samael to make it, as you said, it will speed up a lot the efforts to get to the next generation of tools which are more advanced and all that combined go going towards an ultimate solution that that’s more powerful. [00:10:20][42.0]

[00:10:21] So, yeah, I’m I’m impressed on the progress of A.I.. I’m I’m usually a bit more reserved on their claims. Many others are putting forward for their singularity. And, you know, I’m personally a big fan of Elon Musk as well. I don’t take what he says for granted or not. Nothing like that. I think he’s perceptional one of a kind. But, yes, he’s still a person that can say silly things or could be wrong. [00:11:00][38.9]

[00:11:01] So. So but he has a point on on that this trend will keep growing up to a singularity. [00:11:11][10.7]

[00:11:12] I’m not so sure that is going to be in 20 years, 30 years. But at some point, if you allow enough time. Yeah. This artificial intelligence or machine intelligence will surpass human intelligence. So, of course, you need to find what’s intelligence on and in which sense it supposes. But if. Yeah, essentially it I would say if if if you if in any domain that you can imagine a computer or that computer competing with a human, it’s a pass as a human. I’m vastly southerlies that passes by. Bye bye. But by an age it’s got to be a material or six like exponential. [00:11:56][43.4]

[00:11:56] So. Yeah. [00:11:58][1.3]

[00:12:01] Yeah we will. We’ll get there. But it’s it’s going to take time. Probably way more time than people thinks. Probably hundreds of years. Well we we could be all done by then. Yeah. Yeah. On Pandemic. Sound like I gave it to an authority. Yeah. Climate change. Whatever that shit that hits us next. So yeah, I’m more reserved. I was, I would love to merge with machines and I loved all of them. I’m, I’m, I’m not that technical like you on like a front for my programming point of view, but I built the companies. I’m early adopter of technology as well. So eLong was targeting Joe Rogan on the symbiosis with machines, I think. Yeah. Yeah. That’s that’s that’s great. I think I would definitely go for it. Yeah. Yeah. Fold without hesitating too much. But again, even with that I think we are many years away from it. And also we will know when it’s coming. So that is going to happen overnight. This is looking more like an incremental improvement of things. I don’t think it’s exponential as its claim. Yeah. Certainly there has been progress, but usually we are. Yeah, we are not that good at thinking long term. [00:13:33][92.0]

[00:13:34] Yeah. When it’s gone it’s gonna happen. [00:13:37][2.5]

[00:13:39] Yeah. Bit like nuclear fusion. It’s always been 20, 30 years away. [00:13:42][3.3]

[00:13:42] But what was it. Yeah. I can go like other examples as well. Like it is so different but. Well Peter Thiel, you know this well known investor or entrepreneur from the PayPal mafia, they call them the founders of PayPal. They went on then to build great things. He was the early investor in Facebook. Now he’s listing volunteer and so on, which is very controversial, but good for him. And he’s always this contri and guy putting forward like different ideas. So, again, I’m not saying he’s right, but it’s good to come out like contriving opinion because on this, because everyone you will always hear almost inevitably everyone, all media outlets as well, saying, oh, my God, technologies like growing exponentially. And he puts the example that actually probably is not. [00:14:42][59.4]

[00:14:42] And we are stagnated so that that only on them on the field of bits like computers, that the growth has been exponential. And he attributes that mainly because of their business model love like software companies, you have an incremental cost to build a larger audience. [00:15:05][22.4]

[00:15:05] So usually if you sell cars, you come out cost like cost of producing that car. So if you sell more cars, then you will grow. But it’s not exponential because your costs would grow attached. To do that, you can improve profit margins. So it’s not. But and then whereas in software you if you have a cost of hundred for one hundred customers, but then you can have a cost of hundred fifty for 10 million maybe. So that is like extreme example. But the point is that with little incremental cost you can deliver to a wider audience, increase sales. And that’s why they became so popular and lots of investments went there and developed a lot. But in their world of atoms, as he put it off like things, we actually went backwards in many, many of them. Like he puts some of the examples from top of mind may say so. He said in the 60s we were. The moon 70s, we were sending rockets everywhere. And then, yeah, we thought, OK, then we’ll go to Mars. But instead, NASA had their space shuttle orbiting Earth. So not very exciting. And in 2011, they just landed. So, so, so, so now that they were until now. Which Space X is sending some some some astronauts to a space station in U.S. soil. But until since 2011, NASA was hiring rocket launches from Russia. So they didn’t even have the capability to send a rocket despite our people to the space station. So so it clearly went backwards and then other examples. So you are saying as well. They promised that in the year 2000 we will have flying cars and instead we got 140 characters as originally written and a bottle some day. The human genome. So it was discovered in 2000. And it was, you know, the holy grail of the next wave of many scene discoveries in biotech, nanotech and so on for curing cancer, all sorts of other diseases. But 20 years since then, and that has been very little progress in the cure of cancer or any major other disease, really. So. Yes. So it’s I don’t know, I think with a A.I., there is hype still in their world of bits. But we should be careful long where we think this will go, especially linking to a previous topic we’re talking about on the education system. So don’t just think this will keep growing forever, exponentially. I think eLong would say as well that don’t assume things will. Things won’t improve by their own or on their own. I just need people doing a lot of effort and committing resources over a long period of time to really make it happen. So, yeah, things won’t improve on their own. So, yeah, that the the fallacy on the exponential technological progress. I mean, there is progress. I’m not sure if it’s exponential and maybe in some fields even just this is something very basic example. But like the iPhone. I mean, one what one was the first one lunch like 10, 2007, maybe 2007. Yeah, seven. So 13 years since then. And it’s pretty much the same piece of hardware. I mean. Yeah. A little more memory and camera here and there. But he has not improved much. Of course. Yeah. I would like to see those companies doing what they can do. Definitely way more. And if they are not, I mean they’re driven by profits, though, quarterly, short term profits. So for them that they it’s a cash cow. They just keep producing the same marginal increments. Just more people will buy it because too many people don’t have it. Or what? Well. [00:19:33][267.9]

[00:19:36] But they yeah, they could be doing more. I would say I’m Kudo’s for open A.I. is great. I know as well. [00:19:45][9.4]

[00:19:46] Elon Musk was involved at the beginning and he just step step down or just left. [00:19:53][7.0]

[00:19:54] He was not part of the team, but was kind of invest or some direct or something like that. So he argues that the insurance and vision. I wonder what that meant. But yeah, I’m still to be more impressed on them on the path towards artificial intelligence. I think brawly, for example, the approach on your link. I think it’s it’s way different and more like a radical way of thinking. And probably that could be a light, certainly a shortcut to Janet. I or at least on paper makes more sense to me. I don’t know if it’s gonna work at all. But if it if it does work to some extent, like increase the bandwidth where humans or the human brain can connect to a computer, that then that’s a completely different story. So probably, yeah, more research. [00:20:54][60.7]

[00:20:55] So neural nets are you know, they are one thing you can grow the potential of neural nets exponentially beyond throwing or doing the bigger data sample. So with more like machine power and this and that, but rather it’s the path to a I would rather be more like a combination of stuff. Yeah. So, yeah, I agree to some extent with a counter argument, but I’m still impressed you get plaids here. [00:21:29][34.5]

[00:21:29] So to summarise, kids think twice before you sign up for. Yeah, nothing grand Harvat Zankel and secondly, sex robots. They’re coming. I can feel it. Thanks everyone for listening. We’ll be back next week with some more nuggets of knowledge. In the meantime, please do cheque out our YouTube channel. That’s where we post this and our other podcasts. You can go to YouTube and search for networkers or you can find the link in the show notes if you’re interested in a deeper dive in all things entrepreneurial, including more detailed information. Help mentorship courses. Cheque out our Web site. And that is a network as DOT code. Right. See you next hour, guys. [00:22:08][38.1]

[00:22:08] Thank you to. [00:22:08][0.0]


Share This