00:00Why London? We have a huge pipeline actually by the way thanks for having me
00:05I'm very excited to be here. We have a huge pipeline of researchers, engineers
00:11from Europe and also DeepMind that want to join Luma and also London is the
00:20gateway for the business in Europe as well as Middle East so it seems to be
00:24the right place to have our second office outside of Palo Alto and Bay Area
00:28so yeah today is is our launch of the London office. I have a feeling Demis
00:32Hassabis over at DeepMind and and the AI labs at Google will be having his ears
00:37pricked by that Amit so why would people currently working or having been trained
00:42at DeepMind and the like want to jump ship to help build Ray3 your reasoning
00:46video model? Yeah I think that's a great question so Luma already has actually a
00:50deep bench of researchers from from Nvidia from from DeepMind from from like you
00:57know schools like MIT Stanford and Berkeley and and the reason that these incredibly
01:03exceptional people actually joined Luma is because one while it is a very very well
01:09resourced AGI lab we're only about 150 people so we get to do and get to have
01:15resources per person that like you know it's unheard of in rest of the industry
01:19actually and second Luma is so ultra aligned on this one goal which is to build multimodal AGI
01:26there's actually no second there is no second project happening at Luma this is what we do so
01:33people researchers engineers and and product uh are people who believe strongly in this mission
01:40Luma is the best place in the world for them to be looking at some of the amazing video that you create
01:46I can see how it goes into the creative sphere into advertising and the like but Amit it feels as
01:51though this model these world models has moved to perhaps more physical AI applications must open up
01:58different industries yeah uh so I mean video and and video is actually the path to AGI and let me tell
02:06you just very briefly why that is the case language gives us reasoning and language gives us the human
02:10abstraction that is necessary video shows us the entire universe right like you know how things behave how
02:16water behaves how the laws of physics work when you combine them together then then video audio and
02:22language together is basically a chance to build a universal simulator so one application of that as
02:29you point out is is obviously for entertainment and to generate video to automate or or to to make
02:34digital the act of creating video but in addition to that it is the gateway to be able to build
02:40actually general purpose robotics that is the second area that we are very deeply focused on and now
02:46starting to build out a team and and and the research area of that and and if we have if you want to have
02:52any shot at building general purpose robots then the only way that happens is by giving them this level of
02:59general understanding of the universe so that they can they can you know reason in their head they can
03:03actually simulate every scenario right in their head like okay what would happen if i do this what would happen if i do
03:08something else this level of learning and reasoning is essential so video as it advances and as we
03:14scale these models they will not only get better and better they will get more accurate in simulating
03:19physics and this is the path to building physical intelligence and that's why luma's model mission
03:24is to build multi-model agi that can generate understand and operate in the physical world and that is
03:30the end goal amit what is the biggest blocker to that or the thing that keeps you up because at the
03:35the moment you got the money boy do you have the money 900 million raised we've also got promise of
03:39compute coming from saudi arabia is it about compute is it more about the talent that is what you need
03:44to push through to get to your mission yeah so there are two really important things for luma uh over
03:51the last couple of years we worked tirelessly to to solve the research problem for like you know really
03:56really great video now the next frontier for us are these omni models that are able to like you know
04:01reason and audio video language text together so the the research is is is like you know the first
04:07one that we pay immense amount of attention to now to solve the research of course you know people and
04:11and really really brilliant people is is one of those bottlenecks so we hire from some of the best
04:16places see we don't need thousands of people we need like you know maybe 200 or 300 really really
04:21brilliant people to to work on these problems so yeah that's that's the first one the talent and and and
04:28great aligned people that is the first one second is obviously compute so alongside with the series
04:32c announcement we announced that we will build uh with humane and collaboration with humane a two
04:38gigawatt compute cluster that is the largest compute build out in our space of world models and and video
04:45models and like you know this kind of physical ai and it is also one of the largest compute build out in ai
04:51period right so compute is one of the other things that is a big limitation ultimately multimodal ai
04:59will be a super set of llms will be a super set of the ai we have today it will require more compute
05:04than llms do right now so compute is the second very important input to our business that like you know
05:10we are working to not only shore up but solve in a way we can serve these models economically to a lot of people
Comments