[REDACTED] 2024 | Scaling Efficient Decentralized Research for AI, With AI

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

1. 0:00

   so I'd like everyone to think about a time when they tried to contribute to an open source project they thought they were really bringing something to the table and they eventually gave up because while it was not that easy to find information and it was not that easy to connect with the team also now think about a time when you had a crazy idea and it turns out well you were lacking some skills so for example you were making Maybe lacking a Quantum physicist on your team to uh

2. 0:33

   make that that project or a marine biologist well I just one giant lab or jle we are enabling Science and Tech communities our mission Making Science and Innovation 10x more impactful 10x more impactful per dollar invested 10x easier to interact with communities and build things and 10x more pleasant to actually do science and do Innovation so we started few years ago um in an open source context uh we were

3. 1:09

   working on covid-19 then antimicrobial resistance and then uh in general in uh vaccine uh implementation and research and we just built a very rary platform for funding uh people could come with their project apply for funding they could discover other project we made it very easy for people to merge and for people to access information and learn if they could not contribute they could

4. 1:36

   learn and so we invested 100K in um in 50 projects and all of it was uh Community Based a community Corum and the output was 22 functional products 10 PE reviewed papers and one ex prise finalist that was against a series B company uh they had received 15K and what happened is that a lot of the projects that we worked with ended up merging and helping each other and they ended up you know we ended up finding a

5. 2:09

   laboratory assistant launching a company on raising millions of dollars we ended up seeing uh high schoolers actually building very credible product uh we ended up seeing crazy things from unexpected people just because there was really like a strong help within the community on access to information so 10x is our aspiring Target uh but it's it's way beyond that it's 10x better collaboration experience in open source

6. 2:41

   communities to summarize what we learned is that emergent was much better than top down information access was key for people to feel like they are belonging and not feel rejected and really like feel like it's worth insisting in this community to contribute and meritocratic and learning based let's do a little review of the different architecture for open source projects so we have the not so good this

7. 3:06

   is 95% of the projects few maintenance few contributors either the project is not that great and you don't have contributors and it's going to die or it's amazing and then it's actually very hard to be a contributor um not a very very strong sense of community the tools are going to be GitHub rmy and slack and you have to figure it out on your own then we have the pretty good pretty good ones uh looks like this they have a community of users and contributors that

8. 3:39

   are here to help it's basically your support center slash after sales service it's what's making your open source product easy to use so tooling typically for the core team it's going to be slack or Discord for the contributors it's going to be zulip and for the overall Community it could be course it could be Reddit it could be anything uh else so what's the issue with that well it's pretty hard to find uh

9. 4:09

   collaborators if you walk in and you're not able to pick up a task immediately what you want is to connect with people is to learn is to be part of that Community until your time comes and with those tools it's very hard to actually search for collaborators on discard you don't have profiles so you don't even know who is doing what you don't know you don't even know who has what skills it's how to find resources uh slack

10. 4:33

    doesn't index data sets slack doesn't index models uh it's hard to find events events is the best way to have your first contact with a community they're not indexed anywhere and it's frankly hard to figure out what to do um I was with maker da as a as a facilitator for two years and a half and today if you told me go recruit some maker D people to do a project I can't even tell you

11. 4:58

    what people are doing I can't even tell you what their skills are it's there is no sense of community or searchability really in this ecosystem so what are the needs of Open Source AI well open source AI is extremely different in the sense that we have many many applications uh so think of like 30 types of application just in synthetic biology then you have mathematics then you have education then

12. 5:29

    you of particle physics and the relationship of those problems and those communities with uh the open source Community uh of AI is very different from the one that between r on a team of researcher or python or a team of researcher they're going to need a much higher level of support to be at the frontier of what we can do in their field with AI um so open source AI needs to be set up for that and it's it's a matter of being competitive against open Ai and Google open Ai and Google they

13. 6:05

    have support team for each of those vertical helping them actually Implement open source AI doesn't have that so if we want to be competitive versus the centralized Alternatives then there needs to be some support communities and it's also a matter of value capture on tokenomics I think our dream is open source contributors can make a living we have toonomic they contribute they gain tokens back uh but for this to happen the value creation needs to be within the community and so

14. 6:39

    here's what open source ey would look like you have a call team you have contributors you have research topics so could be Lan could be synthetic biology it could be anything really and those communities are easy to find they are searchable uh the skills of the people and the resources the data set the models are easily searchable findable and you have peer-to-peer integ uh peer-to-peer interactions that are organic now why is this not happening

15. 7:10

    today uh because to be able to do what I showed the page before you need to have multiple tools and collate them together means multiple login means a huge drop off um resources people task events needs to be found again multiple tools um and multiple tools means lack of searchability also there is a huge information overload uh must have happened to you you're out for a week you go back 175 notification on Discord

16. 7:44

    how do you catch up um myself I've been trying to rec contribute to makeer that couldn't catch up too much so there is an information overload or rather lack of relevant information for people who want to engage think of your har bad professor in say I don't know genetics they want to interact with an AI Community well just give them something to learn give them something to do uh don't give them 200 notifications on Discord this is not the way uh to engage

17. 8:15

    them and the issue for organizers is again many tools and if you want peer-to-peer interaction today if the question is who in this ecosystem is an expert in synthetic biology and I can help them with my AI skills you need to ask the admin hey who is an expert in synthetic biology that should just be a search so here's how we have solving this at just one giant lab so we created objects that can structure ecosystem so very simple events slack like

18. 8:49

    discussions documents library of papers that is crowdsourced resource indexing like data set models C++ libraries any type of libraries uh needs and tasks that can be shared not just with developers but also with product people with designers with people who understand the actual problem that's that's being solved and call for proposal slhck and primitive many things to say about this on how we're dealing with voting on such but I'll be brief and what we're working on currently is a

19. 9:22

    zulip like Forum to be able to handle uh you know broader Community Beyond just core contributors uh governance open source third party integration so here we integrated already with hagging face with GitHub with uh archives and soon with the near AI Hub and bring your own agent so what can we do there the problems that agents can solve for us is what I mentioned earlier um I'm trying to engage with this community

20. 9:54

    what can I do um so those agents are going to ingest juggle object and propose um someone a user events they should go to threads they should look at needs or tasks that they could pick up or that their team could pick up resources that they could build upon that's the first thing the second thing is from our experience running a successful research program takes a person on a half if you really want to

21. 10:24

    scale it to thousands of people and a lot of those tasks can be automated with agents and we're talking simply about you know summarizing meetings making sure people are connected make making sure that there are notes for everything making sure that uh relevant events are created at the right point in the life cycle of a project and so on and the last problem is every single lab in research is asking themselves what can I do with AI every single one

22. 10:55

    and every single one the most advanced one is are trying to build specialized agents and their problem is that everyone is using the specialized agents in the corner so let's invite the specialized agents in the team our goal is for anyone to be able to bring a specialized agent from The neari Hub or from hugging face for example going to give you a quick demo here so oh this is the old video okay I'll do

23. 11:27

    I'll do with that that's fine so this is the public Hub and here you have different discussion topics you're going to have is it plan oh here you go so you have the Hub The Hub is structur in different discussions you just show the libraries of paper this is an archive but that's publishing each time there is a new archive paper here multiple apps I can have a but for all pool requests from hugging face from GitHub I can have an archive butt that's just going to notify me each

24. 12:06

    time there is a new paper within a label and I can add an AI agent that's trained on my library of uh papers so for example here there is this new paper that was published I'm asking the agent should we add this paper to our library and you see what this is so right right now what what we're doing is that we're building the pipe so anyone can come and just create their agents and invite their agents to their to their

25. 12:43

    discussion here we go so what's next for us um we've been building for 18 months now um we've been building with five big uh institutional partners and few open source communities U now we're getting into the alpha to the beta we're going to scale to 10 to 20 institutional clients um and five open source communities this year we made 75k we've been really like charging clients for

26. 13:17

    only few months uh next year objective is 10x we like we like 10x so 700k Revenue objectives next year and in terms of client pipeline we are really not not just focused on open source communities uh we're working with uh top Research Institute National Institute for cancer uh Pastor Institute big Consortium of like hundreds of laboratory in Europe in India um IM foundation on the open source side uh

27. 13:48

    we're going to be the reference platform for citizen science uh Europe uh ignite Life Science Foundation a big network of uh research institutes who wants to have to create open source communities around them synthetic biology communities and xrpl uh xrpl comments what we see from those organization is that every Rockstar in their field wants to spend few hours a week contributing

28. 14:21

    to projects very very few of them are able to find the projects that really excites them to contribute and so what they're asking for is an easy way to contribute to existing teams or just launch projects from scratch and that's what we that's the opportunity that we're giving them and that's why for us it's very important to not just focus on open source communities or traditional institutions but do

29. 14:44

    both that was it thank you if you want to reach out this is my email lj. uh we're both recruiting on raising