[REDACTED] 2024 | Venice and the Fight for User-Owned AI

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:00

   great to be here what an honor to follow that spirited conversation about the the stayfree ethos and I think in many ways this conversation is flows naturally from that my name is Jeff wilser I'm the host of the AI curious podcast and of course I joined by Tina Baker Taylor co-founder of Venice AI Tina great to speak with you oh Jee it's great to see you too no pressure right tough fact to follow so on that note I would love I think for

2. 0:33

   our session to to be most productive would love to kind of move from poetry to Pros we just heard a a romp through the in a way the intellectual history of of self- sovereignty going back to John Stewart Mills right so it'll be tough to outdo the philosophy there I'm not going to bring us through a deart rouso Plato session so instead I'd love to kind of focus more on okay that's those are

3. 1:02

   incredible ideals what does it actually mean in the pros of decentralized AI and privacy so starting from that point talk a bit about about your foundational principle of privacy as I understand it venice's Cornerstone ethos is privacy talk about kind of on a pragmatic level almost why that's not just a platitude but urgence right now yeah so I think um um there was a there was a study that came out um maybe

4. 1:33

   10 days ago two weeks ago from the St Louis fed which is weird um and Harvard and looking at the use of generative Ai and uh the the uses are just explosive basically it is the fastest adopted technology over um personal Computing over the Internet um than anything that we've seen to date and so we know that you know chat GPT came on the scene a couple of years ago it was very accessible people started using it um and it has very quickly become part of a

5. 2:07

   lot of people's lives more more than I expected people are using it for work people are using it for personal reasons and what people don't necessarily understand is every prompt every question you ask chat GPT every image you create um all of that information is held by open Ai and it is sh sh shared on their servers stored on their servers and the response that you get back from the AI is also monitored by them there's a whole host of safety features that they put on top of their models to prevent you from getting

6. 2:41

   information that somebody within their organization deems you shouldn't be thinking about or you shouldn't be asking about so they're heavily censored but most importantly if a government agency just like Edward was talking about came to them and said give me all of Jeff's information for the last six months they'll hand it over over because every question Jeff has ever asked is tied to his identity forever so when you

7. 3:06

   ask Venice a question I don't see it nobody in my organization sees it um the inference is sent to a decentralized GPU somewhere that doesn't know who you are and it sends the response back to you via proxy service so nothing about the question or you persists anywhere and it never ever will so if the government comes to ask me I can't provide them any of Jeff's information because I never had it in the first place um and so that creates a lot of challenges tactically um how do you build a product how do you

8. 3:39

   you know provide customer support how do you ensure that you're able to even verify your paid users like there there are a lot of things to navigate when you're using only decentralized infrastructure only open source models um and a privacy infrastructure that basically relies on only using the user browser so that's the only place this information is ever stored but when you

9. 4:04

   think about the questions that you put into chat GPT for example um would you want that to be attached to your identity forever would you want it sold to advertisers would you want it handed over to a government I mean maybe some of the things are innocuous but for example um I had a a a welln CEO of a company that everyone in this room has heard of say to me they had a heart condition they had had a serious incident um a year later they had a followup and they had to have a whole bunch of tasks they saw the doctor gave

10. 4:36

    him this report and we all know what it's like when we go to the doctor and they're very quick right you don't get the time to ask all the questions that you want to ask and you don't always understand exactly what they've said and so because Venice is private and he has this assurance that this information isn't going to go anywhere else he uploaded the entire PDF of all of the reports of his um tests and he asked Venice to tell him what questions he should dig

11. 5:02

    further into with his doctor were there holistic treatments that he could be looking into um on his follow-up visit you know what should he raise again what could he be doing and take you know his own health into his own hands um and he said it was just extraordinary and the next time he went to his doctor he felt far more informed he asked all these extra questions um and he really felt like he had a really better sense of his

12. 5:27

    own kind of Health Care and the ability to take charge of it and he would not have felt comfortable uploading that information we're saving lives here this is great so that and uh to your point Tina on on top of how sensitive this data is today I I would Piggyback in your point I would argue it'll become even more so tomorrow two years from now I I think that virtually everyone in

13. 5:51

    this conference on this live stream in in this nation EV eventually most people will have some kind of AI companion that could act as a chief of staff or a therapist right we might be sharing things like oh I'm not sure what to do about my partner I'm not sure which person on my team to fire we we'll be sharing the most intimate details of our life who's going to own that data how can that data be weaponized monetized I

14. 6:19

    think this will become only more and more urgent as Time Marches forward well if you think about it too like there is the school of thought that we are already giving all of this information to social media for example and so we have become a bit desensitized to sharing information our location our payment information who our friends are um but that information that we share typically especially on things like

15. 6:42

    Instagram for example is what we want people to see about us what we want people to be influenced by how we want to be seen the conversations often times that we're having with AI and certainly in a in a character type environment or or a type environment are what we really think not what we want people to think but these are very intimate and they're very personal and so I always say to people if you wouldn't want your journal or your diary floating around on the internet why are you plugging all this

16. 7:14

    information into chat GPT because that's exactly what you're doing and more importantly we can't see it today but I do think it's going to happen very rapidly in two or three years all of this data that we've put into AI um is going to know more information about us as as Humanity then we know about ourselves and if you extract that down distill it to a very personal level I mean it is possible that your AI companion will know you better than you know yourself and so to me that feels like something worth preserving and

17. 7:47

    protecting let's now get to the how that happens you touched upon earlier uh decentralized governance and how it all fits together talk a little more about how blockchain fits into this and then what are the challenges to truly build this at scale because I imagine it's a lot trickier to do this than collocation of chips right there are many advantages to build things a centralized way so what are the biggest pain points you have to actually make a competitive AI products that does this in this

18. 8:18

    decentralized fashion yeah so um my my co-founder our founder is Eric Vorhees so if anybody um is familiar with Eric you will have a sense of the philosophy with which we started um so the the first kind of Revelation from Eric was that these tools were really cool he was using them um and he did have these concerns about kind of sovereignty open AI um and and preserving privacy um but he thought it would be really hard to build um and he's not an AI person and

19. 8:53

    neither am I and so we thought well you know this is this is going to be too hard um and when you look at how much money you know open AI or anthropic has spent on building their their kind of blackbox proprietary models like you know we weren't prepared to do that but it turns out7 trillion doll no one do that right but it turns out that the open-source Community has been creating

20. 9:17

    models open source models um and they were becoming increasingly more competitive and we've seen over even the last couple of months um certainly like meta put out their 405b model well they put about three models but the the latest open source models that meta put out basically are more competitive um or equally um performance as chat GPT open AI anthropic Etc so the open-source models have become more competitive and so if you have access to those models

21. 9:48

    then you're able to basically build a front end right so anybody can run an open source you can go on hugging face today find a model that you can run on your laptop or potentially your phone very small you know models out there um and so anybody can do this and so if you want truly truly private AI then I encourage you to do that right um but if you're not prepared to run those gpus

22. 10:12

    yourselves we are an excellent alternative um and probably the closest that you're going to get um to something that is you know completely private um so uh we use open source models we use decentralized infrastructure um which then um reduces any type of like the the permissioning that you have in in a centralized ecosystem um one of our providers is AOS so if anybody's familiar with with AOS which is a a

23. 10:38

    great example of we were talking backstage around what role will tokens play um in you know the AI ecosystem and certainly they are contributing incredibly to the AI economy so I pay for my inference um and my nodes um via AOS tokens um so shout out to Greg um building an incredible business and so the the decentralized nature of that is then all of those contributors can be using unused compute they can put them online you can bid for them um and

24. 11:12

    unlike having a centralized infrastructure you need to be really on top of what's going on right you if you let the meter run dry on your servers they will go offline right there's no backups so um that's that's another kind of critical element to that and then I think um looking you know into the future we're starting to see um announcements around smart agents and being able to have you know independent AIS that undertake activities for us and I think that is going to be another great example um of where tokenomics is

25. 11:45

    really important and that that's not to say that again we we had a debate backstage you know could an AI agent not connect to your City bank card well potentially but it would sure be a whole lot easier if it was just trans trans acting on near eth so Tina um no conversation about this topic will be complete without a question I'm sure you hate and get all time but for things that are truly open source and

26. 12:10

    censorship resistance there's a risk there there a trade-off of okay what if someone uses it to build nuclear bomb to create biot Terror weapons how do you and you and Eric are very thoughtful people about this stuff how do you think about that how do you grapple this issue that tension that tradeoff yeah so I mean the we we do not have a board of directors um or a regulator that we are having to um report to that that tells us that we need to put all of these

27. 12:41

    safety features on and and truthfully um there there are no regulations now demanding that people put these safety features in and so that is you know a a cultural judgment call um we believe that people should have access to as close to Raw machine learning as possible um and that all technology can be used for you know nefarious or um colorful uh purposes um and you know we have terms and conditions that say that you can't use our service to you know generate information or or um you know

28. 13:22

    things that could hurt or harm people um we have an image generator we don't um censor that so it does create not safe for work images if you are a paying subscriber um and there will be people that have an issue with that right but again we're not here to kind of police um AI uh we don't think it's our job um we don't think anybody should um and that people will act responsibly and for those that don't then then yeah I mean you can go to the internet and and ask

29. 13:59

    it how to make a bomb and it will tell you if you're diligent enough right so we know that people ask those questions especially when you say anytime you say oh our platform is uncensored you know there's a handful of questions that everybody always asks and I would just say come on people rise above like ask build ask a real intelligent question right like this is incredible technology that can expand your intelligence that

30. 14:23

    can help you you write or or or debug incredible code I mean the ability to create and build with this technology is extraordinary and so my guess is that most people will use it for that last question for you real quick Tina what is the most important variable getting back the theme of our of our session the title The Fight for user owned AI what will it take to win that fight is it

31. 14:49

    about getting the tech right is it about cultural awareness is it about a political struggle what in your mind is a single variable we need a crack to win that fight yeah so we built Venice with the intention of being able to serve a mass Market audience with a simple to use interface anybody in the room could go on your phone or your laptop right now just type in venice. a you'll get a screen you can type in a prompt um and

32. 15:15

    it'll take you into the platform and you can create images and and talk to AI um it is very user friendly the reason that we have tried to create as easy a service to use as possible is that we want people to start as they intend to continue and it's a slippery slippery slope if you start using all of these other services that are not protecting your privacy and are indeed censoring the information and you think you're talking to a machine and you're actually talking to someone's Safety Committee

33. 15:46

    like the ability to change how you think and what you believe is very possible so we want people to have sovereignty of thought sovereignty of intelligence and do that all in a private space um the tech is here right so but it is a concerted choice on your part to choose that um and so I think that for me it it feels like if we go back to the beginning of social media if you don't like the way it's turned out um then start as you intend to continue today and use the tools and services that will

34. 16:19

    protect your privacy and won't Nanny you into thinking what they want you to think great not to end on Tina thanks so much thank you all