AI, Humanity, and What's Next ft. Andrea Isoni

Can AI innovation outpace our ability to regulate it? In this episode of AI Rebels, Andrea Isoni, Chief AI Officer at AI Technology, dives deep into the rapid evolution of artificial intelligence, the complexities of synthetic data, and the tricky legal landscape surrounding AI-generated content. From the surprising global landscape of data centers to the shifting market demands post-2023, Andrea shares sharp insights into how smaller AI models are challenging industry giants. Plus, discover why compliance and AI safety might be the hottest trends in tech—and what investors need to know to stay ahead.

Transcript

0:18

hello and welcome everybody to another episode of the AI Rebels podcast my name is Spencer and I'm Jacob we're very excited to have Andrea with us Andrea you do so much in the NFL as I've was looking into your background you're the chief AI officer at AI Technology who have written machine learning for the web which is very popular you have your hand in so many different things I thought we could start Andre giving us you know who you are a quick summary of who you are what you where you spending your time mostly these days and we'll we'll start there and we'll see where it takes us yeah sure first of all thank you for having me straight that quick correction yeah he was popular the work was popular I mean every every brother that's uh you know are starting an end he still sell like 50 copy per year or something and just to clarify it was said if way more but because Masha Lenny evolved over a time right and uh that that's a that's just to jump in our thing I don't know if you're gonna talk about that but that's one of the problem when AI move that fast whatever gonna have write as a book today uh it's not gonna be relevant to money the moment is going to be published and bring to this already old this is kind of thing now I wasn't at that moment when I wrote it I think it was 2,016 but hey things got you know all pretty quickly and therefore it lasted for a few year very well and as a testament how quickly we are moving definitely it's still relevant in the sense that whoever want let's say you're a software developer if I was designing tender for that reason so if you're a software developer and you want to become a machine learning engineer or the the scientist they will give you the initial tool to understand the mentality where you need to train where you need to learn the data pipeline our training the model which is still relevant today it just doesn't have all the new huge new thing to be discovered because there's so much it just partially relevant and that makes sense yeah that's an interesting thing we haven't talked much yeah sorry go ahead yeah I was gonna say that's an interesting problem you're highlighting with the book where it's moving in AI or anything in AI that's the problem that said apart from this correction for was important who am I I'm a doctor science well I'm academic translate into a data scientist and then I'm into kind of a business person to still have their hands on sometime therefore what I do in a daily basis is talking to client understanding the problem and structuring a project and a team to to solve it usually we we develop a software record using AI and the IP goes to apply IP intellectual property goes to apply that's what I do in a daily basis awesome uh so how long has have you been doing that gig well business I should say yeah so I started in 2,017 since 2,017 I kinda do the same job as it now I'm talking to different kind instruction different project and then I manage a team to deliver so what would you say is kind of the biggest difference between what projects people are looking to accomplish now versus in you know 2017 2018 back before you know the rise of of really large language models and yeah wave of gen AI that we have now okay true distinction the way we work didn't change too much as because at the end of the day we are a consultancy our goal is to keep screening for the best legal AI solution we get up whoever is familiar we get up just to mention our one you know families rifle and and then find out mm that is interesting for this product let's to try it out something like that right that kind of mentality never goes away yeah so the way we work entirely hopefully is creative is not tell me because sometimes you give for granted things for granted for me but not necessarily granted for the public so the way we work is the same now what change it definitely is the maturity of both our clients and maturity of the it infrastructure which means certain project we no need to implement everything we can just integrate with ready made solution of things like that the easiest example is sometime you need really to train uh not necessarily large but definitely I'm older now yes you can use an IPI with a ready made model just find tune it with that that person comes from there I had to add to that variety including securities at the cost long term cost may not be as efficient as in you do your own model and you maintain yourself pros and cons yeah liability is discerned because the virus is either yourself or into the vent I mean there are many the job be changed now you can use a P I integration to the a P I which we do with certain project you cannot do if you want an IP obviously right because yeah an intellectual pro leading a project in drone technologies to the test certain things and control it dramatically you cannot do it not even today automatically for example well certifics but no specific things therefore you need to specialise more than blah blah the battery that occurs way different uh now client when you talk about it I understand you well what I cannot cannot do they roughly understand what you're talking about before was I said pitch black I mean it's not an offense it just take time to understand what you do yeah what happens and stuff like that yeah yeah and I think that's you know representative of the population at large as well where AI has always been this SCI fi term for people where they they didn't really understand is this black box idea artificial intelligence that can do all these things and now it's kind of becoming more solidified I feel like for yeah for the mask consumer I said on top of them many people remember even in 18 or 19 they say the AI is actually doesn't really work they were kind of right what I mean by that they were trying some commercial chat but they were just couldn't find I mean they have just a root of a graph of things it was very centralised it didn't actually work they kind of were right but they didn't understand that there there is a way to make it work it just very expensive a very sty consuming you cannot do with already made a ipads were too like there was the chartboard back in the day in 1819 which by the way was an aide back then so now the aide catch up with the reality let's let's put it that way mmm hmm and therefore these less doubt in the market somehow yeah yeah okay that's awesome I know and this is kind of related where I know you've been working a lot on the state of state of AI right that report that you put together that big comprehensive high level overview it's not that big that's spacious I think that's yeah no I mean it covers a lot a lot in it I'm curious what for you personally where you've helped you know put that all together what's been one of the a few of the big takeaways from 2024 for you for me what there are think they are I didn't like too much because what is interesting for me usually is not really interesting for the public and usually what's interesting for me is think they are yet not sober I give an exam it's so it's completely not interested to the public usually I most bravo are not interested in fact so I discovered something that I still cannot make my mind around but you asked me I reply but I'm not sure how many visualisation we have on that quickly and then I move for something that probably more interesting to the public but for me was is very interesting the moment as still it's a riddle for me is you know data centers right okay okay how many how many data center do you think there are in us Rafi then I say the number I give you the number let's let's go quickly no 500 no about 5,000 now 5,000 no that's the that's the center are equal different side different side different out of the different futures blah okay fine let's do 5,000 now who is the second who would you say is the China China Belgium okay I would say China too it's not but it already is a surprise right how many do you think who is the second weather I I tell you the second how many do you think yes the second uh 2000 oh they say the same thing now the second is actually second and third second third and fourth are I don't not remembering which order but let's say is uh I think is second is Germany third is UK fourth is China with about 500 third time less now I mean there's something that I still don't understand about this how gay is possible a country like China with a population way larger which by the way as a sophisticated tax sophisticated everyone paid with the phone more than us I don't know man you know they use a lot Internet and service and Status Center somehow yeah they do with a fraction of that yeah a fraction of I don't still understand this I still don't yeah I think for me is this is one of the thing important I confess I did a post on the last week not your visualisation but it's one of the posts that nobody care about doing is a very thirsty pig important to understand by the way since they're not there many so have an interest on that I think it's quite fundamental I would say yeah no that's fascinating that's I'm gonna go look that up afterwards cause now I'm wondering the same questions like yeah how are how are they doing that that's a lot fewer what I expected I I wrote the post in that give you oh it's too long give you a few things I don't have the solution I say that even if also say I don't have the solution but this this is for me one of the most interesting things I don't think that let's talk in let's rise the views for you guys because this I don't know why to me it should rise a lot especially from investor I mean if you're investor you should understand that right big question yeah no yeah that's huge I don't know even in the fact that nobody talks about it's strange to me yeah I don't know dude I don't want to get well yeah anyway well I'm curious then with this report what for someone who's not familiar with AI how it works what's going on what would you say are like the top one to three things people should know that happened in 2024 like the big big things in your mind that hinge points so in 2034 the most important kind of uh overall thing that happened is that has been proved and by the way this is one of the thing we got arrived from the Freedest report because that happened is so back one sentence so in 2023 there was this kind of mentality the largest is the better I mean quantity versus quality the more you are the you know billion of parameters to the trillions g t P4 to the trillions of parameter the better the more sophisticated and the more powerful the mother is this was 2023 now we are can in 2024 as we already saying 2023 cannot predict cannot plateauing the number of parameter is still rising I know massively there are really reason for that one of the major reason is you can rise a lot but if you don't have comparable amount of training data to do it you cannot rise to the infinity and not scale into infinity as the same level the data you see what I mean so that was a massive limitation still is somehow so you can all do it even if you want to on top of that there's marginal uh marginal training on that and therefore one other thing that happened is smaller model get very close or actually closer to the top performance available is I think is all free mini is one of the if not still the best slightly better there are one from DC Infamous 1 but still if you find if you see the benchmarks even smaller more when I say more ladies hopefully many who knows how many as but let's say as 5 500 billion to 1 trillion perimeters something like that or more something like that we don't know it's not disclosed I mean on 10% or so from the top performance yeah yeah model with 50 billion parameter a fraction there's a many many meaning for that there are many reason for that one of which is once you create a very very large molder they ask performance what you can do is this is called distillation so you take it out and you distille a compression and kind of a smaller molder there are many ways to do that by the way and therefore you have a smaller model which also means if you don't do the bigger model first you cannot do the smaller it's not like they say to the because one of they I have today is like oh deep sick I was so clever the other were kind of stupid doesn't work that way because without the biggest one you couldn't do those more one later I see I see what you're saying yeah no you don't understand this without them you couldn't do them there's no way we are out it's not that they were following something without no no this doesn't work yes it's more complex than that but just to give an idea yeah yeah so you've you've talked about synthetic data before and I felt like this was probably a good segwaying moment to talk about that Phil as we talk about scaling the models as we start getting these better and better models so for example chat G P t four and a/2 obviously was just announced today do you see the role of synthetic data growing in future training of models at this point hundred percent but the problem is now what not you see are more differently in true for years there was a there will be a problem of definition what is synthetic what I mean by that I mean if you prompted them as also legal definition I think one month ago or something there was a law court on someone that was using an imaginary generation 2 I don't remember which one and the court decided that whatever e generated he has no intellectual property yeah which means that is his dating is not him from a legal perspective right synthetic means it's not you it's not you man who owns it nobody ha ha it's for everyone yeah yeah so even from a legal perspective that is my even in practice system is differ because imagine right you start and you do a prom and then you have a prom yes you refine I would argue the more you if you're refining consistently with the apple walls they still add add the value from you right by blurring so what synthetic I was is not synthetic is a this is a big problem because one of the niche sector in the eye is growing the most or at least I think it will grow even further for various reasons if you want to we can talk about is what is called data labouring scale AI familiar with scale AI basically these companies have a large pool or what they call a trainers which basically mean there are people that teach a model something teach a bit of math teach a specific thing of accounting or certain to defence whatever it is for the clients yeah so they have a specialized they have a platform that allows user with some knowledge counting math whatever to train the models automatically yeah online whatever eggeneration data to to to the modelator this thing when this happening disease is syntated data in a way but from the human I don't know if very very soon what is will be so the first point for me of for everyone is what is synthetic and not synthetic at the moment what is considered synthetic is you print you do one prompt the output whatever the output is that is completely synthetic or you have a generation model what I mean by that is you define such statistics they say people are certainate etcetera then you add some randomization of variability deep random height of 2 foot 2 inches sorry 2 foot is a little bit much 2 inches of the person a randomly adjusted and they mm hmm and the mother can be even a role and model whatever the model is yeah yeah but there are many ways in in which an AI even a large language model or a large model for image etc becomes that synthetic data generation model that can be used in other in a yeah yeah and they still use it today they will be using in the future but at some point there's two things I figured that we are gonna there will be a very difficult problem to understand cause this is dating it was assume right right I see and why why is that so significant do you think to understand what first of all is the legal issue and that I made it and it's not trivial in term also of um liabilities but many court many or many lawsuit at the moment is a lot will depend on how this lawsuit will land up to it's funny because okay so one of the now I remember one of the I think this is an alt Altman that said so one of the way they deep seek train one of their model is somehow somewhere like there are many techniques but same basically they have an API from open AI so deep seek use the API open AI to basically prompt the model Allah so they can get specialised data they can use it there this is actually so basically they use syntated data from another model but the point is that model as infringement copyright allegedly nobody knows because there are almost real data for other coverage right right it's a disease for deep seek synthetic to open AI to open AI to the company is not synthetic right right I see yeah is your concern kind of that like you know generated content even will end up like so say you know say Open Eye AI loses one of their lawsuits your concern is that some of this generated data could then be considered infringing as well generate a French arrangement yeah ha ha oh it is a very complex a rule can be okay after the second step is synthetic anyway or I don't know if it's legally possible to say something needs to be a demonstration of adding value from the human but imagine generation okay I do up from fine but then after I not correct that I don't like the blue no I don't like the blue a little bit of it and then oh I am also refine myself with you know digital pencil whatever if there is an additional value from the human and that is legally defined I don't know can be illegal legally defined I know I'm legal asper I've argued that is still or supposed to be considered human generated yeah interesting so that is you see we are getting a very it's very where is it gonna land yeah we we we we are getting a very because no doubt is a question of contribution I think everyone will agree right both the human and the machine contribute probably also is a question of proportion right right is the proportionate 99% of the machine did every year you say yes and that it yeah but if you actually add something they very distinctive than should be the case this the human contribution was the main factor of the image or the text whatever that is hence is human yeah but then it brings the question like how is that gonna be enforced how's that gonna be that's a problem I don't know if it's legally possible to define oh no yeah I you know no it was because of me that this came to be when in reality that AI did 99% like but how do you it's gonna be messy it's gonna be very messy it is yes yeah it needs to be okay the image is unique image of the text the text is unique fine first of all now where this unique mess come from machine OU because if he's no unique nobody cares is fine but if he's unique his uniqueness come from the human or formula mash yeah so how would what would you define what's your definition of synthetic data synthetic data needs to be to me well it's something that is being created of the machine for the if not entirely or at least quote 90% of the contribution now how to define that if isn't italic if I create a statistical model and then I press the oh I use the statistical model a supposed to be to done to today right a you put in a large language model or whatever model is and then you press the button so to speak and create more prom more data etcetera that is in 30 no doubt about it uh unfortunate Brom is more complex if this model as underline eh usage of data there wasn't synthetic for stuff with which is money many case now the problem is most of the model we use today as in the most successful one layer layer and layer of other data that come from other stuff everyone is building on top to the others it's like saying the internet is built the built on top of the Talco infrastructure you cannot have the you cannot have this goal if you remove the copper line back in the 18 I'm sure some part of the world still has it it is the same thing the all every new model will come back will come in the future somehow somewhere there is still Wikipedia hardcore post on it somehow they come back to even in my book there was a large so every and everyone is there is a large a lot but are the core are the initial a still there somehow yeah yeah yeah that's fascinating well I'm sure this will become more and more of a problem I mean we're already seeing it with you know things like deep seek and all this these lawsuits that are happening but I feel like there hasn't been a definitive resolution to anything yet and no we'll see in the coming months what what happens what happens it's also difficult to get a final again remember when I started not even presenting myself say oh my book is already open it's already quite useless because we are going so fast it's been cancelled it's very difficult to find a definition when we move that fast yeah so you need to have it finished should the future proof which is a bit even better yeah it's difficult yeah it's not needs to be valued for three years and then oh no oh go we are already over it let's rephrase you need to finish these principle days enough then no matter how future prove no matter how advance and the perform will be in the future is the old it's difficult hahaha open ended definitions are a little bit paradoxical you're right yeah well okay this just brings more questions to mind but we'll uh we'll move on cause I don't know if there's a good answer yet Andre you this is kind of a different realm of what you what you do but I know you advise um I think it's Antler Venture Capital is that right I did that Antler now is what had sound the other uncle but I was advisor for Antler yes that's and and was Antler very focused on artificial intelligence investment okay got it well I'm curious then from that experience where I'm sure you saw so many different tech and AI companies we we have so many listeners and guests on who were working to start a company have you noticed any trends that successful companies have in the tech and AI space what what sets it AI or tech company apart from others very difficult in that also because trends you need to be a trend a trend doesn't last long I keep I can give you a trend that I think is going to be well is happening now so if you're fast enough guys you can do something with that AI safety real quick exactly AI safety but ISA 50 two l models there is a huge need for finding structure that is independent for the eye model for example during there are few they already done but there is still space if you if you move fast so for compliance as in so let's say you have a machabo can be everything really well Lachaba is these things to say you want to protect it from uh giving you know derogative output for example or even being ingest injected like jailbreak and stuff but input that then come back you know better should the the way to do it is to create a so the user doesn't interact directly with a model there's a layer in between it can be a model itself but it's not the main model you've looked this layer in between as the communication on like okay if you read derogative term or he has ask you something in a subject you shouldn't ask you don't go to the motor you stop before yeah same for obviously image of generation etcetera etcetera etcetera so and you need to somehow link that to the compliance now I I see you're in US of Canada kind of regulator right as far as I know I could be wrong but there is no AI regulation as such apart from few state like I think guidelines taxes or yeah California some but overall you don't you have it fair enough even in China you have it um a day of trust funny China has placeability of the output I can anyway so if you create structures and product and solution around the compliance of the AI this is a trend that is happening right now and that's that's something that people buy because if you find a good product a good price etcetera people will buy because they cannot use it and they cannot is is a bit silly to think that the same you know Vandal they give you open AI it will also do the security because there is a comfort or interest they can if they do one they can be just on the other you need an external part I see you soon so in your perspective what impact do you think regulations like the regulations that you have passed have on AI innovation do you think that it's the type of regulation that fosters innovation or do you think it's the type of innovation that limits it let me grab my cat off the counter while you're answering it's okay he jumped off okay cool and yes or no so the UAR Act is what is called a horizontal trigulation which means a regulation is across any by the way the UAR Act is the principle of UAR is not about business is to protect life and especially human life what it means is what is regulated what is going to be ever governance in if you are a company that does resemble education or healthcare so your AI assistant can you know impact heavily the life of the person because you you know suggesting medicine or your schooling their exam this is a huge impact of the life of the person right and you should be highly regulated that's the kind of the principle if you're talking to a chat bowl for customer service for your cattle broken in the kitchen you can survive it's okay small regulation for you hopefully though that's the kind of principle okay but again is the same for every industry horizontal okay there are others who don't regulate or and or like the UK as a vertical approach and or California as a vertical well as a guidelines now the issue with vertical approach or no guidelines the problem is they is kind of saying you don't need to regulate but any authority like the Financial Conduct Authority for anything related to fine fee and tech or payment or the stock whatever or the health organization authority any county they has one if they feel there is a need specific for AI they will add something their their own regulation so it's gonna become a vertical regulation because there are already industries they already regulated per se forget about AI insurance is very regulated of course healthcare is very regulated for production just to give you example today at weather is um operating this industry one of these industries already regulated by one of these authorities they need to follow a lot of compliance for production got it right now that means they have also mandate they they feel if you use a INL care needs to be regulated in some way or need an extra additional addendum on the governor you're already doing they can do it this is kind of a vertical approach yeah what is no clear to date is let's say in the future for whatever reason healthcare authorities say Ncon Dsi can be dangerous producing this kind of medicine so you need to add this the food uh I don't know the food of the what is called it food production authority or whatever it is they say in food production this is no well this can be dangerous you should add if you're using a assistant to you know scan the food before whatever you need to add this or gurlish all of a sudden these are two different authority do their own regulation different way all of a sudden you know the technology provider not only need needs to be compliant no one mind 2 if these two two two examples different 1 right or 3 or 4 so the advantage of having one cross industry body of law on AI then it become exactly an event make sense but only only if that will become the case yeah only if the vertical regulations will take if the vertical any authority somehow they will feel the need to do one at that point the good one with the cross and you prefer that you use approach to a horizontal cross industry regulation not me it will be in practice the the most efficient way because otherwise look at the mess every health authority will be there on and the companies to do different ones yeah and there could be conflicting instead of doing regulations yeah yeah and then you know they get mad you multiply the costs for our for our companies to be compliant hey you know what the Americans love our lawyers so well you know the only one but the problem is is that is not what is unclear today that's why there is ambiguity uncertainty unclear the unclear situation because it's no certain that every industry should have an addendum on their own regulation if they are regular regular regulation just for they are that's not clear and not clear even to me probably not but if it's yes yeah I would win yeah interesting so if you could can I give an example so it's clear one thing let's say that's everyone should understand this you buy a flight ticket go lie you check the price okay oh the price is 50 dollar you check okay I check this evening who knows when you got this evening can you say oh the price now is 65 dollar 15 you know dollar smell ey what do you usually think boys they are lying the nose by the cookies or whatever they aren't checking the fries they rise the price yeah and by the way this is illegal yeah you cannot before I no no airlines in the world cannot individualize or there's a word for anyway special they cannot target this individual they can target across a different site is illegal to do that one if they change the price they need to change forever it's illegal so what I'm trying to say is that when they rid of the flying authority there is a name of it right I'm sure aviation authority at 50 school they want to do something for AI well they have already said something that forbid the user just sophisticated the clones to target individuals so AI or not you cannot use it for stop so there's no need that although you implicative who say there is there is yeah that makes sense which might be a good approach if you were to regulate is have a more broad like deviation authority where they say like you can't do this because it hurts the individual allows you to take advantage and do these things that was before the eye yeah that's what I'm trying to learn to make you understand I mean way before the eye was already yeah no problem and now they just in fact in a way the new regulation especially U a I R are kind of um only form of what's that were like harmonizing yeah with previous regulation including deviation you cannot target people citizens individually at least okay Spencer is there anything else you wanted to ask right here before no go for it take a pivot okay um so you agree I think you're unique as far as the other guess we've had because I know that you speak regularly you travel and you you talk about AI to audiences I'm really curious as you've gone through this experience if you've noticed a cultural difference in different areas of the world where you've spoken on people's view of AI on their approach to AI what what are some maybe funny but also like interesting key differences you've seen as you've talked to different people okay I can give you the funny one because straight on the head and then we can go low on that so I as you say I talked in many different conferences around the world but one of the thing that obviously in Saudi Arabia I give a few talk down there and even in very very technical conferences like inside the Security Conference Center now all these a voices I mean real surprises in Saudi Arabia at least in the conference I went which were fairly big 50% let's say 40 40% of the audience were female in a cyber security conference the UK doesn't happen US dubbed it Europe Hundred percent doesn't it does happen in Saudi Arabia and interesting it's a bit strange for me to see it because you know I mean yeah uh yeah so you take a page out of their book to figure out how to how to make that happen they are interested because you should look I have a bachelor in physics and then PhD blah blah blah few women are interested in such a technical subject I don't think it's annoying offending anyone I believe it they just my experience is they are just not interested why why not it's not me to say but definitely they are not interested or less interested than men on technical subjects yeah that's at least what I could say now in Saudi Arabia apparently they are way more interesting typical sagitted or the counterpart in western society yeah luckily inter mode intermal audience that's that's what I I definitely I can say now in term of um other differences the one of the good thing of the tech word is is very very international I mean obviously I travel physically I see people that but again when a gitab is being released whatever the Gitab is everyone has access to that everyone everyone who is interested in to that has probably seen it and so in that perspective I don't see much of a difference in term of knowledge or quite your knowledge sometimes it's a bit different the mentality for example I met few Asian people that way the reason on such a specific topic is way different than us because they have different limitation than us we are born and raised at talking about general western society yeah to use whatever you can use and express yourself in whatever way you can you can easily feel the Asian culture that's not the case and that's reflect also when they talk about coding or you can do with coding or coding in general and that's also Japan by the way is not I'm not talking about you know China you know yeah politically different countries but in that way Japan it kind of more similar to us yeah in televipe political but it's not in culture that's the point yeah and that's where cause it is so interconnected but when you obviously a big a big issue in AI and tech with software and technology based on data sets is implicit biases right that get woven into these data sets that then are manifested later by the AI I know there's been a lot of different organizations out different struggles Google out a big one so it will be very interesting to see different models developed by different countries and then to compare them I think it'll be fascinating as more and more countries have you know like China I don't know if their big one is deep seek right now but like well something that's been really interesting to me is the number of just completely state of the art and uncensored video models coming out of China to my account there's been I think two or three in the last like six months that we're all you know interested at state of the art or near state of the art levels when they came out there's too much much less censored than any of the American hosted models just been fascinating to me honestly when it comes to visual arts I would have expected probably the opposite to be true but I guess you know that the laws around copyright regarding American copyrights are a lot looser than that probably has a part play yeah so that now you make me remember is a kind of a start starting different because another culture is the let's say the Arabic culture the middle is in a broad sense right but then religious speaking is difficult to to interact with image of people because it's kind of against their tradition and culture to give an example even if you want to use Tinder or things like that you don't show the face on Tinder disease yes showing the faith or even on the block this is easier yeah LinkedIn okay if you see people on linking from Saudi Arabia centre some of them quite a few they don't put the pace they put the initials or things like that because I noticed that yeah because it's against the tradition because the only one God is not represented nobody is as God so I know God nobody can do it this is kind of I simplify a lot of I know I'm Muslim scholar or any sort but hopefully I don't I'm not offending the but ha ha it's against the traditional culture to represent the face of a human person because it's basically saying you're close to go in which you are not that's the that's the kind of it's kind of an insult therefore I would suspect that this kind of culture will not do imagination yeah no because they can't yeah so that'll be interesting because they they you know it's not exactly culturally closer to them yeah yeah but then it'll be interesting to see what they they will do out of I feel yeah cause it'll I wonder if certain countries will thrive with certain types of models as time progresses like yeah maybe they wouldn't do image generation but maybe they'll thrive something on pros or something yeah pros or something like that it could be because writing is something that elaborate writing something like that they they are probably more specialized than us culturally I I would not be surprised in fact one of the come back to the prediction you want some predictions so for this year I'm expecting so one of the no only Dixie by the way there are other reason but one since we are talking about this week as well you know when I think it was Jesse Owen when days the hundred meter time you got below the barrier everyone goes over because it's kinda also psychological barrier kind of thing right you you heard this store kind of reasoning about it I think it's when he went off under the 10 sec was it 10 second remember I believe the 10 second barrier and then yeah similar story with like the four minute mile as well yeah exactly once you overcome the barrier she's definitely physical but also psychological many other believe it's possible they actually do it mm hmm so I think DC acted is not the only reason but you definitely acted like that when you show that other company they can do that there will be many others AI brought in that there are definitely I'm expecting that this year top 10 benchmark will come from other let's say emerging countries India is one great great country that it could not be India I would not even surprise Indonesia will do it Brazil someone is the mental the mentality has been broken other can do it so we could do it too yeah just the mentality has been broken I think it's pretty bad though there are other artifact right I want to go to it but yeah definitely this is one and I would be very surprised if one on American countries does at least reach top ten in some original yeah yeah that's at least what I want on my prediction yeah yeah I'm curious do you have any other as you put together the the report do you have any other predictions uh for 2025 that you think might happen oh yeah well there are many to said in fact they'll remember all of them by art but definitely if you're talking about artwork uh in video we still stay on top internal top performance but I suspect a meat market is going into opening up because okay if you want top performance and you don't care how much you cost you go to video it will it will still go for the video however I think it's going to emerge a series of companies that still want to train certain models they not necessarily you know won the top performers this a performance is a price in other words is opening up a meat market which doesn't exist today but it's going to open it up yeah and the meat market other player like that Andy or others will play apart so is kind of a differentiation of the market in this way right before you have just the top and then the market just diversified you did not know I have one the cover alright then you have the top kind of premium product then you have a meat market product and who knows maybe one day we have also uh but definitely a meat market kind of uh yeah yeah brother I'd be curious do you see uh I mean there's the the cancer processing units who I don't recall the company building them and then Google's video processing units do you see either of those kind of starting to fill larger roles within this meat market that you're speaking of or do you think that those are still both kind of a few years off from from play major role okay for disclosure okay I need to do this brand but sometime I getting love a little bit with the technology so no too rational there's one of them there if I don't remember the name which is weird but I really love it's busy down a mixture of blockchain coming back to my mind that does one of different way to processing unit come back in my mind to do the farmers and you know cheap designer the working that anyway yes or no what I mean by that DCL will probably be more technology why more standardized ways to produce chips but would not be surprising in few years from now we'll have new ways to produce chips they are well better Marcos the fatty etcetera but easier because to ramp up production etcetera so watch late very much but definitely hundred percent I believe there will be new ways to do all this kind of competition well Andrea as we as we're wrapping up here as we're wrapping up the episode is there we have a lot of listeners as I mention that are working in AI whether they're starting a company they're at a startup do you have any general advice to people to really thrive in the AI space I feel like you've had a longer experience than many in AI and you've thrived you've done really well in which position or job which what was in what was ideally the career they want to have God Spencer I was gonna say let's pretend that there's someone who is not an AI but wants to you know wants to play a role in the field whether that's cool yeah yep whatever that was okay okay fine it's not yet in the eye but I want to look okay first of all yeah I started from the basic in that case it's starting with the basic we will be surprised how many soft to develop as less people they already know they still unfortunate this problem at least that what I said is type fast I know it's something very trivial but learn how to type fast it's not too difficult and far and saying that but some of the my employees when I noticed that there are online course you can do and they make you do the you know the gaming or whatever and they make you type fasting few months yeah it's better even for your life it's not you know in company time in my case they get better in all their activities anyway so the first thing is type fast second obviously experimenting with all this um AI solution better even if you can I don't know what's your as being your path in life but hey I now applies to everything so if you can find a way to you know to to using your own job or whatever it is is obviously the way another suggestion I give it to you is a bit clanky but it's a good thing for you if you want to get into the either the fatty is clanky is what is called agentekii again to come back on the prediction just do that I should many have the the word agentekii is coming many said that yeah no it's not coming this year just to clarify that and the reason is because of this because the technology is clanky now this is a great opportunity for you since this clanky go in one of this platform and try to implement your own assistant or your calendar automation when you go on Telegram WhatsApp whatever you use you send a voice note automatically you can fill up your calendar with stuff try to do be able to do this yeah it's great advice Clanky great for you because if we wasn't it doesn't have value right no value for you so if you have no technical skills that's I would suggest be fast in typing this experimental that I can but do some adjectic you know of this just for fun and to understand how all that works and what are you know the you know the solution about how clanky is which is again is a good thing for you because yeah if he wasn't clanky once once you learn and nobody you know will care about no that's great advice to your point on the on the fast typing um I in high school I cheated my way through my typing course cause my friend and I discovered a bug in the system for you okay press I don't remember the series of keys but you press a certain series of keys and it marked all of the all of the assignments complete for you anyways and that that held me back the first year so of being a software developer just not being able to type fast um anyways just had to it's good advice reminding me of that had to get that story off um Andre thank you so much for for joining us and and sharing your thoughts on on AI and the AI space it was really fascinating to hear and it's it's awesome to hear from somebody who's been around in the space for so long and you know since since before the attention is all you need paper really and that's that's you know these days that's an eternity so leave what seven years but that's that's an eternity um maybe so yeah if if our viewers want to follow you see what you're up to what is the best way to do that uh look I'm very active on LinkedIn if I miss your message by the way I'll link green it's not gonna want to reply just been under the you know the list and politely yeah send it back I reply even we are no thank you by always reply so this is link green so if you wanna email is Andrea at AI technologies dot co and and yeah if you're interested in all the silly things I say there is also a linkling as well there is AI newsletter and I buy a human um yeah you can follow that and we'll drop links some as well for people to follow you thanks Andrea we'll stay in touch thank you thank you