Partnership Panel—Genesis Mission Summit 2026

U.S. Department of Energy
60 followers
Jul 29, 2026
AI Generated·May contain errors
Video Summary
AI GeneratedThe Genesis mission is a collaborative initiative between the U.S. government, national laboratories, and the private sector to secure leadership in AI for science. By integrating hyper-compute infrastructure and fostering tripartite partnerships, the mission aims to compress scientific discovery timelines from years to months, driving breakthroughs in energy, biology, and economic prosperity.
All right. You know, the word partnerships, not just internationally, but our partnerships here
in the United States, the word partnerships has been thrown around a lot today. And it's because
we would be doing ourselves an injustice by not recognizing and amplifying and utilizing the
amazing talent and ability and capabilities that exist in the private sector here in the United
States of America. And that's why partnerships are so important. So to lead a discussion on
how essential partnerships are, please welcome me in joining back Secretary Wright back to the
stage. Alongside with Secretary Wright, we're honored to be joined by partners from industry
and national labs. Misha Laskin, the co-founder and CEO of Reflection, a frontier AI lab.
Karthik Narayan is the chief product and business officer of Google Cloud. And finally,
director of Slack National Lab and the director of the National Lab Directors Council,
Professor John Sirag. Please join me. Don't worry. You won't have to hear from me much at all.
I am here with the honor to sit on the table with the head of Stanford Linear Accelerator, Slack
National Laboratory. And he also leads all of the lab leaders right now. So the director of the lab
leaders. Well, actually, I'm going to have everyone introduce themselves. But the idea of
the panel is how is industry, big business, new entrepreneurial startups, the national labs,
universities, how are we all going to work together in this mission? And I'll give a tease
for the next session. You'll hear Stephen Schaefer, who's the head of Oak Ridge National Labs,
very early on in this project. We're like, we need to move fast. We need to do things
differently. And we made a reach out then who wants to build computers fast, not a normal
procurement process fast. We have a system already running, you know, less than 12 months
from discussion to running an awesome new machine that you'll hear about. Hats off to Lisa Su from
AMD who jumped on a plane to come say, we're ready to build starting tomorrow. And HPE,
the integrator who built this system. It's a tremendous story. But it's the story of collaboration,
labs, industry researchers. But why don't we have everyone introduce themselves up here? And we'll
start with you, Misha. Well, first, Mr. Secretary, thanks for having us all here. It's a real honor
to be a part of the Genesis mission and supporting science in America. I'm the co founder and CEO of
reflection. We build open models, American open models. So you can kind of think about the
difference between the closed models and open models coming out of China. We're building
that variant in America to support the government and our enterprises here.
Dario says we have to lead in open models and proprietary models. Are we going to do it?
I think we will. I think we have all the capability here. And there's the right kind of
commercial now incentives and community incentives to
build an open source frontier intelligence laboratory. Fantastic. Kartik, you work for
Google. I think I've heard of that company. But tell us about yourself about your focus.
Thank you, Secretary Wright, for having us here. Kartik Narain. I lead product engineering and
all the go to market capabilities for Google Cloud. And I represent Google and proud to be
a founding member partner of the Genesis mission. And we genuinely believe like the Genesis mission
believes in that, you know, the in the age of AI, scientific discoveries and economic prosperity
are an eventual outcome. And we want to be part of it. And we've seen this firsthand,
when, you know, government labs, academia and industry come together, what could be achieved.
We had an early example with AlphaFold, which got the benefit of publicly available scientific
protein data from the protein data bank supported by Department of Energy,
and bringing the capability of that data and private AI capability resulted in a Nobel winning
outcome. And we genuinely believe that this mission, by being part of it, we would be able
to jointly enable 20, 30 more Nobel winning scientific discoveries. That's, that's possible.
And the last point that I want to make is, you know, an introduction is, not only are we a
founding member, but we are proud to be making an in kind commitment today of $40 million
on a couple of things. Number one, number one to provide trillions of tokens available for
scientific researchers through the Gemini for government, you know, platform that that we have
provided, and also, you know, cloud credits for access to specific models, like AlphaFold3,
AlphaEvolve, AlphaGenome, that, you know, individual scientific researchers, labs and
cross lab, you know, as John drives, can can leverage with and accelerate this,
this progress to faster outcomes. Thank you. Thank you. And John, tell us about.
Yeah, no, thank you, Secretary Wright, for your leadership and for bringing us together.
Some would say I'm not that excitable sometimes, but you can't not be excited by this. I mean,
this is amazing. As you said, I'm the director of SLAC National Accelerated Laboratory. So I
definitely have a favorite of the national labs, and that's SLAC. But more importantly,
and I think my role on the panel is that I am this year's chair of the National Lab Directors
Council. And so I think a real feature of all 17 of the national labs is us working together.
I think all of my 16 other lab director colleagues are in the room. So that's just
a measure of how important Genesys is to us. And with no disrespect to my colleagues,
what's even more exciting is to see all of our colleagues from the labs who are participants
in the Genesys proposals that you announced this morning, sir. It's truly amazing. It's
part of why we're all here. And it's just really looking forward to what comes next.
So, John, we seem and everyone's in agreement that we want the United States to lead in AI.
That's a high challenge. But tell us, what do we need to do to achieve that? And what are maybe
we blind on? We're not thinking about and it's going to trip us up if we don't change it. And
then I'm going to come to Kartik and Misha after that with the same question. What does it take to
be the leaders? So, well, I stand by the comment that I just made about how exciting things are.
This is actually the least exciting day for Genesys, because tomorrow is going to be even
better. And the day after that and the day after that. So I think this really is just the tip of
the iceberg. And so how we get started on all the projects we announced, how all of those national
challenges challenge the overall platform to make it even better and harder, it's just going to
spiral up from there. So how we bring together the challenges and the platforms grounded in the data
really is the opportunity. And I think success will look like Dari articulated some metrics
early this morning, but when we look back a month from now and say that was kind of slow and not
that exciting, that's actually the real progress to be had. And I'm totally bullish that's what's
going to happen. And Kartik, what are we getting right? What are we missing? How do we create an
ecosystem that leads the United States to lead in AI? I think the Genesys mission and the real world
use of AI are the two X factors that has come into an already strong scientific foundation that
the United States already has. There's not a lot of places where there's history of these many labs
with specific focus and outcomes that they've delivered over a period of time. And I'm sure
in scientific research and acknowledgement, whether Nobel or various such recognitions,
the United States is already leading. But AI has come to an age where suddenly you can unlock the
private data that was available at the disposal of an individual researcher to all the researchers
in the same lab. And now with the mission of the cross-lab collaboration and Genesys,
this will be going across the board, across all the labs, so that every researcher can stand on
the shoulders of all the progress that is being made. So that is the progress of AI that can be
used. And I'm loosely using the word AI, which means that the ability for data to be readable,
understandable by AI, the advancement in the hypercomputer or the high-performance computing
capability like TPUs and GPUs and the likes is one dimension of it. But the other dimension is
the Genesys mission in itself. And John and I were talking and when we made the investment,
this additional in-kind commitment of $40 million, we said that this is for the Genesys mission. But
if you look at Genesys mission, there's no one lab or there's, I come from a commercial world,
so I use the term P&L. It's not one P&L to go to. This naturally forces in a good way,
all the labs to collaborate so that they're able to build things that are reusable, so that you
are able to get more for the R&D investment or the scientific investment, federal investment
that is going into scientific research. I think this goes to the exact statement, I think,
Secretary, probably you, coined that to double scientific discoveries in the next decade. For
that to happen, you need both efficiency and effectiveness, and that is what Genesys mission
would bring to the table. Great. That was Dario's goal, but I'll take credit for it now.
Thank you, Dario. Misha, the same question. What is it that we need to do to make this environment
so we thrive here across our country, and of course, with our international partners as well,
but how do we make the U.S. the leader in AI and AI for science?
I think the difference between science in the U.S. and anywhere, say 10 years ago and now,
is that there's a new foundation on which everything is being built, and that foundation
is AI. The thing that is missing in that foundation today is the ability to have great
open American models that scientists can then use and customize for the various workloads.
I'll give an example of what I mean here, an example of where this has worked very well,
and an example of if this thing existed, how it could power American science more broadly.
Before I was an AI researcher and now started a company, I was a theoretical physicist.
These models have actually gotten very good at math and physics type of calculations,
because there's a lot of data on that, and you can do a lot of simulations there.
I had this prompt that was basically my PhD thesis that I used to track the progress of
these models. Two years ago, it was falling over entirely when I basically gave it the
question of what my thesis was. A year ago, it got it at, I'd say, an undergraduate level.
Six months ago, actually entry-level PhD, and then two months ago, it one-shotted my PhD thesis.
It took me four years to do that. You could see that as a negative, but actually it's a positive,
because now, if I was back as a PhD student, I could crank out a thesis a week. That's the
production of knowledge, because it takes a while to ask the right question, but then,
because the system is calibrated to the data, it can produce the answer quickly.
How do we take that and drive that into fundamental progress, especially in
more experimental science like bio, development of photovoltaics, chemistry,
where maybe that data is more locked up in the national labs? Because the physics thing worked,
because there's a lot of physics data on the internet, you need to be able to customize
them towards the data that you have in the national lab. I'll go for another example, which
is, I actually moved to this country because my parents got jobs in a national lab, the Pacific
Northwest National Lab. I'm Russian-Israeli, moved here as a kid before the accent set in,
but we moved to Washington State, where my parents were both chemists at this lab.
An example of the kind of research my father did, and still does, is material science for
better photovoltaics. How do we make better solar panels? Well, what does that look like today? You
have to basically assemble materials together, run experiments in the lab, and it's a
pretty manual process. There's a large search space, and so it's pretty slow
and manual. But this is the kind of thing that you can imagine, like an AI system, closing the loop
and actually greatly accelerating the process in which we get cheaper, more scalable photovoltaics.
But that will only be unlocked if there's an open model that labs can use and a scientist can
customize. So that may be some kind of roundabout way of saying, if you have great open models,
then scientists will be able to customize it for their needs. And instead of, it took me four years
to do the PhD thesis, now it's a week, it took my father 10 years to discover some new piece of
photovoltaic material, that could be a month. I think that's the future we want to go to.
So I think you raised a great question right there, which is a question I hear a lot.
You know, hey, well, it took you years to do your PhD, and you could do it in a week now.
People think, oh my god, then all those jobs are going to be gone. All those jobs, you know.
I can give an example that never resonates with people, that 80 percent of Americans used to be
farmers, and now it's less than two percent, and everyone got a better job. You know, it didn't,
it didn't, jobs change, but they don't disappear. But it's a big question right now. Everyone in
this room, we can feel it in our bones, some of us maybe even a skin tingle, how exciting
what we're doing, and how awesome the benefits are going to be. But a lot of America doesn't
get that, they don't feel that, they don't know that. How are, what are we going to do,
or how is it going to work, that these, you talked about some specific innovations,
how are we going to, how are those benefits going to translate to the average American feels like,
thank god we got AI, this Genesis mission is awesome, how do we double down on it?
That's, how do these improvements get felt by the American people?
I think in the case of, like, what's, you know, what got me into science is so beautiful about
science, is that you're exploring the frontier of human knowledge, and it's so positive some,
because we probably, we haven't discovered even, you know, I think even a smidge of everything that
there is to be discovered. So what happens when you discover better photovoltaics? Well,
you get scalable renewable energy at much cheaper prices, right? That you're able to lower the
energy costs for everyday Americans. What happens when you have breakthroughs in biosciences, where
new drugs are being discovered faster? Well, you basically increase the life expectancy of your
population, because you're able to cure diseases that weren't curable before. And it kind of,
this percolates to, you know, I think that every scientist is actually empowered to,
it's so much better if you see the impact of your stuff instead of 10 years. You can iterate,
and right within a year, start seeing impact that then diffuses across the economy. I think that's
kind of what it looks like, right? I think it's actually much more empowering to be a scientist,
and I'd be actually really excited to go back and do the PhD in theoretical physics again,
because I think I would discover a lot more than I did then. Absolutely. And Karthik, you're in
the commercial world. How fast can these innovations, you and your team and the teams in
this room, how fast can they be felt in some fashion by the American public? And how can we
accelerate that or emphasize it? I think, you know, you hit a very important point that while
in the bone or in the skin or in the instinct, everybody believes that AI is actually going to
drive better benefits and better outcomes for, you know, every citizen out there. There's this
feeling of the interim state that is a concern. And the reason it is a concern is compared to all
other cycles of change, whether from agriculture, farming to other professions and several other
examples, the curve was a lot, you know, slower, or the pace of that, that change was a lot longer.
And this time, it feels like happening a lot faster, that everybody's worried, whether you
have the time to rotate those jobs or professions or the skill sets and so on and so forth.
The good news is even in the commercial world, there is a significant pivot in the focus from
a year ago, where year 18 months ago, the focus was all about productivity improvements. And,
you know, what can be the cost reduction outcomes that could be delivered from AI to a significant
pivot of what are the new markets? How do I improve the reach to a customer? How do I
create better customer experience between when they come during a sales cycle versus a customer
service cycle? We've all experienced that these are two completely different experiences. One is
a very pleasant experience. The other one is a very, you know, dragging experience, or how to,
you know, create newer products at a faster, cheaper, more novel ways, which is the new
product introduction. I think the same elements are going to be applicable in the scientific
world as well. But in the scientific world, there is also another factor, unlike in the commercial
world for all the other kinds of jobs, there is, so to speak, you know, abundance of, you know,
labor availability, or at least as much labor as is needed is available. In the scientific world,
that has never been the case. You've always been short-staffed on the scientific research,
you know, population or people who want to do scientific research. Now, by focusing on,
you know, removing the rote tasks, and I'll give you an example. In the National Laboratories of
the Rockies, they act, you know, we actually use Gemini agents to automate, you know, electron
microscope calibration. You know, no researcher is going to go back home, you know, after a month
of effort, feeling accomplished that I calibrated the electron microscope. What they are really
interested in is to start doing their research with a calibrated electron microscope. So what
we did with the Gemini agents, we built a full automated workflow that reduced 90% of the effort
to be completely automated by AI agents, and also did this 8x faster. And now the ability to get
started with your research is much faster. And the funding, the federal funding or whatever
funding that they have got, industry plus federal funding, can be used more towards the actual
research rather than the labor hours spent on doing the rote tasks and so on so forth. So this
is just one example of activities that could be done to make things better. And last but not the
least, we talk about all the research scientists who spent four years, you know, I don't know
whether we have the count of the research scientists who dropped from PhD after doing it
for two years, three years, five years, I think that will be a meaningful percentage. And the
reason that they drop out is because they are trying to make progress on one thing over, you
know, and we say that, you know, they don't need as much, you know, validation on a daily basis,
but everybody likes validation of progress. And AI will give them that validation of progress
a lot better, so that they would be at research for longer, and they would be able to do
more scientific discoveries than ever. So AI is going to make it more fun to be a scientist. I
think that's, that's huge. And among that constituency, we've got them. But you think
the public is going to feel it through better services and products, it's going to happen fast
enough that they're going to have better experiences and feel better. People will like that, right?
Absolutely. And there is generally this notion by public on anything that is non-commercial,
whether they're, you know, indirect dollars, tax dollars and other indirect contribution,
whether that is getting its right yield, or productivity or outcome, and so on, so forth.
And they would see and appreciate that there's a lot more that's coming out because of technology.
So that, you know, there would be more investment. And eventually, all of this is going to make
whether that is, you know, medical discoveries, biological, molecular progress, or energy
discoveries, and so on, so forth. This would be absolutely felt. And there's also going to be a
easier correlation to the cause and effect using AI so that, you know, it doesn't feel like, you
know, it doesn't feel like a butterfly effect, that, you know, the butterfly, you know, flapped
its wing and, you know, South China Sea, and you will feel it in the West Coast of the United States
two years later. All right. So, John, right now, there's worry of the crowds with pitchforks,
stopping the data centers from stealing all the electricity and drying up all the lakes in your
town. So we have a public palpable opposition right now. Talk to us about making science better
and faster in Genesis Mission. How is the American public going to say, put the pitchforks down,
this is good? You know, so I think it goes back exactly to the premise of the question. And
especially for we as national laboratories who are principally federally funded, we need to own
the feedback. It's the taxpayers and the citizens who are paying us. So first thing, the question
is fair. Got to address the concerns and meet it there and be responsive. So I think there are
three pieces that really address that. Importantly, this is a mission initiative. It's not called the
Genesis Science Handbox. It's called the Genesis Mission. We're actually trying to solve real
problems for real people. We highlighted an example earlier this morning of we're going to
use AI to make the electric grid better. So it actually, it can be net positive. Yes, it might
take a few more electrons to do the math, but the output is much better. So it's a mission. It's not
a sandbox. That's vitally important. The other part that I think this panel portends is it's
about the partnership. We the labs play, and at least in my judgment, an essential role in the
research and innovation ecosystem, but so do our university colleagues and especially so do our
colleagues in the private sector. We need to accelerate the results that labs produce into
products and impacts even better, even more effectively. That's why we need our private
sector partners working hand in hand. And then I think the third piece is the workforce development
opportunities that was highlighted earlier this morning. I don't know anybody who went to grad
school who said, well, I wish it had taken me twice as long and I was half as effective.
So I calibrated a microscope for a month. I spent the whole year tuning it. People are
motivated to do results. And so it's how do you do science differently to make an impact
that makes the lives of everyday citizens better? That's the mission. If we both articulate that
and walk the walk, I think we'll do just fine on the feedback, but the feedback is real.
Fantastic. Let's, John, just stick with you and come back to this partnership issue.
We explicitly required in the thing that you have to have at least between universities, labs, and
industry, you have to have at least two of those involved. The phase two, the larger grants, you
have to have something from all three. So we're trying to drive collaboration. Is that going to
cause friction or is there going to be people, how's that going to work? What are you worried
about with the partnership thing? Well, and oh, by the way, you said do it in six weeks time.
And so, and like this much time. It's not the worst thing we did. It isn't good,
but it's not the worst. We all figured out, including the 280 odd teams in the room,
how to do that, how to build partnerships among national labs and universities and the private
sector. So that was clearly an inflection point in the community. You could not go to any place
in the research community and not hear Genesis, not hear RFA, maybe another little shaking of
head at Dario, how could he do this? But in the end, it all kind of works. So I think an important
point is this is just the tip of the iceberg. This is a great launching point, but to my earlier
comment, it's only going to get better from here. There are even more proposals that weren't able
to get funded this time that have built teams to go take it to the next level. We need to double
down on that, both through the Department of Energy, but also through the interagency partners
we heard at this point. So there's even more opportunity, more chances to take that partnering
to the next level. So we just, it's a context for it. We need to work it together. I would say one
of the areas I'm concerned about is how do we bring all the data that we, the national laboratories,
produce to our user facilities and both make that broadly available, but also appropriately protect
our data so it's not misused for other purposes. We just need to get after it. Great, great. So Karthik,
you're in a business, a big business, very profitable business. As John just said, it's a contact sport.
Do you worry about that? Hey, we're doing this awesome stuff at Google, but we got to partner
with the universities or the national labs. Is that a, does that give you concerns about
leakage of technology or misaligned incentives? Or how do you feel about this sort of forced
partnership? We actually are not looking this as a forced partnership. Actually, the Genesis
mission makes the ability to partner a lot easier and better. And the reason for that, like I said,
it would have been a very difficult way to figure out this investment that we are just talking about,
the $40 million investment, if there was not a Genesis mission and we had to pick and choose
which lab and what initiative within that lab and how granular do you go? Then do you go and
do an analysis of which researcher is working on something? Now this is like making a startup
investment, seed funding. It's going to be too difficult without this kind of a framework.
You could use the word forced partnership, but it's for the good. I'm just trying to be
controversial. Yeah, I can understand that. And you can afford to say that. We look at this as
a privileged way to partner. And the reason is that there's a lot of commonalities that
need to come together. There are three layers that need to work for bringing advancement through
AI for the labs and research advancements to happen. The first is the hyper compute
infrastructure layer. We need to have a common standard with which everybody is able to build
their infrastructure within a lab and across the lab so that you would be able to figure out
what are the tools that you are building and what are the common standards that you are able to
create so that you are able to progress on the scientific discoveries that you will be able to
build on top of that infrastructure. So that common framework is now possible. And in that,
Google has its own capability across the layers, whether that's the TPUs to
up above the Gemini models. But if you look at it, we are the closest partner with NVIDIA to
bring their GPU stack also through our Google Cloud Platform. Even though we have Gemini as
our own proprietary model, our Gemini Enterprise Agent Platform delivers, provides inference
capability for all models, including open models in the industry. Both are competitive
frontier labs as well as open models. So that infrastructure layer is important. The second
thing is you need a way to unlock data for the age of AI. Data built for queries and analytics
and individual processing is very different from data being available for AI and agents.
So you need common tools and frameworks with which data can be shared between researchers,
across labs, and in certain cases, also internationally for certain specific
needs and so on and so forth. But you need to be able to do that based on open standards,
benchmarks, evals, and also keeping in mind the sovereign needs of a particular lab or a researcher
or a government. So to do all of that stuff, you need a common framework that needs to operate
across the thing. And last but not the least is the actual workforce enablement. This is going to
be successful if the researchers feel that this is enabling them. We need to, we are as of today,
and we need to be the destination. The U.S. needs to be the destination for any scientific researcher
to come to make their scientific discoveries. It is not a secret that there are other countries
that have made progress in the recent past. And the fact that we need to make this place
the destination, and how do you make this place the destination? What do researchers need? They
have all the tools and infrastructure and the investment capabilities that can progress their
scientific research. That's requirement number one. They have the data set and the collaborative
environment with which they can actually work with others so that they can build on something
that others have already built on. And last but not the least, they also need to feel that
they are able to actually create this reverse, the full feedback loop between industry and
academia and labs and government so that they are getting a sense that their research actually has
useful outcomes and is going to be able to help the general public. So you need to create that
environment, and we have worked to do that. Awesome. Well, all right. We're going to go to
the tricky, the intellectual questions, sort of deep questions. So I'm going to start with Misha
since he completed his PhD in a week. Or it took you years, but it was only a week's worth of work.
So maybe I should start with John. But I'm going to start with you, Misha. It's relevant for your
business. So I remember 30, 40 years ago when I was young, there was this movement that open
source software. You know, we had Unix had been around, but there was Red Hat, and we're not going
to need commercial software anymore. It's all going to be open source. And I think it's shaken
out to there's a certain chunk of software that's open source, a certain chunk that's proprietary.
But I look at AI and all the coverage in the news is sort of the proprietary
frontier models from American companies that are awesome, awesomely powerful,
so scary to some people. And then we hear most of Chinese efforts are in the open source domain.
So talk to us about what is the role of going to be open models versus proprietary models? What
should the different roles of those? Is one of them going to win? Are they always both going
to be around? And how does that work? I think it's a really good question. And
I don't think it's a this or that. I mean, you can kind of if you forget where the models come
from, you can like the market demand is for both, right? And if we always will be for both,
they're both going to be around. Very much so. I think that it's actually very similar to,
you know, previous software technology from this perspective, like operating systems,
right? Operating systems that we use on our most kind of consumer devices tend to be closed.
But the operating systems powering most computers in the world actually tend to be open because,
you know, the ease of customization, flexibility and permissiveness around them just makes them
sort of spread like wildfire. And so I think that the one thing that is probably that's
probably happening right now that there is a geopolitical issue between American and Chinese
AI. But it's kind of being muddled with, you know, a separate issue, which is closed and open
technology. And those two things have nothing to do together. It just so happens that we're
in a world where, you know, there's proprietary, great proprietary models in one country,
great open models in another country. But fundamentally, I think the equilibrium state
will be that there are great proprietary and open models in America, in China as well, right?
There's nothing to do with kind of the geopolitics. It just has to do with, hey,
American enterprises want both closed and open models for different workloads. The Genesis
mission will require both closed and open models for different workloads, right? That photovoltaic
customization thing, you'll need an open model and something else you'll need a closed model for.
So it's not a this or that. But I anticipate like open source software and the internet and
that the majority of token consumption now and in the future will be dominated by open model.
Like today, most tokens consumed are, you know, going in are consumed through open models,
just because they're so much more cost effective. And I think they'll only, you know, they'll
continue an increase that there might be a lot of commercial value in the proprietary model
themselves. The open models probably consume most tokens going on in the future. And there
are various business models built out below the model, right, like the compute and infrastructure
layer, and above the model, the application services and customization. And John, you agree
with that is is science is most of what's done at the lab is going to be in this open models,
or or not necessarily. So like, how do you think the mix between open and?
Right? I think it's exactly a both, as was said, and it's and what we're doing in Genesis is just
a microcosm for what we the national labs do in general. The science frontiers we're advancing
are the most compelling parts of science. And if we want to be world leaders, we need to engage
with the international community and be open. And the questions we're pursuing our national security
challenges are doing it. But you've got to get the balance right. Within the labs, we've developed
this rubric called a managed research environment. And I think those principles of what we do openly,
what we do more protectively applies in everything, whether it's open and closed
models, whether it's who we partnership, who we partner with, if we take that overall framework
and strike the balance, we'll come out where we need to be, which is being world leaders.
And Kartik, for you, from the commercial space, you've got a proprietary commercial model,
but you're also open models, you're going to be across that ecosystem. That's just the way
it's going to work. Yeah, the way we think about it is that we're still in the early to mid
phase in the trajectory of how much reasoning can further develop or improve. And to go through that
curve, organizations, frontier labs, are going to require immense amount of investment. And as a
result, there is this frontier closed model that would be required as a business model to be able
to further fund and take the reasoning to what is potentially possible. Whether that's AGI or ASI
or whatever you call it, a lot more is possible. That said, Google also has been a proponent of
open technology ever since. The entire cloud is invented around a technology called Kubernetes.
Kubernetes was invented at Google and open sourced to the rest of the world, based on which
Amazon built AWS. Even though we did that, we have a Google Kubernetes engine, a proprietary
version of that. There are numerous examples of software that we've done.
When you think about open, there are two dimensions that people need to be thinking about.
One is open weights for you to be able to modify those weights so that you are able to customize
certain things, which is open source, and open weights, where you are able to see the weights
and basically put it in your own environment and do that. One is a privacy sovereignty point,
the other one is a customization point. We believe that there will be a need to customize
certain models for specific scientific discoveries and scientific purposes because you need to be
precise. You need for the model to work on a specific set of skills with a high degree of
accuracy. We see a world where a workflow will have a lot of things done by these frontier models
for a high degree of reasoning, grounding on massive amounts of data that are diverse
across different data sets. Then there will be specific activities and tasks that need to be
performed by a highly customized, fine-tuned model. We do have our own open source model
in the Gemma family that we constantly make progress and publish to the environment.
We believe that what is required is a platform that operates this heterogeneous
model and agentic architecture. That is what we are creating so that enterprises could use.
There are challenges and complexities in using a heterogeneous architecture compared to a
homogenous architecture. It's not impossible. You need to do the work. We don't want every
research scientist to be swinging on their own to do those things. We want that to be done
through the platform that we have delivered to the government and now that we are giving to the
Department of Energy, the Gemini for government. John, I'm going to start with you. This is the
final round. What are you most excited about? What's this awesome a-ha thing that could happen
in the next two or three years from Genesis Mission? Scientific breakthrough? What are you
dreaming about? I guess two things. Since I said at the top this is the least best day for Genesis,
I'm excited for what happens next for all the people in the room getting started and doing
the work. I think that will be just amazing. My own personal bias is when we actually link up
the resources, the platforms we're creating that go all the way from large-scale computation
through partnerships with our industry partners in the cloud to actually driving scientific
instruments ubiquitously across the department, that'll be truly amazing. Whether you're operating
a place like LCLS at Slack or the LST camera, that full integration will make the stuff we're doing
now, which is amazing, seem relatively pedestrian by comparison. So the future is remarkably bright.
Fantastic. Misha, punchline. What's something we can't imagine today but that AI and innovation
is going to make arrive? I think it's the shrinking of time horizons to discovery
across every discipline of science that's happening in the world and across the United States.
It's general purpose technology. It's going to be in the hands of every scientist,
both closed and open versions. It's just going to close the feedback loops on discovery.
I think that what we're going into is an era of unprecedented prosperity.
The reason that a PhD is, let's say, four years is because that's how long it would take a human
to do the thing. Human is not a third party concern to me.
With AI, the compression on discovery is much lower. I think it begets a lot of benefits
very broadly. The other thing that I think is not maybe as touched upon is that
when we say science, we assume that it's open. Science used to be alchemy. Newton was actually
famous in his day for both being an alchemist and a physicist. We remember him as a physicist
because at the time, the journal was born. The science that he did in physics was open,
and now people can read for generations. I think this part of the Genesis mission of
writing all this up, publishing it in journals, contributing to software that's open source,
and having these artifacts that then can be studied and investigated for generations to come
is really exciting for me as well. Awesome. Kartik, punchline. What are you excited about?
I think there are three things that I'm excited about. Number one is that
we have an opportunity to give a wholesale upgrade of the capabilities that American
research scientists will get to drive better scientific discoveries and put America as the
clear number one head and shoulders above anybody else in the scientific discovery,
whether that is the AI infrastructure upgrade, modernizing how data is used,
as well as all the tools that they would need for AI to do that. It's going to be like there are
times when the US highway system was created or several such projects. This could be that moment
of a wholesale upgrade of capabilities. That's number one. Number two, I think there is a huge
opportunity in fields of whether that is energy fusion or biological progress. That is absolutely
in the horizon. It's not going to be decades. It's going to be within a decade. That's possible.
Last but not the least is that because of the Genesis mission, there is going to be for the
first time a true feedback loop between research and citizen outcome that could be created.
The mission of Genesis mission should be to hold the labs, the academia and industry accountable
to deliver this and the research scientists and show to the public that they are getting
their money's worth or their citizenship worth. Fantastic. Three pieces of data or
datum that show me the world is going to be bright. Thank you all so much.
