Welcome to Easter to Eg. Okay, our our last session of the morning, then everyone can go grab some food. Uh we have Brits uh when AI whispers. Thank you very much. Thank you.
So yes, this is the the purgatory which is the slot just before lunch. Thank you everyone for you know withstanding the temptation to go grab lunch. Um, what I promise you, what I'm going to do as a reward is not to try and sell you a million. I'm just going to share some thoughts about AI in general. And maybe at the last bit, I'll just spout some technical jargon.
What potentially is Nillian? Um, I knew I was at the right place this morning because when I entered, I saw the little sticker saying privacy is normal. And these are our core ten tenants. Privacy is the default and sharing data is a choice, right? So, I'm knits.
I'm the end boss at million for product. Um, as anyone in product should be doing, you should be working with AI or you should at least have strong opinions around AI, right? Uh, I've been working in AI for about seven years actively. Um, 95% of the time I'm I'm super enthusiastic. Everything that AI will bring us, all the the progress that we're seeing, 5% of the time, however, I'm on more on the dark side or I wake up on the dark side.
I'm in in the pit of AI depression. Uh this presentation was made when I was chilling in that pit. Right. So this is going to be bit of a depressing presentation. Anyway, when AI whispers from assistants to manipulators.
So we all know this, right? All those all those big headlines saying AI may replace 300 million jobs. AI will all will steal our jobs, right? 41% of companies in in the world will uh cut jobs by 2030. Uh most of the people in this room probably don't fear AI too much, but just think about the the broader economy and everybody else outside, right?
Because the reality the reality is that a bunch of content creation uh markets are basically the the bottom of the market is being cut off, right? If you're average, if you're below average, if you're a below average designer, a below average content creator, you're a below average developer, your job is in jeopardy because of AI. So, you need to be above that line. And it's not just content creators, it's also white color jobs like legal, strategy, my job, you know, white collar jobs in general. Um, luckily however and this is this is a little bit old.
Um, everybody well you might not know him but this is Gary Kasparov. He he famously lost to IBM DB Blue way way back. Uh, what what was he good at? What made him a grandmaster? He was very good at tactics seeing immediately on the board like these are the schemas.
This is the next move. I understand this pattern. So short decision-m but he combined that with long-term strategy knowing where he wanted to get to right what I found interesting is that he kept playing and he kept playing in this against uh these hydra teams where normal people do the strategy and then AI would do the short-term tactics and he was beat by a janitor somewhere like a normal person using AI for the tactics but the person himself he was just thinking about the overall strategy so they were able to defeat Gary Right. And the Hydra model, yeah, that's that's probably the model we'll all end up using if AI is going to progress at this pace where where you will use an AI assistant in in a way to be able to keep above that curve. What makes that interesting, and this is a depressing part, one second.
We're in uncharted UX territory because this relationship we can have with this type of tool, with this type of assistant is going to be way more intimate than anything else, right? It's it's not going to be the same like the relationship you have with Photoshop or you know window or or office or or whatever because you already see it in the mannerisms how people are talking to their AI assistant. They're using please and thank you. They're using personification. Um the assistant also has recent memory of everything, right?
So it's not a tool but it's more of a partnership. And I feel that this where it becomes a little bit scary because all those foundational models if you use an a tool like openai like claw like everything else they are phoning home right it's they are using at the back end gemini um llama well not llama in particular but groi openai everything else they're basically using your data and they will use it for something and why I think they will use it for something is that there's been a ton of investments in AI like like I think the last number is 700 billion or something. The ROI is not not that great just yet, right? So, it's it's all future forecasting and hoping it will make money at some point and investors on on the whole are not that patient. So, at some point investors will want to see return on their investment.
grabbing grabbing an example out of uh history again in 2011 Google said don't be evil they were all helping they had this nice browser nice search engine they weren't doing anything malicious they were just helping us out right come today don't be evil unless it makes you a lot of money right that's where they they're at right now and if you look at their uh current and this is 2022 their quarterly income statement um 56 billion of the 69 billion that they're making comes out of ads, targeted ads, customer profiling, commumul commumulative customer profiling and whatnot. And I fear this is where the evolution of AI is going to go to as well. We are we were at the stage that AI is helpful, right? Um I feel like we are are at the stage where I think you might have seen it that OpenAI just added shopping to JPT where it will be very easy to from your uh client be able to shop shop directly. So I think I thought we were there.
I think the next thing is going to be product placement that that somebody at the back end will pay open AAI or somebody to just make sure that the product that you're researching gets put on top and then you can go over the rails, right? Is is if you talk to it to uh in that intimate fashion, you can get political bias influenced and and hold out personalization reinforcing bias and maybe change your opinion about certain topics in an entirely different way. funnily enough yesterday um after I created this presentation OpenAI uh has a new CEO for applications. It's a person at Instacart. She worked for 10 years as the product lead ad meta on ads just and so there's signals out there.
Um, and that's just if if they want to target ads at you and then try to change your, you know, change your mind. There's another one that I found a little bit depressing thinking about it is um from Bertoff, right? We we're studying, we're doing time investment in in different to gather experiences and get all these skills. Um, and this all of this constitutes us as a working person. This is the value that we build up during our lifetime and and I'm now 48.
So at this point I've got 48 of time investment and this is my worth. This is what I can bring to a company. So this is the value I can create for a company moving forward as well. But if I share everything with an AI, right? The worst well the best thing that can happen is they'll just use it to create an AI agent that uses my type of correlations and thinking etc etc and monetize that to somewhere else.
The worst case scenario is of course is it will take my job right and you might think that might not happen but it always starts with content creation. And if you look at content creation actors are already at risk. We we've got digital twins. uh they're using voice actors etc etc. So it always starts with content creation and then it just um goes into other territories.
So what can we do to prevent the a the AI apocalypse? And this is where I'm going to do a little bit of million selling just a little bit right. Um so what you could do is instead of just storing everything in the cloud and at Google etc etc is create mouthful decentralized encrypted permission scalable storage. That's what we're doing. You can also next to it have a decentralized secure arbitrary compute platform.
If they communicate with each other then you actually have the central compute unit. It's completely secure. You can do stuff on also run AI. So that is our that's our main component and obviously well no no no tech uh drawing cannot have input and output right so put those there as well um but this whole system of course has a trust model that that has a trust me bro vibe right as in trust me it's secure so what we're adding is a immutable layer posting on the blockchain that has trust verification um through the whole end to end call. And this whole thing is multi-user.
Multiple users can use the blind computer. And we also allow for agents. So that's it. If you're interested in checking us out, dogs.nillion.
com and I'm within time. Thank you. There any questions or commiserations as in we're all going to hell? That's fine as well. Uh I have a question.
Sure. So I I agree with a lot of the sort of worries of and potential avenues that AI could go down. What are you using dayto-day in place of things like chat GPT? Yeah, so I'm still using chat GPT. I'm sorry to say um but but we are and this is what we're working well that is what I'm f focusing on at million is I'm trying to create expert systems that are running in a TE for instance we moved away from uh public uh meet noting meeting note takers and we created our own using the whisper model in a TE and then using a summarization engine within a TE and you can talk to all the meetings that I've had and I have this attestation that this was done in a privacy preserving way and it's secure So, so I'm trying I'm what I'm personally trying is not to rely on foundational models that just have all the answers, but then try to choose expert models that I can train and I have control over and are running in a privacy preserving way.
Okay, nice. That makes sense. Anybody else? Yes, sure. I'm just curious what I I really like your presentation.
I mostly agree. Um, but I'm very curious what what do you what does Nillian do now daytoday for their clients? Right. So for our clients, we're we're already running uh ex uh expert models in TES in combination with with yeah the the nerdy tech jargon that I try to keep as short as possible. So we are doing this already.
Yeah. So you can store stuff in a decentralized way using secret sharing. Uh so it can only be reconstructed uh in one spot and then we have very opinionated communication between our decentralized TE environment and that decentralized storage. All right, I'm going to look it up. Great.
Great. Thank you. All right, thank you very much. It's uh lunch. Thank you.
Lunch time.
Automatic transcript — names and jargon may be misspelled.