WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=FWMJQDH3iK0

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/FWMJQDH3iK0.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.560 --> 00:00:18.200
So, I hope everybody had a great lunch and you got to check out some of the amazing demos that we have.

s3
00:00:18.200 --> 00:00:26.320
Uh we're going to begin the panel the first panel of the afternoon here where we're going to be talking about of course the engines that are actually powering the stuff that

s4
00:00:26.320 --> 00:00:38.400
you know could remotely be used for things like local sovereign any kind of ownership over your own artificial intelligence and of course the engine powering those in addition to the hardware is the models themselves.

s5
00:00:38.400 --> 00:00:41.360
And so for this panel we have excellent guests.

s6
00:00:41.360 --> 00:00:45.120
We have Vincent who's the CEO and founder of Prime Intellect.

s7
00:00:45.120 --> 00:00:53.520
We've got Lucas the CTO of RCAI and we've got Chris who is the senior product research engineer on the Neumotron family of models at Nvidia.

s8
00:00:53.520 --> 00:00:59.520
Now, what's really cool about working in this industry is really cool companies like this we all get to work together.

s9
00:00:59.520 --> 00:01:08.080
And so this is one panel where we all directly get to work together on both models infrastructure some of the ways that we think that the direction of the industry should go

s10
00:01:08.080 --> 00:01:20.920
and each of us kind of play a different role in that stack but I want to leave it to you guys to to introduce yourself and be able to talk about sort of the the charter that you see the problem of the stack that you guys are working on.

s11
00:01:20.920 --> 00:01:21.360
Awesome.

s12
00:01:21.360 --> 00:01:21.960
Should I kick it off?

s13
00:01:21.960 --> 00:01:22.560
Kick it off.

s14
00:01:22.560 --> 00:01:32.560
Yeah, so I'm I'm Vincent as you mentioned and then really the goal with Prime Intellect from the beginning was like to ensure that basically frontier intelligence will be open and accessible

s15
00:01:32.560 --> 00:01:36.000
not just the models but also the full stack to to train the models.

s16
00:01:36.000 --> 00:01:46.280
So kind of like this was like our motivation from the beginning and we we've yeah like worked also together with a lot of gentlemen on the on stage like on the one side is like

s17
00:01:46.280 --> 00:01:50.920
we we work with folks like Lucas and RC to help them train frontier open models.

s18
00:01:50.920 --> 00:01:56.840
We we we help also um like Nvidia on the Neumotron coalition help their uh open models.

s19
00:01:56.840 --> 00:02:05.560
And I think like I'm actually um think both like an Eleutheron and and Trinity might be the best like two open models right now outside of China.

s20
00:02:05.560 --> 00:02:11.800
So, I think it's actually uh like we need to fact-check that, you know, but this is actually from my I think they they might be.

s21
00:02:11.800 --> 00:02:15.680
It's our our marketing says they yeah.

s22
00:02:15.680 --> 00:02:18.000
But yeah, so so that's the high level.

s23
00:02:18.000 --> 00:02:20.040
Um my name's Lucas Atkins.

s24
00:02:20.040 --> 00:02:22.840
I'm happy to be here and and thank you for joining.

s25
00:02:22.840 --> 00:02:33.959
Um very similar to Vincent, RC was uh you know, founded with the idea of um domain-specific owned models are are are going to be needed.

s26
00:02:33.959 --> 00:02:40.680
Um you know, we were founded early 2023 uh jumping on the custom model uh train quite early.

s27
00:02:40.680 --> 00:02:47.680
Um you know, you have all these people who are excited about AI and all the things these new generation of LLMs can do.

s28
00:02:47.680 --> 00:02:59.280
Um but they're using these monolithic very expensive closed APIs uh for at the time and still like very narrow tasks that don't require uh you know, at the time it was $100

s29
00:02:59.280 --> 00:03:01.160
per million tokens out.

s30
00:03:01.160 --> 00:03:07.160
Um and through doing that, we were building on top of open models and we were releasing a lot of our tooling in the open.

s31
00:03:07.160 --> 00:03:17.239
Um and uh we noticed that in the United States um and in the West, you know, in general, we were starting to lose uh leadership in the open model space.

s32
00:03:17.239 --> 00:03:19.519
A lot of it was coming out of China and that's amazing.

s33
00:03:19.519 --> 00:03:20.400
I love those models.

s34
00:03:20.400 --> 00:03:21.720
We learn a lot from them.

s35
00:03:21.720 --> 00:03:24.360
Uh we're close with a lot of the people building those.

s36
00:03:24.360 --> 00:03:35.920
But when you're working with large enterprises and companies and um geopolitics gets involved, whether you like it or not, you have people that become concerned about where those models are coming from.

s37
00:03:35.920 --> 00:03:46.320
And we uh decided that, you know, we had a a good group of people and we had a good group of partners like Nvidia and like Prime Intellect where we could probably try to pre-train ourselves.

s38
00:03:46.320 --> 00:03:47.600
Uh so, last year we did that.

s39
00:03:47.600 --> 00:03:56.000
We kind of uh reoriented the entire company towards let's figure out how to pre-train a you know a 400 billion parameter model in 6 months.

s40
00:03:56.000 --> 00:04:05.120
And a lot of people said it was impossible and by in many ways it was, but we figured it out and now we are an open model lab working with our wonderful partners

s41
00:04:05.120 --> 00:04:17.440
and our customers to build western open models that are permissive and you can own those and customize them or run them wherever you want and and that's kind of where we're at right now.

s42
00:04:17.440 --> 00:04:19.079
So thanks for having me.

s43
00:04:19.079 --> 00:04:19.560
Yeah.

s44
00:04:19.560 --> 00:04:21.200
And so I'm Chris Alexa.

s45
00:04:21.200 --> 00:04:26.640
I work at Nvidia as a product research engineer and I support the Neumotron family of models.

s46
00:04:26.640 --> 00:04:33.200
I think it's you know we we've talked a lot about why we do Neumotron but just to say it a few more times.

s47
00:04:33.320 --> 00:04:41.480
You know AI should be open open as in weights, data, training methodology, training frameworks.

s48
00:04:41.480 --> 00:04:48.800
Really respect a lot of the work that uh the the the the two other people up here do because they they believe that very strongly as well.

s49
00:04:48.800 --> 00:04:52.840
But the Neumotron family of models is focused on being as open as humanly possible.

s50
00:04:52.840 --> 00:05:07.960
So we we have this understanding or belief that in order for AI to continue to grow and be useful to everybody, it has to be done in the open so that we can build off of each other, we can compound on each other.

s51
00:05:07.960 --> 00:05:20.000
And part of what we do because team green, this is always true, is we we think that the the rate that you can squeeze tokens out of models is very important.

s52
00:05:20.000 --> 00:05:30.400
So we kind of have this mantra that like faster models are smarter models and so a lot of the decisions we make when designing a model like Neumotron is

s53
00:05:30.400 --> 00:05:33.600
built around how fast can we make it go.

s54
00:05:33.600 --> 00:05:48.800
As especially you are going to see in the next however many months local AI take off, we we need to make sure that models are well supported on uh hardware that doesn't just exist in massive buildings,

s55
00:05:48.800 --> 00:05:51.040
you know, thousands of kilometers away from you.

s56
00:05:51.040 --> 00:05:56.920
Uh and so that's uh you know, for AI to be very useful, it should be quick uh and and open.

s57
00:05:56.920 --> 00:05:59.720
So, that's kind of the the vibe of Neumotron.

s58
00:05:59.720 --> 00:06:02.520
Who makes those buildings with the massive processors?

s59
00:06:02.520 --> 00:06:09.440
Oh, that's uh a lot of excellent people in the world that they use a lot of excellent hardware from a pretty cool company.

s60
00:06:09.440 --> 00:06:10.880
Yeah, I heard I heard anyway.

s61
00:06:10.880 --> 00:06:11.480
Yeah.

s62
00:06:11.480 --> 00:06:14.160
And I'm Carter Abdallah, I'll be your moderator for today.

s63
00:06:14.160 --> 00:06:23.960
Uh something that you know, we all kind of talked about is is this you know, building on top of each other, learning from others, whether it is people, you know, across the the big pond of the Pacific Ocean from us.

s64
00:06:23.960 --> 00:06:30.360
Um but really it is kind of like a collaborative sort of research effort and I imagine that a lot of the people here in this room share that sentiment.

s65
00:06:30.360 --> 00:06:39.720
But as it was brought up during the, you know, inaugural panel this morning in the state of the union, there there is a growing sentiment um potentially on the other side that that paints

s66
00:06:39.720 --> 00:06:46.080
uh open source to be something that is actually more chaotic, that there there's less trust involved.

s67
00:06:46.080 --> 00:06:55.520
And I think trust ultimately as Lucas, you and I were talking about before, um depending on who the party is and depending on what lens you're looking at at it from, I think it kind of means different things.

s68
00:06:55.520 --> 00:07:05.560
But ultimately from from the the end consumer, the somebody who's using this intelligence, um or somebody who's you know, more of a business and is actually customizing something to maybe monetize

s69
00:07:05.560 --> 00:07:07.480
tokens in the in their business.

s70
00:07:07.480 --> 00:07:22.400
Um can you comment, we'll start with you, Lucas, a bit on how open source and open source models are actually key to building that trust so that when these people walk out of this room and somebody does come at them with that other angle, they can they can sort of steel man this side.

s71
00:07:22.400 --> 00:07:23.000
Certainly.

s72
00:07:23.000 --> 00:07:29.840
You can weaponize any term uh and certainly trust is is has been weaponized, that word.

s73
00:07:29.840 --> 00:07:36.520
Um and the reason I say that is because it means something in based on the context and with your speech you're speaking about it.

s74
00:07:36.520 --> 00:07:44.400
Uh often in AI, people like to conflate trust with safety um and those are not the same thing.

s75
00:07:44.400 --> 00:07:47.240
Uh, and I'm happy to speak on safety, uh, you know, later on.

s76
00:07:47.240 --> 00:08:00.560
But when it comes to trust, um, I I think that, you know, you hear a lot from closed model providers or uh, politicians or people out in the space who are advocates for or

s77
00:08:00.560 --> 00:08:05.760
uh, against open source that you can't trust these open models cuz you don't know what went into them.

s78
00:08:05.760 --> 00:08:09.400
Well, the same is true for these closed models, uh, even more so.

s79
00:08:09.400 --> 00:08:17.280
Uh, the benefit of uh, uh, of of open models is that uh, we can very easily validate what is inside of them.

s80
00:08:17.280 --> 00:08:24.400
Uh, they are you can there is a whole bunch of files with a whole bunch of matrices in there and you can view them and you can see the code

s81
00:08:24.400 --> 00:08:25.800
that is running these models.

s82
00:08:25.800 --> 00:08:32.080
You have implementations from Prime RL, vLLM, SG Lang, the provider themselves.

s83
00:08:32.080 --> 00:08:34.240
These models are inherently trustworthy.

s84
00:08:34.240 --> 00:08:41.360
You know much more about what's going on when you hit and talk to these models than you ever will what's going on when you hit an arbitrary API.

s85
00:08:41.360 --> 00:08:52.080
Now, that being said, um, certainly there is fear that people can um, reduce uh, you know, the you can't trust that these models are writing safe code.

s86
00:08:52.080 --> 00:08:55.000
Well, again, that is the same thing with any model.

s87
00:08:55.000 --> 00:09:07.120
You need to you need to use your uh, your judgment and you need to make sure that you have the proper um, you know, safeguards in place and you're viewing the outputs of these models as the outputs of

s88
00:09:07.120 --> 00:09:13.120
uh, an inherently random system that we are working very, very hard to make less and less random.

s89
00:09:13.120 --> 00:09:17.480
I think that uh, a telling thing is a lot of people said, "Well, you can't trust Chinese models.

s90
00:09:17.480 --> 00:09:18.680
You can't trust Chinese models.

s91
00:09:18.680 --> 00:09:20.160
You can't trust Chinese models."

s92
00:09:20.160 --> 00:09:23.920
That was often uh, for the last few years meant you can't trust open models.

s93
00:09:23.920 --> 00:09:34.640
Well, as soon as uh, Anthropic had to put Fable away and people realized that, "Oh, our access to these frontier systems might not be universal anymore.

s94
00:09:34.640 --> 00:09:37.280
Uh, there's probably going to be a lot of checks and balances.

s95
00:09:37.280 --> 00:09:47.440
You had a tremendous number of enterprises and developers and companies start going to these new Chinese models because they could trust that they would always have access to them.

s96
00:09:47.440 --> 00:09:59.800
Um and so when when it comes out of trust in the way I view that word as it relates to open models is do I know that what I am running and can I be as sure as possible that when I send something to this model that I am going to get the output that I expect?

s97
00:09:59.800 --> 00:10:13.560
Um and the only way uh currently to be 100% sure that what you are getting is what you were expecting is by hitting an open model either that you are running yourselves or you're working with a partner like Prim Indelec or RC or Nvidia

s98
00:10:13.560 --> 00:10:14.160
to validate.

s99
00:10:14.160 --> 00:10:16.840
So that's my take on the word trust.

s100
00:10:16.840 --> 00:10:25.160
I think too something you mentioned is like we don't get to know a lot about the data that goes into these models and that's something that I'm really happy

s101
00:10:25.160 --> 00:10:40.240
you know that that we're trying to do which is not it's not you know the incentives don't exist for everyone to do this right so it's not something that I think is mandatory or should be mandatory thanks to the things that Lucas mentioned which is that it's rather straightforward to validate what data did go into a model

s102
00:10:40.240 --> 00:10:47.920
without seeing the data sources originally but I'm happy that Nvidia continues to release data sets along with our models.

s103
00:10:47.920 --> 00:11:00.760
Release environments along with our models to make sure that even even if you can't go through the work of determining what went into the model which you can do with the the weights alone for the most part you have like a spreadsheet

s104
00:11:00.760 --> 00:11:12.800
you can look at that says here's you know a couple trillion tokens of this data set a couple trillion tokens here and I think that that helps to educate people on why

s105
00:11:12.800 --> 00:11:20.080
it's much easier to trust open models than than models that we we don't get access to any of that.

s106
00:11:20.080 --> 00:11:22.280
helps people see what that data looks like.

s107
00:11:22.280 --> 00:11:22.760
Yeah.

s108
00:11:22.760 --> 00:11:29.280
You know if you don't have someone releasing it openly when someone says data is going in I mean data can take many different shapes.

s109
00:11:29.280 --> 00:11:34.360
You can but you can go to Hugging Face you can go to Nvidia or you can go to Prime Intellect, or RC's uh HuggingFace.

s110
00:11:34.360 --> 00:11:40.640
You can look under our data sets, and you can see exactly what that looks like, um and that can help you uh inform your priors on it.

s111
00:11:40.640 --> 00:11:44.560
Yeah, I think that trust also um you know, there's some angle of of a reputation.

s112
00:11:44.560 --> 00:11:46.760
Do I believe that your your intentions are pure?

s113
00:11:46.760 --> 00:11:57.560
And I think that a lot of people, again, in this room believe that intelligence is kind of this next layer of almost, you know, infrastructure for for us to progress as a species, and I believe that everybody

s114
00:11:57.560 --> 00:11:58.520
should have intelligence.

s115
00:11:58.520 --> 00:12:08.120
So, uh on on that front, I want to uh hand it over to Vincent because uh you kind of have this um almost like founding thesis that you this this stack should be the open, right?

s116
00:12:08.120 --> 00:12:09.680
The open superintelligence stack.

s117
00:12:09.680 --> 00:12:12.160
You want everybody to have a lot more intelligence.

s118
00:12:12.160 --> 00:12:23.280
Um can you talk about how uh the this is kind of moving into the era of control, but uh beyond just data sets, how important it is to have the knobs and dials

s119
00:12:23.280 --> 00:12:28.720
of the of this industry also be uh available in an open way for people who are building this?

s120
00:12:28.720 --> 00:12:36.600
Yeah, like I I think it's a really important point is to set um so I can say it like be able to take those open models, like customize them, be able to like build on top of them.

s121
00:12:36.600 --> 00:12:43.083
And I think like all the different components are going to it, like especially from like the pre-training to mid-training to post-training, I think

s122
00:12:43.083 --> 00:12:43.240
[music]

s123
00:12:43.240 --> 00:12:44.600
like need to be more accessible, right?

s124
00:12:44.600 --> 00:12:50.080
Like so more people can also like take those uh amazing models and like make them work for their specific use cases.

s125
00:12:50.080 --> 00:12:57.000
So, I think when we started like we we also took a look at the whole stack that was out there and and tried to figure out like what is missing

s126
00:12:57.000 --> 00:13:00.600
for ourselves to train open models and for like helping our partners to do so.

s127
00:13:00.600 --> 00:13:02.800
And a lot of this was around the RL and post-training stack.

s128
00:13:02.800 --> 00:13:14.240
So, we basically went deep into building out like a lot of infra around that, like around our environment, Evals, around like making it much more accessible to do post-training also because it's like the most economically

s129
00:13:14.240 --> 00:13:24.760
viable way to maybe like customize those models, to take an open model, and um to have like a specific Eval environment, and the um specific domain and dimension that you want to improve it on.

s130
00:13:24.760 --> 00:13:31.680
And this This of like what we're are doing with Prime Intel now, is like enabling people to post train specialized agentic models.

s131
00:13:31.680 --> 00:13:40.880
Um so, being able to take models like um Trinity, for example, from RC or Nematron or others and and specialize them, post train them for the use cases that ultimately enterprises care about.

s132
00:13:40.880 --> 00:13:49.960
So, good example of this was like a company like, for example, Ramp or Saber to like take an open model and like specialize it to automate finance within like a week or two

s133
00:13:49.960 --> 00:13:53.800
to get like better performance than like Opus at a fraction of the cost of Haiku.

s134
00:13:53.800 --> 00:14:02.320
And I think really this Pareto frontier of like being able to create these specialized models that are much better than the frontier, but also faster, cheaper.

s135
00:14:02.320 --> 00:14:05.600
Um I think it's like a key thing enterprises care about increasingly.

s136
00:14:05.600 --> 00:14:08.520
It's really like just making it work for their use cases, basically.

s137
00:14:08.520 --> 00:14:14.960
If you go back to trust, it's how you can make your CFO trust you by knowing exactly how much something's going to cost all the time.

s138
00:14:14.960 --> 00:14:27.960
Um that's That is increasingly becoming very important is uh you hear a lot, you know, all these companies have unbelievably large token spend and um they're having to cut back on their Opus usage because they burned through it all in a couple months.

s139
00:14:27.960 --> 00:14:34.880
Um and that is going to continue to be a problem because yes, the cost of an individual token has come down drastically.

s140
00:14:34.880 --> 00:14:47.080
You can look at it, you know, the difference between GPT-4 when it first launched and GPT-5.5 is is is much, much cheaper per token, but at the same time the amount of tokens in an individual session has gone up exponentially as well.

s141
00:14:47.080 --> 00:14:53.000
So, we're kind of um we're spending more uh on a on a total session.

s142
00:14:53.000 --> 00:15:05.560
And so, the ability to uh bring in-house or or or at least work with partners to ensure that you are controlling your cost and you're not at the whims of uh when a company releases a newer model

s143
00:15:05.560 --> 00:15:07.760
uh that might be better, but also more expensive.

s144
00:15:07.760 --> 00:15:09.760
They might deprecate a model.

s145
00:15:09.760 --> 00:15:16.280
Um owning that and being sure that, you know, same way is what you what out input goes in, you know what output's going to come out.

s146
00:15:16.280 --> 00:15:21.320
In the same way, um when it you know, an input goes in, how much it's going to cost.

s147
00:15:21.320 --> 00:15:23.560
Uh having assurance on that's is important, too.

s148
00:15:23.560 --> 00:15:31.480
Yeah, and maybe like one thing to add to this is like almost like I like this new term of like instead of speaking about token maxing, you know, speaking more about like the outcome maxing of like

s149
00:15:31.480 --> 00:15:31.840
Yeah.

s150
00:15:31.840 --> 00:15:42.440
Ultimately, it's like you you want to have like more than a dollar worth of value come out of like a dollar of input and I think this is sort of like Jevons paradox of like if you can create more value

s151
00:15:42.440 --> 00:15:45.040
for like your flop for your GPU, basically.

s152
00:15:45.040 --> 00:15:49.000
I think this is sort of like how you get like the most adoption also of like agentic models.

s153
00:15:49.000 --> 00:15:52.360
Like if they can like be able to create as much value as possible.

s154
00:15:52.360 --> 00:15:57.240
I think the cheaper those models get, the more usage they'll get like for those specific use cases.

s155
00:15:57.240 --> 00:15:58.080
It's funny you say that.

s156
00:15:58.080 --> 00:16:12.200
I have a and I think a lot of us in this room, but especially on this panel believe this to be so that the most meaningful AI applications the next couple years, even this year, are the ones where the harness and the model and the product, they all kind of blend together.

s157
00:16:12.200 --> 00:16:22.360
If you think back to at least for me, the first like truly game-changing agentic experience I had was when Deep Research from OpenAI.

s158
00:16:22.360 --> 00:16:36.920
And that was because they spent a tremendous amount of time doing reinforcement learning on O3 with test time compute to do these longer running research tasks that people had tried previously, but they were kind of just doing a for loop over search.

s159
00:16:36.920 --> 00:16:40.120
Whereas I kept coming back to to Deep Research.

s160
00:16:40.120 --> 00:16:49.240
And you know, you saw for a very long time that OpenAI and Anthropic and Google, when they'd release a new product, they'd release a custom version of their model for that product.

s161
00:16:49.240 --> 00:16:57.200
And if they're doing that, if their off-the-shelf GPT-5 isn't good enough for, you know, their Atlas web browser, why should it be good enough for our apps?

s162
00:16:57.200 --> 00:17:09.120
And that's why I appreciate the work that, you know, Vincent and Nvidia are doing for for giving people the tools to customize their own model and allowing us to focus on how we get a good model to start from.

s163
00:17:09.120 --> 00:17:19.839
So it really is you know, it's it's extremely important as you look at developing applications and experiences over over next few years that you're taking into account that you can

s164
00:17:19.839 --> 00:17:24.560
uh make the model do something that maybe your harness isn't fully able to do alone.

s165
00:17:24.560 --> 00:17:28.640
Well, that's something I think that's really important to just like reiterate, right?

s166
00:17:28.640 --> 00:17:32.040
I mean, like Nemotron's great, I love it.

s167
00:17:32.040 --> 00:17:33.800
Trinity's great, I love it.

s168
00:17:33.800 --> 00:17:39.600
Like we design a model that's supposed to be as good as it can be across a number of harnesses, right?

s169
00:17:39.600 --> 00:17:41.440
You can see this in the technical report.

s170
00:17:41.440 --> 00:17:48.880
The idea is like we want the the model to work as well as it can in Pi compared to you know Hermes compared whatever you're using, right?

s171
00:17:48.880 --> 00:17:52.120
But like you're you're not using all of these tools at once.

s172
00:17:52.120 --> 00:17:53.760
You're using one of these tools.

s173
00:17:53.760 --> 00:18:03.080
And so when you have open models, you can do things like news research can create a post train of whatever model for their harness, right?

s174
00:18:03.080 --> 00:18:05.560
That you know will be extra good.

s175
00:18:05.560 --> 00:18:14.200
And you know, this this this thing from from from you know, I I can't remember who who who originally wrote it, but this idea of like the mismanaged genius, right?

s176
00:18:14.200 --> 00:18:25.000
We're we're leaving a lot of like a lot of important capability on the table because we're just not we're not fitting the models into the harness, right?

s177
00:18:25.000 --> 00:18:27.680
You can do a bunch of stuff with closed models.

s178
00:18:27.680 --> 00:18:32.120
Like you can change your prompts and your skills and all kinds of other nido things, right?

s179
00:18:32.120 --> 00:18:40.040
But nothing will let you get the the the level of customization or customizability that you can achieve with open models.

s180
00:18:40.040 --> 00:18:52.600
And I think that is something that is going to become increasingly and increasingly more important, especially thanks to folks like the others on the panel where you know, I can just straight drop like my favorite coding and

s181
00:18:52.600 --> 00:19:01.400
you know, an agent environment spin up the CLI and suddenly my my model feels way better with very little effort, right?

s182
00:19:01.400 --> 00:19:10.520
Like that is that is something that is already at our fingertips and it is only going to get easier and easier as as time goes on.

s183
00:19:10.520 --> 00:19:11.280
Yeah.

s184
00:19:11.280 --> 00:19:18.000
And it's maybe also the most concrete like info for the builders in the audience like call to action of like if you kind of want to build the next like

s185
00:19:18.000 --> 00:19:26.480
cloud code, the next like cursor or perplexity, I think the easiest way to get started is like take the best open model like and and then post rate on your harness like that you care about, right?

s186
00:19:26.480 --> 00:19:30.400
Like basically like create a product that is like truly AI native, right?

s187
00:19:30.400 --> 00:19:36.440
And I think this is sort of like I think one of the most exciting like unlocks for builders um like out here.

s188
00:19:36.440 --> 00:19:48.800
That's a big thing about the the framing of control is is um similar to like, you know, when when the cloud um explosion started to happen in the in the, you know, the late uh 2000s,

s189
00:19:48.800 --> 00:20:00.520
early 2010s and the um social media world kind of took off and apps became extremely popular and more, you know, cloud-based and and, you know, managed by these these bigger companies,

s190
00:20:00.520 --> 00:20:09.840
uh you know, you the data that they were collecting from you, whether anonymized or not, was how they were monetizing their platform through ads or or or or in other ways.

s191
00:20:09.840 --> 00:20:18.400
Uh and a very similar thing has always been happening, but I think it's becoming clear to people in in the space is that the the data that people get from from you using these models

s192
00:20:18.400 --> 00:20:29.080
is how these companies largely make their models better, whether it's through actually training on that data or um by using it as a signal for what data they they go out and find or generate to train.

s193
00:20:29.080 --> 00:20:42.280
And uh with closed models, there are terms of service that keep you from being able to um well, I could get into a a discussion about what terms of service is an agreement between you and the provider, it's not a legal

s194
00:20:42.280 --> 00:20:53.560
uh anyway, but you you you shouldn't be training on a, you know, a Claude Opus output or a Fable output or a GPT-5 output, and they do a lot to try to obfuscate to to make that not great for you.

s195
00:20:53.560 --> 00:20:56.840
If you're using an open model, you can save all of those traces.

s196
00:20:56.840 --> 00:21:04.400
All of those traces of you using it inside of your harness uh that will allow you over time to if you say, "Hey, I want to go train a custom model,"

s197
00:21:04.400 --> 00:21:15.200
you can take all of that, and again, either use it to directly do like fine-tuning on a smaller model, so you're not spending as much, or to have a model help you find signals, so that you can go out and use Verifiers

s198
00:21:15.200 --> 00:21:20.880
or Nemo RL or Nemo Gym to create these environments, so that you can hill climb and make your models better.

s199
00:21:20.880 --> 00:21:28.440
So, um as much as using open models is like owning your stack, owning your intelligence, it's also owning your outputs, right?

s200
00:21:28.440 --> 00:21:29.080
Owning your data.

s201
00:21:29.080 --> 00:21:30.920
That's going to be extremely important, too.

s202
00:21:30.920 --> 00:21:34.480
I do want to I do want to plug the license for a second.

s203
00:21:34.480 --> 00:21:35.604
So,

s204
00:21:35.604 --> 00:21:36.160
[laughter]

s205
00:21:36.160 --> 00:21:49.520
uh so, AI is very different than traditional software, uh which is why recently uh Nemotron, as well as uh Trinity, I know, uh has adopted the Open MDW model

s206
00:21:49.520 --> 00:21:52.320
uh data weights uh license.

s207
00:21:52.320 --> 00:21:59.240
Uh the idea is like, we we need a way to really make it very clear in the license that you can use the outputs to produce a model.

s208
00:21:59.240 --> 00:22:02.680
You can use the outputs to train uh right?

s209
00:22:02.680 --> 00:22:08.680
All of these TOSs and stuff like that that that that that have language that's meant to dissuade you to do that.

s210
00:22:08.680 --> 00:22:17.320
Uh we wanted to make sure there's a license that exists that not encourages you, but makes it crystal clear that it is it is permitted it is permissible.

s211
00:22:17.320 --> 00:22:22.040
Uh and and I think, you know, the licenses maturing, right?

s212
00:22:22.040 --> 00:22:30.160
To fit the use case better should be extremely positive signal uh for for the way that the ecosystem is thinking about open models.

s213
00:22:30.160 --> 00:22:32.640
Uh to to the fact where even even the lawyers are on board.

s214
00:22:32.640 --> 00:22:34.555
know how much lawyers cost?

s215
00:22:34.555 --> 00:22:35.000
[laughter]

s216
00:22:35.000 --> 00:22:35.800
Yeah.

s217
00:22:35.800 --> 00:22:52.520
Yeah, it's uh as we move on to the, you know, kind of the third topic which here is of course optimization, um something that you know, we've implicitly said, but haven't said it quite explicitly yet is uh effectively that I I think that for a lot of people there's this preconceived notion that when you're deciding to use

s218
00:22:52.520 --> 00:23:03.880
an open model for whatever the use case, um there there are the tradeoffs that come in the form of performance at the benefit of getting things like, you know, maybe data sovereignty and so forth.

s219
00:23:03.880 --> 00:23:15.280
Um, but what we are now talking about is that with the with the right customization and optimization, depending on the the use case that you're and the harness that you're applying it to, you can actually exceed

s220
00:23:15.280 --> 00:23:21.720
and build the model against the tool to to get better performance than even frontier models.

s221
00:23:21.720 --> 00:23:30.560
Um, I'd love to hear a little bit more about the cuz I think that uh another thing that we would probably agree on is that the the current level of intelligence

s222
00:23:30.560 --> 00:23:35.680
um, already has so much uh left to diffuse into society.

s223
00:23:35.680 --> 00:23:41.040
And so, where are those areas where that diffusion is is is happening in the specific industries?

s224
00:23:41.040 --> 00:23:54.840
I know, for example, things around uh again, kind of fundamental pieces of infrastructure, whether it's like browser use um and how you can start to train models to to be able to use uh you know, the the internet better when looking at a computer and so on and so forth.

s225
00:23:54.840 --> 00:24:07.920
So, what are the what are the some of those examples to where you think that the uh post-training of open models will uh see new use cases basically unlock compared to just paying full price for the the frontier models?

s226
00:24:07.920 --> 00:24:09.840
Yeah, I think I like like I can start on this.

s227
00:24:09.840 --> 00:24:24.360
Like I think the the power what that we've seen with a lot of different customers is is really kind of this idea that like if you want to make a specific use case work, like we can take the example of like if you want to figure out like a way that agents can actually automate your text.

s228
00:24:24.360 --> 00:24:30.200
Like the the the most like concrete way you can do it today really is like build an RL environment for that use case.

s229
00:24:30.200 --> 00:24:33.240
Like train on it and then deploy it into production with those users, right?

s230
00:24:33.240 --> 00:24:38.640
Like let's say with a a million accountants that then now use this agent to ultimately get it towards full autonomy.

s231
00:24:38.640 --> 00:24:48.760
It's a bit like almost like Tesla's levels towards full autonomy where like you kind of need to deploy it into like do the last mile of actually like training for that specific use case, but then also

s232
00:24:48.760 --> 00:24:50.640
um deploying it to those specific users, right?

s233
00:24:50.640 --> 00:24:56.080
So, like there's a reason why like uh chatbot isn't good at self-driving because like it's not trained on that.

s234
00:24:56.080 --> 00:24:57.960
It's not deployed into that context, right?

s235
00:24:57.960 --> 00:25:08.320
And like I think it's the same even for these specific like knowledge work use cases, where it's like if you want to have the perfect like financial agent, it's much more likely that you'll be able to get there if you have like

s236
00:25:08.320 --> 00:25:12.520
our environment for that use case, if you deploy it into production, for example, as a bank, right?

s237
00:25:12.520 --> 00:25:17.600
Like to millions of customers, than if you're there's like one got model chatbot.

s238
00:25:17.600 --> 00:25:25.360
Like And I think this is sort of like what we've seen now with a lot of verticals and customers that like um there's like a huge unlock there to

s239
00:25:25.360 --> 00:25:32.080
um really go into this like specialized domains, post train on them, deploy into them, and then continuously learn from production traces.

s240
00:25:32.080 --> 00:25:43.600
So, we work with like some um also big AI natives on things like computer use, where ultimately having like millions of of traces from production data really can help you to to um continuously improve

s241
00:25:43.600 --> 00:25:44.840
uh those those agents.

s242
00:25:44.840 --> 00:25:56.440
And I think this kind of applies to almost every single domain, and I think it's sort of the the white pill for like the AI application builders and and the AI startups to actually have a a huge opportunity to

s243
00:25:56.440 --> 00:26:02.560
build kind of their modes and and to get to this data flywheel um of like specialized models even in a broader sense.

s244
00:26:02.560 --> 00:26:05.280
Like just going after like let's say computer use agents, right?

s245
00:26:05.280 --> 00:26:14.800
And I think um yeah, this is something where I think um we're just seeing a lot of like movement especially now with like all models catching up to the frontier.

s246
00:26:14.800 --> 00:26:24.920
Um and I think the other piece is like optimization, where like I think um like GM is a great example or like also like Trinity and NeMo Triton is like you have the whole ecosystem sort of like driving down the cost and optimizing it further, right?

s247
00:26:24.920 --> 00:26:34.960
Like like we very heavily work like very closely with all the teams here, but then also um like deeply also with Nvidia and with teams like ViLM to really drive down

s248
00:26:34.960 --> 00:26:44.680
uh the cost and and make the for example inference and training for models like GM or like models like Trinity and NeMo Triton extremely efficient, so you can basically

s249
00:26:44.680 --> 00:26:46.320
drive down the cost like further and further.

s250
00:26:46.320 --> 00:26:51.680
And I think this is something you don't obviously get with the closed APIs, where like they have like a huge margin on top.

s251
00:26:51.680 --> 00:26:56.040
Like they might drive down the optimization, but then might not pass through those savings.

s252
00:26:56.040 --> 00:27:04.360
So, I think in general, like the open models are only getting through the open ecosystem like more and more efficient like and and cheaper and cheaper to run and train on.

s253
00:27:04.360 --> 00:27:06.600
So, I think there's like this element as well.

s254
00:27:06.600 --> 00:27:18.320
I think too like a couple things that I I want to make sure we're very clear about is you you like most people probably do not need frontier level intelligence for like 90% of their tasks, right?

s255
00:27:18.320 --> 00:27:22.680
Like like not to say that you're not doing cool smart stuff.

s256
00:27:22.680 --> 00:27:32.120
Not to say that I'm I'm sitting here trying to do not cool smart stuff, but like a lot of the time these models are just overkill or they have like this really smooth

s257
00:27:32.120 --> 00:27:42.120
you know capability horizon that means they're they're also quite good at chemistry, but like most people are using models to do one or two things very well.

s258
00:27:42.120 --> 00:27:53.200
And open models let you choose those one or two things and then make the model just very good at those things at the expense of at the expense sorry of almost everything else.

s259
00:27:53.200 --> 00:27:55.640
And that that is great.

s260
00:27:55.640 --> 00:27:59.080
I mean that's exactly what we should be doing, right?

s261
00:27:59.080 --> 00:28:10.080
To to to use this model that is hyper generalized and able to you know perform well across like 90 different axes is is dope and cool, but it is not really

s262
00:28:10.080 --> 00:28:13.160
you know using the model effectively.

s263
00:28:13.160 --> 00:28:24.240
It makes sense for someone who is trying to ensure that everyone can use this one endpoint to do their task, but it makes much less sense when you're a person who's trying to do that task

s264
00:28:24.240 --> 00:28:25.600
yourself.

s265
00:28:25.600 --> 00:28:32.040
What what what was just said about efficiency is also deeply true, right?

s266
00:28:32.040 --> 00:28:37.920
I mean the idea that you are all here at a local AI summit.

s267
00:28:37.920 --> 00:28:40.440
Presumably you are running AI locally.

s268
00:28:40.440 --> 00:28:43.200
Presumably you would like it to be faster and better.

s269
00:28:43.200 --> 00:28:47.720
And presumably many of you are quite uh, quite cracked engineers, right?

s270
00:28:47.720 --> 00:28:55.880
This is a whole room of people who is going to contribute in some small part to making the ecosystem just a little bit faster, just a little bit more efficient.

s271
00:28:55.880 --> 00:29:06.400
And while it's true that closed companies can, uh, afford to hire great amazing teams of people, as we saw with Linux over the uh, whole time that it's existed,

s272
00:29:06.400 --> 00:29:07.000
right?

s273
00:29:07.000 --> 00:29:20.440
Uh, Linux is the thing that runs the internet, it runs networks, it runs all of these services that, uh, that require it to be hyper optimized in a way that I think you can only get when you have people who are trying to run

s274
00:29:20.440 --> 00:29:22.680
as resource constrained as possible.

s275
00:29:22.680 --> 00:29:35.520
And, uh, all of that to wax poetic and say this idea that like local AI and open models in the most efficient version of the of the model ecosystem,

s276
00:29:35.520 --> 00:29:37.800
uh, is is necessary to do it in the open.

s277
00:29:37.800 --> 00:29:48.280
I think it's, in fact, not possible to do it behind closed doors cuz you're shutting too many people, uh, that could make that one small contribution, uh, out of the room.

s278
00:29:48.280 --> 00:29:58.280
I I think that that it's important to state, too, that I don't think any of us agree that or or or of the mind that closed models or frontier,

s279
00:29:58.280 --> 00:30:08.440
you know, what OpenAI and Anthropic, just to name names, you know, Anthropic and others are doing it is, uh, not extremely beneficial or that not don't use them.

s280
00:30:08.440 --> 00:30:12.320
Um, I I I certainly, uh, I use those models near every day.

s281
00:30:12.320 --> 00:30:17.800
It's it's just that it's where does it fit in the, uh, in in the future of this ecosystem.

s282
00:30:17.800 --> 00:30:29.960
Um, and just like, you know, Chris is alluding to, you can think of open models and self-hosted or kind of owned intelligence or LLMs as like the Linux layer, which you're beginning to see kind of take place.

s283
00:30:29.960 --> 00:30:33.600
Linux runs enterprises, you know, it runs the cloud.

s284
00:30:33.600 --> 00:30:40.600
We're seeing a very similar thing take place with hyperscalers and neo clouds and providers like Fireworks, Together, Base 10's models.

s285
00:30:40.600 --> 00:30:54.360
Um, but just like Macs are one of the best ways to get work done individually in the same way that maybe using open AI and chat GPT is the best way for you to do the vast majority of simple

s286
00:30:54.360 --> 00:30:58.880
check my email, help me rewrite, you know, check for grammar, those kind of things.

s287
00:30:58.880 --> 00:31:05.360
It's accessible, it's easy, and for you know, your average consumer, an individual, it's pretty cheap.

s288
00:31:05.360 --> 00:31:08.280
Um, if you're using like the $20 a month plan.

s289
00:31:08.280 --> 00:31:23.600
In the same way that you know, Microsoft helps the the world of medium size to large businesses run on Microsoft and Windows because you know, they're not as expensive as getting everybody a Mac,

s290
00:31:23.600 --> 00:31:28.800
and you're going to see a similar world play out there for for some closed and open open frontiers themselves.

s291
00:31:28.800 --> 00:31:43.160
So, it's all an ecosystem, you know, I don't want to give the impression that you know, I think anytime you log into chat GPT or Claude that that you're committing a sin, only that as you are, you know, this is a a conference for AI builders, AI engineers,

s292
00:31:43.160 --> 00:31:54.240
as you're looking at the best way to engineer your product or your service, that there is another layer you can go down into and it's becoming way more accessible than it used to be.

s293
00:31:54.240 --> 00:32:04.640
I I think that's a great point, and I think that the relationship between closed frontier models and open models will be one that is it's constantly there, right?

s294
00:32:04.640 --> 00:32:15.800
I think that we have it's never you'll never get the headlines to apply nuance and say that both will coexist and gain more usage and are going to be useful to everybody,

s295
00:32:15.800 --> 00:32:21.400
but that is kind of the de facto state that that will not only currently exist, but will continue to exist.

s296
00:32:21.400 --> 00:32:33.320
Um, I want to spend the last few minutes here to really give the audience something that only you guys potentially can answer.

s297
00:32:33.320 --> 00:32:48.920
Often times I reflect about my time at Nvidia and I think I I feel as though I have a clear vision outside into there's definitely still a fog of war out there, but I have a vantage point that many people don't have and you guys because of the positions you are in as well.

s298
00:32:48.920 --> 00:33:03.080
And so what is the the thing if we're looking forward towards AI engineer World Fair 2027 that you think if you were to make a a bold prediction, let's say, let's not be conservative

s299
00:33:03.080 --> 00:33:11.000
around the intelligence in in the open source and ground it it was some frame of reference.

s300
00:33:11.000 --> 00:33:15.360
What do what do you think we can look forward to by by this time next year?

s301
00:33:15.360 --> 00:33:27.320
I think one key aspect obviously that people are closely tracking is like sort of like the just like capabilities of open frontier models and I think they'll they'll keep being very close

s302
00:33:27.320 --> 00:33:35.560
to the general frontier potentially even like what like what like now the the speed or like the of those close frontier models like slowing down.

s303
00:33:35.560 --> 00:33:44.040
I think like they'll they'll catch up even more and I think the the most concrete thing that I think will be very exciting is like seeing the world move from sort of chatbots

s304
00:33:44.040 --> 00:33:47.920
and now coding agents to like just general knowledge worker agents.

s305
00:33:47.920 --> 00:33:56.480
I think over the 12 the next 12 months right like to see more and more of like kind of like everyone across every knowledge worker domain like adopt agents in their

s306
00:33:56.480 --> 00:34:04.920
workflows which I think like developers have with coding agents have probably done better than any other domain in the world, but I think we'll we'll see over the next 12 months like

s307
00:34:04.920 --> 00:34:11.760
a lot of like domain specific like knowledge worker agents, but then also I think domains like computer use agents and I think others will take off.

s308
00:34:11.760 --> 00:34:14.320
Like I think in a similar way that like coding agents have taken off.

s309
00:34:14.320 --> 00:34:24.360
I think we'll just see almost like in some ways you could say like almost like the general intelligence for like the knowledge worker in digital domain before then hopefully maybe moving on to the physical

s310
00:34:24.360 --> 00:34:35.240
and I think very concretely like I think we'll like in in 12 months I think it's pretty likely that we'll have like better than fable metals level capabilities and open models.

s311
00:34:35.240 --> 00:34:43.240
And I think this is like a huge opportunity uh to ultimately enable a huge crop of new like AI startups and companies.

s312
00:34:43.240 --> 00:34:47.360
So, there's some ways like you want to almost like write the the levels of capabilities.

s313
00:34:47.360 --> 00:34:52.360
Like to some extent like cursor really only took off when like Opus was good enough to do coding, right?

s314
00:34:52.360 --> 00:34:54.000
So, it's like this is when like cursor inflected.

s315
00:34:54.000 --> 00:35:00.400
And I think we'll see like hundreds of these inflections for like startups getting started like now, like in over the next year or two.

s316
00:35:00.400 --> 00:35:13.120
Um once like open model and and I think we've seen this literally a month ago with like GM 5.2. I think like uh it was I think legitimately one of those moments when people were like, "Okay, this is now like similar to the Opus inflection point.

s317
00:35:13.120 --> 00:35:19.680
Feels like an inflection point for open models to be like extremely strong and ultimately enable a ton of new businesses.

s318
00:35:19.680 --> 00:35:23.040
And I think this will only continue like

s319
00:35:23.120 --> 00:35:29.280
Uh not so bold prediction is that uh Primin Electin RC are going to have a combined valuation of a trillion dollars.

s320
00:35:29.280 --> 00:35:31.160
Uh that's all obvious.

s321
00:35:31.160 --> 00:35:42.000
Uh but but but I think that this is going to be a huge year for um uh this is probably going to be the most consequential year for like the future of how AI gets distributed.

s322
00:35:42.000 --> 00:35:58.120
Um the the the Fable uh and GPT 5.6 um you know, uh uh uh embargo, if you will, um has left a lot of open questions, uh you know, no pun intended about open models and and where

s323
00:35:58.120 --> 00:36:00.720
um how this intelligence gets distributed.

s324
00:36:00.720 --> 00:36:05.880
And at what capability level it starts to be uh politicized and and and kept back.

s325
00:36:05.880 --> 00:36:08.640
And so, sovereign intelligence is going to be very important.

s326
00:36:08.640 --> 00:36:17.080
I I think that um if you were to take, you know, the general population of AI um users, people that are using it every day.

s327
00:36:17.080 --> 00:36:23.760
So, you know, upwards of uh a billion to two billion people if you, you know, include ChatGPT and Google and whatnot.

s328
00:36:23.760 --> 00:36:33.720
Uh maybe 0.000001% have ever used an open model, You know, I think it it or you know, run it on themselves with with the multitude of different tools.

s329
00:36:33.720 --> 00:36:46.760
Um and I would hope that with the work that we're doing and the community's doing and the way that we're advocating um for open science and open models and open discussion,

s330
00:36:46.760 --> 00:36:46.960
right?

s331
00:36:46.960 --> 00:36:55.760
That's probably been the most frustrating thing about the last couple weeks is that all of these conversations around capabilities and who gets to use them and who doesn't have been happening behind closed doors.

s332
00:36:55.760 --> 00:37:11.560
Uh my hope a-a-a-a-and I hope that I can predict that we will be able to have a a 10% to 15% of people that have ever used AI have used a model locally on their system and that that becomes a very important part

s333
00:37:11.560 --> 00:37:14.640
uh of ensuring that you have access to what you need.

s334
00:37:14.640 --> 00:37:17.240
Um and and so it's a prediction.

s335
00:37:17.240 --> 00:37:27.320
Uh it's also something that I know all of us uh up here and and you out there are going to try to fulfill and I hope that uh we can continue to advocate for that because if we're if we're quiet, if we just let this

s336
00:37:27.320 --> 00:37:38.240
uh things play out the way they are, uh open models will, you know, will will be put under the microscope um in the context of untrustworthy, unsafe.

s337
00:37:38.240 --> 00:37:51.080
Uh and as much as there's work and and um vitriol and weaponized terms being out there uh advocating for that, we need to be uh combating uh as much of that if not more

s338
00:37:51.080 --> 00:37:54.360
uh with the reasons that it deserves to exist.

s339
00:37:54.360 --> 00:38:01.320
Yeah, I just uh I could not uh plus infinity what the last part what Lucas said more.

s340
00:38:01.320 --> 00:38:16.280
I think this is going to be the most consequential year for uh open intelligence that uh will a-a-a-a-at least for from where I sit determine the future of uh of a summit like this, right?

s341
00:38:16.280 --> 00:38:21.760
Uh I think it will has a potential to look very different in two radically opposed ways.

s342
00:38:21.760 --> 00:38:32.880
Uh as for bold predictions, I think that we will not be needing to go to an API for most of the tasks that we all do each day with AI.

s343
00:38:32.880 --> 00:38:43.880
I think it's likely to assume that you'll be running a model that is sufficiently capable in let's call it day-to-day work on your on your MacBook within the year.

s344
00:38:43.880 --> 00:38:49.240
It's already extraordinarily close, so not maybe not that bold of a prediction to be honest with you.

s345
00:38:49.240 --> 00:38:59.120
I also think that we're going to continue to see models become the the future of AI, so not model, right?

s346
00:38:59.120 --> 00:39:09.200
Swarms of or specialized systems of models, I think you're going to be increasingly important.

s347
00:39:09.200 --> 00:39:20.400
And lastly on the open model front, I think we're going to see some very large architecture shifts, especially as we start to crack things like diffusion models for text a little bit more

s348
00:39:20.400 --> 00:39:27.320
to to get us models that are better suited for the the hardware that we have in our houses.

s349
00:39:27.320 --> 00:39:37.120
And then last meme one, I think you're going to buy computers with agent operating systems on them instead of traditional operating systems.

s350
00:39:37.120 --> 00:39:42.440
Similar to like you buying a Spark preloaded with with Hermes or whatever.

s351
00:39:42.440 --> 00:39:44.120
I think that's that's likely to occur.

s352
00:39:44.120 --> 00:39:51.520
I also predict that come September when the next iPhone comes out, you're going to get a lot of texts from family members asking about this magical new Siri.

s353
00:39:51.520 --> 00:39:56.760
So, a lot of people who have not engaged with AI are about to in a very real way.

s354
00:39:56.760 --> 00:39:59.080
And the the response to that's going to be very very cool.

s355
00:39:59.080 --> 00:40:04.880
And so, just like when Deep Seek came out, I'm sure a lot of y'all got questions about what's this Deep Seek thing?

s356
00:40:04.880 --> 00:40:07.080
There'll be another one in September, get ready for it.

s357
00:40:07.080 --> 00:40:15.160
Yeah, I think like this might be actually one of the kinds of crunch almost like unlock stuff like I think combination of like basically open models getting good enough

s358
00:40:15.160 --> 00:40:20.240
as well as like the on-device compute getting strong enough to serve the current of like today's frontier models, right?

s359
00:40:20.240 --> 00:40:21.040
Like in a year or two.

s360
00:40:21.040 --> 00:40:31.680
Like Like basically if you can like run all posts like at a decent speeds on your like phone or laptop, I think the majority of humanity will probably like run local models.

s361
00:40:31.680 --> 00:40:44.560
Like and and I think this probably applies more to the consumer than to the heaviest like enterprise agents, but I think it seems pretty likely to me that like there will be some flexion point even then like almost like similar to a new platform shift where like

s362
00:40:44.560 --> 00:40:56.480
you can almost like tap into the local compute of a phone or laptop and then like start a next generation of almost like AI-enabled applications without like that that can ultimately really like leverage the local compute of like

s363
00:40:56.480 --> 00:40:57.720
device on-device compute.

s364
00:40:57.720 --> 00:41:04.840
You can run a 4 billion parameter model on your on your phone right now that is way more useful than GPT-4 was when it came out.

s365
00:41:04.840 --> 00:41:15.680
And I think it's important for us to continue to to to focus on how do we best utilize that in the most meaningful way possible as well as chase

s366
00:41:15.680 --> 00:41:22.520
the newer capabilities that will come from things like, you know, drug discovery and and and scientific exploration.

s367
00:41:22.520 --> 00:41:25.360
It's It's going to be a a fun couple years.

s368
00:41:25.360 --> 00:41:25.880
Absolutely.

s369
00:41:25.880 --> 00:41:36.000
If I were to try and summarize, I think that, you know, we're going to learn a lot more about how to these local models are are in built incredibly in the systems and their relationships as they,

s370
00:41:36.000 --> 00:41:43.760
you know, interact with frontier models over the next panels, but this panel really shows that I think that we are at an inflection point to where if you think about how,

s371
00:41:43.760 --> 00:41:50.840
you know, not even a short 6 years ago it was AI was really for the research crowd and not really many people cared about it.

s372
00:41:50.840 --> 00:42:02.280
And then of course it came into the public consciousness with ChatGPT, um but now there's this next thing which is that open source is now really starting to enter the public consciousness, but very few people have touched and played with it

s373
00:42:02.280 --> 00:42:04.240
and have had that aha moment.

s374
00:42:04.240 --> 00:42:15.360
And it sounds like we have the potential to do that this year and sort of guide the the future wisely, but ultimately it's up to a lot of the builders in this room as well to to to leverage that and represent,

s375
00:42:15.360 --> 00:42:27.800
um, you know, this important inflection point that we're in, uh, on the side that hopefully brings, uh, you know, intelligence, more intelligence to all of us, which is ultimately, I think, what everyone in this room would agree is is sort of the direction of progress.

s376
00:42:27.800 --> 00:42:36.600
And you know, everyone has said it on a panel previously, so I'll just also say it, which is that, uh, and and both of you have already said it, in fact.

s377
00:42:36.600 --> 00:42:40.080
Uh, like you you guys are extraordinarily important to this goal.

s378
00:42:40.080 --> 00:42:48.520
Uh, every one of you who is in this room and your friends and whoever what whatever communities you're part of, uh, without you guys, uh, we we will we lose the fight, right?

s379
00:42:48.520 --> 00:42:53.200
So, thank you for showing up and, uh, I I can't wait to see what we all build together.

s380
00:42:53.200 --> 00:42:59.419
And with that, thanks Vincent and Lucas, RC and Prime Intellect, and of course Chris from Nvidia.

s381
00:42:59.419 --> 00:43:00.840
[applause]

s382
00:43:00.840 --> 00:43:04.200
Always, always a pleasure.

s383
00:43:10.554 --> 00:43:12.554
[music]

s384
00:43:15.140 --> 00:43:17.140
[music]