Feb. 19, 2025
Everything Bagel: How ZKPs Are Scaling AI with Privacy & Ownership with Founder, Bidhan Roy
How can developers monetize open-source AI while preserving ownership and privacy?
In this episode of Everything Bagel, co-hosts Alex Kehaya and Bidhan Roy, Founder of Bagel Network, dive into the revolutionary role of zero-knowledge proofs (ZKPs) in open-source AI monetization.
Bidhan shares how Bagel Network is transforming the AI landscape by enabling developers to earn from their contributions while maintaining full control and privacy. They discuss the game-changing breakthrough of ZK LoRa, a zero-knowledge proof system that slashes AI training verification times from weeks to under two seconds—paving the way for scalable, decentralized AI.
The conversation explores how community-driven ASI (Artificial Super Intelligence) can evolve through modular AI models, making decentralized AGI (Artificial General Intelligence) more accessible and cost-effective. Plus, they unpack how The Bakery, Bagel’s latest product, empowers developers to stack AI components like LEGO bricks for greater scalability and efficiency.
🔥 Key Topics Discussed:
✅ How zero-knowledge proofs are reshaping AI ownership and monetization
✅ The impact of Bagel Network’s ZK LoRa verification system
✅ The rise of domain-specific ASI models and real-world applications
✅ How The Bakery fuels decentralized AI development at scale
✅ The potential for community-built AGI running on crypto rails
🔗 Learn More & Join the Community:
🌐 Website: Bagel Network
🐦 Follow on X (Twitter): @BagelOpenAI
📺 YouTube: @bagelnet
Show Links
The Index
X Channel
YouTube
Host - Alex Kehaya
Producer - Shawn Nova
Chapters
00:06 - Zero-Knowledge Proofs in Machine Learning
12:58 - Decentralized AI and Monetization Opportunities
24:14 - Exciting Developments in Decentralized AI
Transcript
WEBVTT
00:00:06.290 --> 00:00:10.316
Welcome to the index podcast hosted by Alex Cahaya.
00:00:10.316 --> 00:00:19.864
Plug in as we explore new frontiers with entrepreneurs, builders and investors, shaping another episode of Everything Bagel.
00:00:38.240 --> 00:00:54.104
I'm your host, alex Cahaya, with my co-host Bidon Roy, the founder of Bagel, co-host Bidhan Roy, the founder of Bagel, and today we are going to be focused on what zero-knowledge proofs are and their significance in machine learning, especially in the context of monetizable open source AI.
00:00:54.104 --> 00:01:02.822
Bidhan, do you want to kick it off and maybe just talk about the recent announcement that you guys made about the zero-knowledge proof, the zero-knowledge verification of LoRa training?
00:01:02.822 --> 00:01:05.165
Let's talk about that, the zero knowledge verification of LoRa training.
00:01:05.185 --> 00:01:05.686
Let's talk about that.
00:01:05.686 --> 00:01:06.286
Yeah, yeah, happy to.
00:01:06.286 --> 00:01:11.933
So we recently made a breakthrough in research internally, which we open sourced.
00:01:11.933 --> 00:01:14.036
That's what Alex is talking about.
00:01:14.036 --> 00:01:19.938
But before going into that, I want to take a step back and kind of set the stage on what this means.
00:01:19.938 --> 00:01:21.667
Then I can go into the technicals of that.
00:01:22.480 --> 00:01:35.007
So overall, from a bird's eye view, what we do at Bagel is we make open source AI monetizable, and what that means is that we allow open source developers to contribute to open source models.
00:01:35.007 --> 00:01:43.007
That means like training model parameters, and then they can earn the portion of the future revenue of that model in different ways.
00:01:43.007 --> 00:01:57.647
And one thing that's very important to make open source AI monetizable in this way is that we have to preserve the ownership of all of those contributions that are being made by open source developers.
00:01:57.647 --> 00:01:59.486
And what do we mean by ownership?
00:01:59.486 --> 00:02:05.513
By ownership we mean that they have full control of those resources that they contribute.
00:02:05.513 --> 00:02:15.332
In our specific case, this is some model parameter that they contribute, and how can they have ownership and control of their resources if they don't have any privacy?
00:02:15.332 --> 00:02:21.248
Because if it's open how it works today in traditional open source if it's open for the whole world.
00:02:21.248 --> 00:02:25.735
Anyone can copy it and run it locally or deploy it in their system.
00:02:25.735 --> 00:02:27.502
There's nothing stopping them.
00:02:27.502 --> 00:02:31.432
So the open source developers do not have any control over it.
00:02:32.139 --> 00:02:50.992
So one of the first things and one of the biggest problems that we had to solve is allow for the open source contributions private but at the same time we give away for the open source developers to prove that they actually contributed without revealing their contributions to the world or any other model owner.
00:02:50.992 --> 00:03:10.211
So that's the kind of the background on why this relates to what we do, and this has traditionally been a big problem, like one of the biggest problems in machine learning space period, because it's really hard to prove your contribution without revealing it.
00:03:10.211 --> 00:03:12.181
There are a lot of things that goes into it.
00:03:12.181 --> 00:03:16.551
A lot of people have tried using zero knowledge proofs to do that.
00:03:16.551 --> 00:03:29.610
I'll give like quick one line about zero knowledge proofs are they are exactly what it sounds like Like you can prove something without revealing the details of that thing.
00:03:29.610 --> 00:03:43.485
So for our case, the zero-knowledge proof will apply to an open-source developer proving that they contributed to a model, but they provide zero knowledge of the actual contribution to the rest of the ecosystem.
00:03:44.539 --> 00:03:45.181
People might.
00:03:45.181 --> 00:03:53.769
They might be running like, like trying to fine tune Lama 4 or whatever, or DeepSeek right, these open source models and the fine tuning.
00:03:53.769 --> 00:03:57.570
Can you explain a little bit for people who might not be aware of just like, what the fine tuning actually is?
00:03:57.570 --> 00:04:05.849
But before you do that, you know they use parameters to fine tune these models, to improve them, and the parameters could be.
00:04:05.849 --> 00:04:23.069
You could see the source code, the actual code they write to define those parameters, or you could keep them private, like if the source code is available, then any developer can just run it locally on their own copy of Lumbafor without the original creator of those parameters benefiting from them doing that monetarily right.
00:04:23.069 --> 00:04:34.148
And so this is a way for you to keep the parameters that you're doing closed actually and monetize it but still allow anybody else to leverage the LLM and the improvements and add on to it in the future.
00:04:34.148 --> 00:04:36.586
So it's kind of a middle ground.
00:04:37.629 --> 00:04:38.360
Exactly so.
00:04:38.360 --> 00:04:47.523
One thing I'd like to mention that the technology we have developed it's applicable for general purpose training, so you can train models from scratch with that.
00:04:47.523 --> 00:04:54.845
The first use case that we have built around it is the fine-tuning part, because there's a big market need for that.
00:04:54.845 --> 00:05:05.809
But the actual technology that we have, we foresee that being used by millions of developers of coming together and building a massive, massive model from scratch.
00:05:05.809 --> 00:05:07.302
There's that context.
00:05:07.302 --> 00:05:11.112
But I want to go back to the technology, actual technology that we are.
00:05:11.680 --> 00:05:14.384
We just open sourced, so we call it ZK-LORA.
00:05:14.384 --> 00:05:26.903
What it does is it allow developers to prove that they develop some kind of plugin, like small parameter set for a model, which are called LoRa.
00:05:26.903 --> 00:05:32.747
In traditional machine learning, lora stands for low-ranked adaptation of large language models.
00:05:32.747 --> 00:05:46.492
So they create a LoRa and those developers can send a zero-knowledge proof that they trained this LoRa to me and I can verify and why this is big.
00:05:46.492 --> 00:05:51.125
So some context on the zero knowledge space in general.
00:05:51.125 --> 00:05:52.930
Zero knowledge is very expensive.
00:05:52.930 --> 00:06:00.132
It takes a lot of time to generate zero knowledge proofs and it takes a lot of resources to build the proofs and all of that.
00:06:00.620 --> 00:06:03.930
Yeah, physical compute power, right, like it's a lot of energy, a lot of compute.
00:06:04.901 --> 00:06:13.370
Not just that, like, even if you throw a lot of energy and compute, even then it takes like hundreds of hours, if not days or weeks sometimes, if you want to do that.
00:06:13.370 --> 00:06:30.550
And the zero knowledge proofs has been used for model inference before, like if a model has been generated, an inference if it's coming from a certain model, but it has never been used for model training, what we are doing.
00:06:30.550 --> 00:06:31.072
Why?
00:06:31.072 --> 00:06:34.831
Because it's very, very slow to use for model training.
00:06:34.831 --> 00:06:42.394
You cannot use the zero knowledge proofs for proving model training until now.
00:06:42.394 --> 00:06:46.781
But why it wasn't possible model training until now?
00:06:46.781 --> 00:06:47.362
But why it wasn't possible?
00:06:47.362 --> 00:06:48.024
Because it's like it take.
00:06:48.024 --> 00:06:56.249
It is to take like weeks literally to be able to do verified model training, even for very small models, the models you cannot even use like, let alone the state-of-the-art ones.
00:06:56.249 --> 00:07:04.512
That's where we are and from that we took a leap with our publishing of our research.
00:07:04.512 --> 00:07:16.254
All of a sudden you can do this ZK Verify training on state-of-the-art open source models under two seconds.
00:07:17.922 --> 00:07:18.987
Wow, that's incredible.
00:07:19.439 --> 00:07:19.942
Yeah, thank you.
00:07:19.942 --> 00:07:21.988
That's a massive, massive leap.
00:07:22.680 --> 00:07:30.651
Can we go into the details a little bit, like just generally speaking, it's really interesting startups in general when they have these kinds of breakthroughs.
00:07:30.651 --> 00:07:37.439
You know it's often not the team that has like all the money and hundreds of employees.
00:07:37.439 --> 00:07:58.749
A lot of times it's like the small scrappy startup with a couple of engineers and that's kind of you guys Like yes, you have, you guys raised a pre-seed couple million bucks, but you're like nine people on your team and this kind of discovery is like people have worked on this for a long time, right, like they've tried to figure this out and the best they could do was like thousands of hours.
00:07:58.749 --> 00:08:02.223
So how did you guys figure this out?
00:08:02.223 --> 00:08:03.826
It's kind of crazy.
00:08:03.826 --> 00:08:04.648
How did you do that?
00:08:05.149 --> 00:08:09.204
I would like to say that this is the kind of a deep seek moment.
00:08:09.204 --> 00:08:15.605
So all the folks who have been working on this problem they have been kind of hammering at it.
00:08:15.605 --> 00:08:27.483
They are trying to solve, from a certain way, like we, we do not have that many resources or that many people or that many GPUs to run, have that many resources or that many people or that many GPUs to run.
00:08:27.483 --> 00:08:28.365
We had to be creative with it.
00:08:28.365 --> 00:08:32.561
We had to go back and we had to think from first principles like what else can we do to make it fast?
00:08:32.561 --> 00:08:45.860
That's how we came up with a totally different solution and the unique insight we had is, like all the folks who were spending so much money and they are trying to prove it for so long and most of the companies failed to be able to do so, they, all of them, are trying to prove it for so long and most of the companies failed to be able to do so.
00:08:45.860 --> 00:08:58.851
They, all of them, were trying to prove this is a little bit technical they were trying to prove that some compute was used in some gpu to train some model.
00:08:58.851 --> 00:09:10.174
So they were trying to get a proof from a model trainer that they burned some compute right and that makes sense, like that's the first kind of problem, the way you want to solve it.
00:09:10.174 --> 00:09:14.712
And people have been trying to make it faster, like again and again forever.
00:09:14.712 --> 00:09:16.259
And it became a lot faster.
00:09:16.259 --> 00:09:18.423
It became thousand x faster as well.
00:09:18.423 --> 00:09:23.655
But even after thousand x acceleration it's still thousands of hours for very small model.
00:09:23.655 --> 00:09:25.659
It's far away from being useful.
00:09:25.659 --> 00:09:28.565
But we could not do that.
00:09:28.565 --> 00:09:30.989
We had to think differently.
00:09:30.989 --> 00:09:38.455
So we thought, like, why we ask people to prove, prove that they burned some compute?
00:09:38.455 --> 00:09:39.778
That does not make sense.
00:09:39.778 --> 00:09:42.284
What do you do in traditional machine learning when we train models?
00:09:42.284 --> 00:09:50.730
Do we ask people like, hey, do we ask open ai that, hey, did you burn some gpu hours to build this model?
00:09:50.730 --> 00:09:53.205
No, we just use the model and see if the performance is good.
00:09:53.205 --> 00:09:54.410
Right, that's how it works.
00:09:54.410 --> 00:10:03.004
And even when they're training the models, a lot of training runs are thrown out of the window just because they do not meet the evaluation that's required.
00:10:03.004 --> 00:10:14.561
So what the question they're asking is that if the training run met the desired evaluation score from some specific evaluation dataset, that's it.
00:10:14.561 --> 00:10:19.988
So that kind of was the unique insight for us.
00:10:20.788 --> 00:10:30.351
We decided, instead of asking the provers to prove that if they burnt any compute, we will ask them if they improved the model.
00:10:30.351 --> 00:10:33.524
Does not matter how much compute they burnt, we don't care.
00:10:33.524 --> 00:10:36.005
So that's the first insight.
00:10:36.005 --> 00:10:41.690
And second insight we had is that you do not really prove the whole model.
00:10:42.510 --> 00:10:57.822
We built this modular architecture for open source models where, instead of being a monolithic thing, all the developers can kind of provide smaller contributions, which kind of stack on top of each other like Lego pieces.
00:10:58.581 --> 00:11:09.113
Which kind of stack on top of each other like Lego pieces, right, and all of a sudden the contributions they have are so much smaller that the proving on top of that is way more efficient than proving on the whole model.
00:11:09.113 --> 00:11:24.535
So those are the two unique insights we had and that's how we were able to come to a solution where we allow developers to contribute their own Lego pieces and they are proving that they actually improved the model with those Lego pieces.
00:11:24.535 --> 00:11:30.258
It does not matter if some compute was burnt or not burnt at all.
00:11:30.258 --> 00:11:34.653
We care about the performance and we ran with it.
00:11:34.653 --> 00:11:47.275
We figured out how to do it, we did the benchmarks and we were able to get under two second verification time for LAMA 70B, which is the state of the art model.
00:11:47.275 --> 00:11:55.359
Previous ones was like the maximum we have done on LLMs, zkml or whatever was like 1.5 billion or something.
00:11:55.359 --> 00:11:57.231
Even that took hundreds of hours.
00:11:58.304 --> 00:12:03.027
There's a similar story I think I might've told you this before of something like this with Instagram.
00:12:03.027 --> 00:12:06.355
Right, the founders of Instagram used to work at.
00:12:06.355 --> 00:12:26.791
One of them used to work on the Gmail team at Google and when he was there they kind of knew that they figured out that, like if you started entering in your username your email address, there was like a 90% chance you were going to log in and so they started preloading your account for you as soon as you started typing.
00:12:26.791 --> 00:12:36.773
And he took that insight over to Instagram and and basically, if you started writing the caption on a photo, they just uploaded the photo to their database for you and so you'd hit submit and be like bam, the photo was there already.
00:12:37.215 --> 00:12:41.191
And people were like blown away because they were like, what kind of CDN are they using?
00:12:41.191 --> 00:12:44.969
Like there must be some crazy tech, you know, but it was really just a process thing.
00:12:44.969 --> 00:12:50.850
It wasn't even like they invented some new fast way to upload, you know, data on the internet.
00:12:50.850 --> 00:12:53.687
Like the pipes didn't get any bigger, they just use statistics.
00:12:53.687 --> 00:12:55.893
Yeah, it was just a process innovation.
00:12:55.893 --> 00:12:58.179
That was in hindsight.
00:12:58.179 --> 00:13:06.333
It it's like you're kind of like, wow, that's so obvious, but it wasn't right, like no one was thinking about it and I just find it interesting, like the path to these insights.
00:13:06.333 --> 00:13:10.245
There's this forcing function that gets you to this insight.
00:13:10.245 --> 00:13:11.653
That's super interesting to me.
00:13:12.841 --> 00:13:13.345
No, definitely.
00:13:13.345 --> 00:13:18.639
I mean, I can talk from our experience, right, I can talk from our experience right.
00:13:18.639 --> 00:13:27.274
We could have just taken the easy path and, you know, try to kind of bang the hammer on the nail forever and try to squeeze out a little bit of performance.
00:13:27.274 --> 00:13:28.630
We're like, no, it doesn't make sense.
00:13:28.630 --> 00:13:49.197
We have to come from the first principle and we believe that we have started a new era in verifiable decentralized training, which is the verifiable part is important because only then these decentralized training providers can monetize their work.
00:13:49.197 --> 00:13:51.932
Otherwise it's not possible.
00:13:51.932 --> 00:13:59.775
All the other attempts that has been made on this problem they are decentralized training but they are not ZK verifiable.
00:13:59.775 --> 00:14:02.714
They're like kind of put together with some duct tapes, which is great.
00:14:02.714 --> 00:14:11.971
But we opened up this new era where we showed the industry where these does not need to be attached together with duct tapes.
00:14:11.971 --> 00:14:21.615
They can be verified with zero knowledge, proofs at scale, with great latency and with 100% security guarantees.
00:14:22.524 --> 00:14:24.188
Yeah, you can be super confident.
00:14:24.188 --> 00:14:35.711
No one's gaming the system, like this contribution actually did improve this model by this much, and then you can use that data to calibrate how much a developer should be compensated when that thing is used.
00:14:35.711 --> 00:14:39.469
I've actually been really curious about that part, if you can elaborate a little more.
00:14:39.469 --> 00:14:44.549
So, now that you've got this proof, do you guys have an algorithm that weights these contributions?
00:14:44.549 --> 00:14:50.409
How does the system know my contribution is worth 50 cents and yours is worth 25?
00:14:50.951 --> 00:14:55.298
So we are going to release more information on how it works.
00:14:55.298 --> 00:14:57.008
It's really easy to do so.
00:14:57.008 --> 00:14:59.414
It's possible to benchmark those contributions.
00:14:59.414 --> 00:15:00.336
That's how they work.
00:15:00.825 --> 00:15:07.967
Yeah, like once you have the proof, you can benchmark it, you can prove that the benchmark is accurate really is what it is, Exactly, exactly.
00:15:08.249 --> 00:15:10.254
Yes, Okay, cool.
00:15:10.254 --> 00:15:18.375
Those are some of our additional proprietary research that we're doing internally, but we're going to open source it as we roll out our protocol.
00:15:24.264 --> 00:15:46.272
One of the things that I keep thinking about about ZKP in general, but I also think, just like crypto and the investment that's gone into things like distributed systems and trying to decentralize these different networks and stuff, have like other use cases that people might that might be like non-obvious, like what's the impact in AI to AI, with this ability to verify, outside of just being able to prove for monetization purposes?
00:15:46.272 --> 00:15:53.975
Are there other use cases that this technology, this insight that you guys gained about this, could be used for?
00:15:54.828 --> 00:15:57.672
First of all, I think monetization is the killer use case.
00:15:57.672 --> 00:16:12.753
So we, as a decentralized AI industry, have been trying to sell verifiable inference or whatever, like lots of other things, to the Web2 users, and they don't care about verifiability, but what they care about is monetization.
00:16:12.753 --> 00:16:19.298
So if we show them that this is what you need to do to monetize your contributions, then it makes sense for them.
00:16:19.298 --> 00:16:22.975
If you just tell them, like the technology, like this, is verifiable, you say blah, blah, blah.
00:16:22.975 --> 00:16:27.437
A lot of people care, the researchers care, but their users don't care.
00:16:27.437 --> 00:16:32.110
Consumers don't care right, that's how we see it.
00:16:32.913 --> 00:16:35.018
A lot of other things that's possible through this.
00:16:35.018 --> 00:16:39.947
Like we have built the first product which is leveraging this technology.
00:16:39.947 --> 00:16:42.023
It's called the bakery and it's blowing up at the moment.
00:16:42.023 --> 00:16:48.708
We have had more than 14,000 signups on it since it was launched a month ago, so it's actually blowing up.
00:16:48.708 --> 00:16:54.167
And the second there are some frontier business models that we can enable with this.
00:16:54.167 --> 00:16:56.092
I would love to talk about them when we are in process of launching them.
00:16:56.092 --> 00:16:59.240
So I want to make sure I talk about them when we're in process of launching them.
00:16:59.240 --> 00:17:01.870
So I want to make sure I talk about them when it's public.
00:17:01.870 --> 00:17:04.605
I want to make sure that they kind of match the time.
00:17:04.605 --> 00:17:10.106
But there are some frontier business models that we're enabling which was never possible before period.
00:17:10.106 --> 00:17:13.413
So it's not just not research.
00:17:13.413 --> 00:17:19.596
This research opens up new business models, entirely new, entirely innovative business models.
00:17:20.765 --> 00:17:24.868
Well, so you guys have the bakery, but you also have your newsletter where you're sharing all of this research, right?
00:17:24.868 --> 00:17:29.808
You guys have done a ton of research and that newsletter has got like 50,000 subscribers, you know.
00:17:29.808 --> 00:17:32.576
So who are these like on the subscriber list?
00:17:32.576 --> 00:17:36.349
I'm curious, like, who are those people and who do you think they are that are reading the blog?
00:17:36.349 --> 00:17:38.374
And then the same thing with the people using the bakery.
00:17:38.374 --> 00:17:43.153
Today there's 14,000 subs, or they're not subscribers, they're just users of the bakery product.
00:17:43.153 --> 00:17:44.388
Like, who are those people?
00:17:44.388 --> 00:17:45.452
You know what are they doing.
00:17:46.085 --> 00:17:48.253
There's a big overlap between those two, by the way.
00:17:48.253 --> 00:17:55.976
So the readers we have on our blog research blog there is just people who are interested in machine learning.
00:17:55.976 --> 00:17:57.547
They want to learn about the technology.
00:17:57.547 --> 00:18:00.451
Interested in machine learning, they want to learn about the technology they want to learn about.
00:18:00.451 --> 00:18:01.994
They care about monetizing open source AI.
00:18:01.994 --> 00:18:12.326
That's why they subscribe to what we say and it's not something an assumption we have made.
00:18:12.346 --> 00:18:19.732
We did a survey among our 50,000 subscribers and 73% of them gave us the answer that they are actively looking for a solution where they can monetize their open source contributions.
00:18:19.732 --> 00:18:22.452
So we are like very aligned, mission aligned.
00:18:22.452 --> 00:18:26.674
Open source developers are like really behind what we do and what we say.
00:18:26.674 --> 00:18:30.614
So there's that, and some of them convert to the users.
00:18:30.614 --> 00:18:32.351
So that's how we kind of monetize.
00:18:32.351 --> 00:18:32.825
We don't.
00:18:32.825 --> 00:18:38.178
We have actually gotten offers to put ads on our blog for 20 grand or so.
00:18:38.178 --> 00:18:41.574
We're like that's not the way we're going to monetize.
00:18:41.574 --> 00:18:48.724
We're going to monetize that the readers who happen to be our target customers and know about us.
00:18:48.724 --> 00:18:52.554
They trust our brand and then they use the products that we build.
00:18:52.554 --> 00:18:55.146
That's how we gain the trust among them.
00:18:55.146 --> 00:19:07.338
So the summary of what I just said it's basically kind of the same demographic of users and developers on both our newsletter and the product.
00:19:08.025 --> 00:19:13.530
So, okay, let's talk about that a little bit more, because you have 14,000 of them that have converted over there now using the bakery.
00:19:13.530 --> 00:19:14.815
What kinds of things are they cooking up?
00:19:16.365 --> 00:19:17.911
A lot, a lot of things actually.
00:19:17.911 --> 00:19:23.896
So we are seeing a lot of agents and we're seeing a lot of, as I said, like some frontier business models that they're exploring.
00:19:23.896 --> 00:19:27.974
I'd love to talk about them as we work with them and release some of them.
00:19:27.974 --> 00:19:36.232
But there are some frontier business models, like some developers are coming together and they're going against giants not just open AI and stuff.
00:19:36.232 --> 00:19:40.622
Like they're actually building domain specific models, not just open AI and stuff.
00:19:40.622 --> 00:19:46.951
Like they're actually building domain specific models which are somewhere between 25 to 50% better than in their specific domain, than the state of the art model.
00:19:48.317 --> 00:19:54.414
So some people are building some model in a specific domain and let's say it's like some domain.
00:19:54.414 --> 00:19:56.663
I'm going to give an arbitrary example.
00:19:56.663 --> 00:20:12.736
It's a domain where the model knows about podcasts, let's say and then that model performs better 25 to 50% better than state of the art, like cloud anthropic, like whatever state of that model there is cloud anthropic, like whatever state of that model there is.
00:20:12.736 --> 00:20:18.428
That model performs 60% better than the state of the art models when it talks about podcasts.
00:20:18.428 --> 00:20:22.505
Those are the domain specific use cases that we're seeing.
00:20:23.796 --> 00:20:36.521
What I'm really excited to see and really curious to see how it plays out is where you know, I come in, I fine tune a model with my parameters and then you take that same thing and you fine tune it further and then we both get compensated.
00:20:36.521 --> 00:20:42.244
Like that, that mixing of that cross pollination between things is going to be really cool to see.
00:20:42.244 --> 00:20:45.220
Is that something you're already starting to see amongst some of the users?
00:20:46.221 --> 00:20:46.623
Exactly.
00:20:46.623 --> 00:20:49.436
Yeah, I mean the best part about the architecture that we have.
00:20:49.436 --> 00:21:04.675
It's not even like traditional fine-tuning like fine-tuning does not explain the architecture that well, but like the, what's happening here is that they add their modular contribution on top right and all of a sudden, let's say there's a new shiny model came out.
00:21:04.675 --> 00:21:11.719
Traditionally, what happened is that people have to redo the work from scratch on the new base model.
00:21:11.719 --> 00:21:16.127
But for what we build, all of the contributors already had the Lego piece.
00:21:16.127 --> 00:21:20.586
They just take it out and put it on top of the new shiny model.
00:21:20.586 --> 00:21:23.564
They built it for DeepSeek V2.
00:21:23.564 --> 00:21:30.645
Now V3 came out, they take it out and put it on top, and that reduces cost for them exponentially as well.
00:21:31.315 --> 00:21:32.380
Yeah, increases speed too.
00:21:32.380 --> 00:21:34.701
Time to market right, boom.
00:21:34.701 --> 00:21:37.442
It's like how long does that take somebody to do that switch?
00:21:38.204 --> 00:21:41.284
Exactly that's what we're enabling with the architecture.
00:21:41.284 --> 00:21:46.207
The first piece of research that we published is the verifiability part.
00:21:46.207 --> 00:21:54.638
That is revolutionary even today and we think it will open up new era in decentralized, verifiable training.
00:21:54.638 --> 00:22:00.155
But at the same time, that's just tip of the iceberg of the things that we're working.
00:22:00.636 --> 00:22:03.164
Well, let's talk about AGI Artificial General Intelligence.
00:22:03.184 --> 00:22:21.444
One of the things you and I have talked a lot about like the last couple of weeks, is this idea of open intelligence, open intelligence being AGI running on crypto rails this community owned, and I think Bagel has built the core component to that right, especially with the ability to monetize these open source contributions.
00:22:21.444 --> 00:22:32.626
What do you think are the components, the other components that need to happen or like how can this community actually, like hypothetically, achieve building an AGI?
00:22:32.626 --> 00:22:35.833
Can this community actually, like hypothetically, achieve building an AGI, even if it could?
00:22:35.833 --> 00:22:45.688
And also like another sub question is like could it be like a point specific thing, like a niche AGI or I guess general intelligence can't be that niche, but I don't know what your perspective is on that or what's what's like actually makes any sense.
00:22:45.688 --> 00:23:00.801
But I guess mostly I'm concerned about or want to understand, like, what the actual parts of it are that the community can contribute to, to actually make that if we want to see something like open intelligence exist agi, asi like those are broad concepts, right.
00:23:01.403 --> 00:23:12.761
and if you want an asi as a community, let's say you want an asi which knows about everything, about every domain there is definitely can build it but it's not the most cost effective.
00:23:12.761 --> 00:23:31.835
But as a group, like some community members, you care about building an ASI which has ASI level intelligence for math or ASI level intelligence for machine learning, asi level intelligence for podcasts, but all the other stuff it does not necessarily need to know about.
00:23:31.835 --> 00:23:39.747
So all of a sudden it's like we're more cost effective and those are the use cases we enable with our first product that we have built, the bakery.
00:23:39.747 --> 00:23:44.006
So that's the kind of domain-specific ASI thing.
00:23:44.335 --> 00:23:54.282
Yeah, it'd be really cool to see how fast this snowballs for something Like there's going to be one specific use case that snowballs to like super intelligence level ai.
00:23:54.282 --> 00:23:59.931
Are you seeing any trends where there's like lots of focused attention on you know?
00:23:59.931 --> 00:24:07.450
It's like you said, agents are one, but it's like a specific domain you're seeing where, like, lots of devs are like working on this and compounding each other's parameters?
00:24:07.450 --> 00:24:10.038
Is there one area that you guys have insight into?
00:24:10.880 --> 00:24:14.087
we actually have have a really good, interesting thing cooking.
00:24:14.087 --> 00:24:21.708
I am so excited to talk about it, but I want to wait until it's there for the community to see, just to keep up the hype, Right.
00:24:21.788 --> 00:24:22.710
Okay, all right.
00:24:23.436 --> 00:24:24.761
I am very excited about this.
00:24:25.415 --> 00:24:26.862
I am too, because I don't even know about it.
00:24:26.862 --> 00:24:30.162
We talk like daily, so not daily, but we talk a lot.
00:24:33.035 --> 00:24:34.221
So I'm interested to figure out what that is.
00:24:34.260 --> 00:24:37.074
It's going to be exciting when it comes out for sure, like how exciting are we talking about?
00:00:06.290 --> 00:00:10.316
00:00:10.316 --> 00:00:19.864
00:00:38.240 --> 00:00:54.104
00:00:54.104 --> 00:01:02.822
00:01:02.822 --> 00:01:05.165
00:01:05.185 --> 00:01:05.686
00:01:05.686 --> 00:01:06.286
00:01:06.286 --> 00:01:11.933
00:01:11.933 --> 00:01:14.036
00:01:14.036 --> 00:01:19.938
00:01:19.938 --> 00:01:21.667
00:01:22.480 --> 00:01:35.007
00:01:35.007 --> 00:01:43.007
00:01:43.007 --> 00:01:57.647
00:01:57.647 --> 00:01:59.486
00:01:59.486 --> 00:02:05.513
00:02:05.513 --> 00:02:15.332
00:02:15.332 --> 00:02:21.248
00:02:21.248 --> 00:02:25.735
00:02:25.735 --> 00:02:27.502
00:02:27.502 --> 00:02:31.432
00:02:32.139 --> 00:02:50.992
00:02:50.992 --> 00:03:10.211
00:03:10.211 --> 00:03:12.181
00:03:12.181 --> 00:03:16.551
00:03:16.551 --> 00:03:29.610
00:03:29.610 --> 00:03:43.485
00:03:44.539 --> 00:03:45.181
00:03:45.181 --> 00:03:53.769
00:03:53.769 --> 00:03:57.570
00:03:57.570 --> 00:04:05.849
00:04:05.849 --> 00:04:23.069
00:04:23.069 --> 00:04:34.148
00:04:34.148 --> 00:04:36.586
00:04:37.629 --> 00:04:38.360
00:04:38.360 --> 00:04:47.523
00:04:47.523 --> 00:04:54.845
00:04:54.845 --> 00:05:05.809
00:05:05.809 --> 00:05:07.302
00:05:07.302 --> 00:05:11.112
00:05:11.680 --> 00:05:14.384
00:05:14.384 --> 00:05:26.903
00:05:26.903 --> 00:05:32.747
00:05:32.747 --> 00:05:46.492
00:05:46.492 --> 00:05:51.125
00:05:51.125 --> 00:05:52.930
00:05:52.930 --> 00:06:00.132
00:06:00.620 --> 00:06:03.930
00:06:04.901 --> 00:06:13.370
00:06:13.370 --> 00:06:30.550
00:06:30.550 --> 00:06:31.072
00:06:31.072 --> 00:06:34.831
00:06:34.831 --> 00:06:42.394
00:06:42.394 --> 00:06:46.781
00:06:46.781 --> 00:06:47.362
00:06:47.362 --> 00:06:48.024
00:06:48.024 --> 00:06:56.249
00:06:56.249 --> 00:07:04.512
00:07:04.512 --> 00:07:16.254
00:07:17.922 --> 00:07:18.987
00:07:19.439 --> 00:07:19.942
00:07:19.942 --> 00:07:21.988
00:07:22.680 --> 00:07:30.651
00:07:30.651 --> 00:07:37.439
00:07:37.439 --> 00:07:58.749
00:07:58.749 --> 00:08:02.223
00:08:02.223 --> 00:08:03.826
00:08:03.826 --> 00:08:04.648
00:08:05.149 --> 00:08:09.204
00:08:09.204 --> 00:08:15.605
00:08:15.605 --> 00:08:27.483
00:08:27.483 --> 00:08:28.365
00:08:28.365 --> 00:08:32.561
00:08:32.561 --> 00:08:45.860
00:08:45.860 --> 00:08:58.851
00:08:58.851 --> 00:09:10.174
00:09:10.174 --> 00:09:14.712
00:09:14.712 --> 00:09:16.259
00:09:16.259 --> 00:09:18.423
00:09:18.423 --> 00:09:23.655
00:09:23.655 --> 00:09:25.659
00:09:25.659 --> 00:09:28.565
00:09:28.565 --> 00:09:30.989
00:09:30.989 --> 00:09:38.455
00:09:38.455 --> 00:09:39.778
00:09:39.778 --> 00:09:42.284
00:09:42.284 --> 00:09:50.730
00:09:50.730 --> 00:09:53.205
00:09:53.205 --> 00:09:54.410
00:09:54.410 --> 00:10:03.004
00:10:03.004 --> 00:10:14.561
00:10:14.561 --> 00:10:19.988
00:10:20.788 --> 00:10:30.351
00:10:30.351 --> 00:10:33.524
00:10:33.524 --> 00:10:36.005
00:10:36.005 --> 00:10:41.690
00:10:42.510 --> 00:10:57.822
00:10:58.581 --> 00:11:09.113
00:11:09.113 --> 00:11:24.535
00:11:24.535 --> 00:11:30.258
00:11:30.258 --> 00:11:34.653
00:11:34.653 --> 00:11:47.275
00:11:47.275 --> 00:11:55.359
00:11:55.359 --> 00:11:57.231
00:11:58.304 --> 00:12:03.027
00:12:03.027 --> 00:12:06.355
00:12:06.355 --> 00:12:26.791
00:12:26.791 --> 00:12:36.773
00:12:37.215 --> 00:12:41.191
00:12:41.191 --> 00:12:44.969
00:12:44.969 --> 00:12:50.850
00:12:50.850 --> 00:12:53.687
00:12:53.687 --> 00:12:55.893
00:12:55.893 --> 00:12:58.179
00:12:58.179 --> 00:13:06.333
00:13:06.333 --> 00:13:10.245
00:13:10.245 --> 00:13:11.653
00:13:12.841 --> 00:13:13.345
00:13:13.345 --> 00:13:18.639
00:13:18.639 --> 00:13:27.274
00:13:27.274 --> 00:13:28.630
00:13:28.630 --> 00:13:49.197
00:13:49.197 --> 00:13:51.932
00:13:51.932 --> 00:13:59.775
00:13:59.775 --> 00:14:02.714
00:14:02.714 --> 00:14:11.971
00:14:11.971 --> 00:14:21.615
00:14:22.524 --> 00:14:24.188
00:14:24.188 --> 00:14:35.711
00:14:35.711 --> 00:14:39.469
00:14:39.469 --> 00:14:44.549
00:14:44.549 --> 00:14:50.409
00:14:50.951 --> 00:14:55.298
00:14:55.298 --> 00:14:57.008
00:14:57.008 --> 00:14:59.414
00:14:59.414 --> 00:15:00.336
00:15:00.825 --> 00:15:07.967
00:15:08.249 --> 00:15:10.254
00:15:10.254 --> 00:15:18.375
00:15:24.264 --> 00:15:46.272
00:15:46.272 --> 00:15:53.975
00:15:54.828 --> 00:15:57.672
00:15:57.672 --> 00:16:12.753
00:16:12.753 --> 00:16:19.298
00:16:19.298 --> 00:16:22.975
00:16:22.975 --> 00:16:27.437
00:16:27.437 --> 00:16:32.110
00:16:32.913 --> 00:16:35.018
00:16:35.018 --> 00:16:39.947
00:16:39.947 --> 00:16:42.023
00:16:42.023 --> 00:16:48.708
00:16:48.708 --> 00:16:54.167
00:16:54.167 --> 00:16:56.092
00:16:56.092 --> 00:16:59.240
00:16:59.240 --> 00:17:01.870
00:17:01.870 --> 00:17:04.605
00:17:04.605 --> 00:17:10.106
00:17:10.106 --> 00:17:13.413
00:17:13.413 --> 00:17:19.596
00:17:20.765 --> 00:17:24.868
00:17:24.868 --> 00:17:29.808
00:17:29.808 --> 00:17:32.576
00:17:32.576 --> 00:17:36.349
00:17:36.349 --> 00:17:38.374
00:17:38.374 --> 00:17:43.153
00:17:43.153 --> 00:17:44.388
00:17:44.388 --> 00:17:45.452
00:17:46.085 --> 00:17:48.253
00:17:48.253 --> 00:17:55.976
00:17:55.976 --> 00:17:57.547
00:17:57.547 --> 00:18:00.451
00:18:00.451 --> 00:18:01.994
00:18:01.994 --> 00:18:12.326
00:18:12.346 --> 00:18:19.732
00:18:19.732 --> 00:18:22.452
00:18:22.452 --> 00:18:26.674
00:18:26.674 --> 00:18:30.614
00:18:30.614 --> 00:18:32.351
00:18:32.351 --> 00:18:32.825
00:18:32.825 --> 00:18:38.178
00:18:38.178 --> 00:18:41.574
00:18:41.574 --> 00:18:48.724
00:18:48.724 --> 00:18:52.554
00:18:52.554 --> 00:18:55.146
00:18:55.146 --> 00:19:07.338
00:19:08.025 --> 00:19:13.530
00:19:13.530 --> 00:19:14.815
00:19:16.365 --> 00:19:17.911
00:19:17.911 --> 00:19:23.896
00:19:23.896 --> 00:19:27.974
00:19:27.974 --> 00:19:36.232
00:19:36.232 --> 00:19:40.622
00:19:40.622 --> 00:19:46.951
00:19:48.317 --> 00:19:54.414
00:19:54.414 --> 00:19:56.663
00:19:56.663 --> 00:20:12.736
00:20:12.736 --> 00:20:18.428
00:20:18.428 --> 00:20:22.505
00:20:23.796 --> 00:20:36.521
00:20:36.521 --> 00:20:42.244
00:20:42.244 --> 00:20:45.220
00:20:46.221 --> 00:20:46.623
00:20:46.623 --> 00:20:49.436
00:20:49.436 --> 00:21:04.675
00:21:04.675 --> 00:21:11.719
00:21:11.719 --> 00:21:16.127
00:21:16.127 --> 00:21:20.586
00:21:20.586 --> 00:21:23.564
00:21:23.564 --> 00:21:30.645
00:21:31.315 --> 00:21:32.380
00:21:32.380 --> 00:21:34.701
00:21:34.701 --> 00:21:37.442
00:21:38.204 --> 00:21:41.284
00:21:41.284 --> 00:21:46.207
00:21:46.207 --> 00:21:54.638
00:21:54.638 --> 00:22:00.155
00:22:00.636 --> 00:22:03.164
00:22:03.184 --> 00:22:21.444
00:22:21.444 --> 00:22:32.626
00:22:32.626 --> 00:22:35.833
00:22:35.833 --> 00:22:45.688
00:22:45.688 --> 00:23:00.801
00:23:01.403 --> 00:23:12.761
00:23:12.761 --> 00:23:31.835
00:23:31.835 --> 00:23:39.747
00:23:39.747 --> 00:23:44.006
00:23:44.335 --> 00:23:54.282
00:23:54.282 --> 00:23:59.931
00:23:59.931 --> 00:24:07.450
00:24:07.450 --> 00:24:10.038
00:24:10.880 --> 00:24:14.087
00:24:14.087 --> 00:24:21.708
00:24:21.788 --> 00:24:22.710
00:24:23.436 --> 00:24:24.761
00:24:25.415 --> 00:24:26.862
00:24:26.862 --> 00:24:30.162
00:24:33.035 --> 00:24:34.221
00:24:34.260 --> 00:24:37.074