Unlocking Value in Open Source AI - Berkeley 2024

[Music] thank you everyone uh so I think this is one the panel I'm most excited about today you know so first of part of the panel want to talk about the industry Evolution you know like first of all what does mean to be like open source AI mean is it open data open weights open model and how the open source AI Evolution has been in the past 3 to five years so you know when people talk about uh about an open model or an open source project in AI there are at least three I think orthogonal Dimensions I could be talking about you hit it there with uh whether the source artifact itself is open source or the weights are open where the data set's open I actually think all eight settings of those bits mean different things and have different set have different strategies um and they all are meaningful I don't think any of them are actually just foolish um one of the ones that I think is relatively underexplored is uh is actually open data uh open source closed weights basically you reserve the right to serve the the Primo version of it the fully trained version of it um I think that'd be a really interesting thing to see people make a stronger go out than has happened so far so M has a great background like it's like from open source Community how you have perceived like with mongodb how open source principal has been driven a lot of the AI Innovation here yeah so um open source the principles of Open Source as well as the ethos of Open Source are qued mongodb's product um and our product philosophy a lot of it's around composability um and making sure that we're not overly opinionated in how a developer uses our product uh that being said you know people think our license is controversial a lot of developers start using mongodb locally um and that ability to download open source mongodb and build what you want has driven a lot of innovation and continues to and so that sort of innovation um is really driven by the fact that we we have an open source offering how you think about monetization commercialization strategy Eric what's your experience you know with MB you're the city investor in Minds to be how's their Journey being involved there it's interesting at this point in the development of AI and and the maturity of the AI Market there are certain considerations like right now for us that are really important and others we see will become important a little bit down the road everyone's terrified of putting these at Large Scale in production because of obvious concerns around security you know promp poisoning and stuff and then then also you know how's this thing making its decisions we're going to get sued right um and so what we found right now given the state of the market you have a lot of large organizations that are finding their way in the dark and the AI stack every week kind of changes like the the best architecture for this and that like you know um uh it develops really really quickly and somebody works in a big you know it director in a big uh insurance company has no idea how to keep Pace with this and so the monetization for us right now is really around um support and services and and especially Consulting the more mature um uh monetization model is really going to be around dual licensing you know open source is nice for now but if you really want to connect this to all of your internal systems you want security you want user provisioning all that stuff um you know you it's not the Community Edition you want the Enterprise Edition and that's kind of down the road maybe for like funders um what type of mode they should build over time as well because I open source you know what's the best way to sing about like building that regard well there um I mean especially in there's some hard problems that aren't easy to solve and if you can raise money and kind of race ahead and and and find a certain amount of maturity uh my CB a lot of it is open source but how you scale it and we have some kind of close source and like really like hardcore research projects around you know well we do some fine-tuning type of stuff much like f pabas i mean there are kind of some secrets in there that we've hired a lot of physicists uh to solve those problems and nobody's going to catch up anytime quickly so we've kind of differentiated through the technology when we put together this open source AI investor surprising is the hardest one like what kind of like you know our strategic to support those f for those programs and it's kind of the same methodology I've always used you just lock a couple Founders in a room and you you have you just dig into the problems they're trying to solve you kind of go through their patent filings you ask them lot of questions and after about two hours you're like yeah these are the guys so I think it's the investment process for me has been just really similar to the way it's always been and I think it's all about asking people difficult questions about the decisions they're going to make and then sharing stories of what you have learned but not necessarily telling them to follow your path and one of the things that we've found with sort of anybody that's trying to serve a model to customers right now is that you know they're more sophisticated multicloud users than like companies that are 10 or 100 times their size because on day Zero having a kind of secure supply of compute and being ready to handle some spike in load whether it's because you want to train more or you want to or you actually have demand um you can't just assume that gcp will have more A1 100s for you tomorrow because they just might not so I mean we're we're you know meeting people who are two guys in their dog in an apartment in San Francisco and they're you know kind of buying 490s for cash in parking lots and setting them up in their in their basement you know um just like it was sort of 1995 or something again which I think is exciting and interesting but it means there's a weird premium again to kind of can these people really get a big pile of computers to do something strange and surprising in a pinch um which has you know not been sort of something people have been actively assessing at at early stage lately which is interesting not every single founder they have the money to buy those crazy like cluster to support their program are we seeing a different type of Entrepreneurship going involved from this like type of a new paradigm I think we'll obviously see business model Innovation um I'm not smart enough to tell you what it's going to be unfortunately VCS who traditionally Provide Capital and some other resources now all of a sudden they have gpus to offer um which is something that I haven't seen in the past it during this time where it is constrained I do assume that we'll see some business model Innovation yeah so I think like you know thank thank all of you for this amazing panel I think 100% this a new paradigm like facing both funer and also for investor as well like we're collaterally thinking this problem like very like proactively and at the same time like for us for for Ed we're launching our open source campus this year to really supporting a lot of the open source AI funders just to a saler your project build your community and working with this amazing like Venture investor to help you grow your journey as well