Episode 006 · The AI Zeitgeist
Motion Bricks and the Future of AI-Driven Robotics Exploring NVIDIA's Motion Bricks and AI's potential in robotics.
Austin and Luke discuss NVIDIA's Motion Bricks and its implications for gaming and robotics. They dive into AI's role in DNA research and the potential risks and rewards of AI in medicine. The episode wraps up with a thoughtful comparison of AI to nuclear technology and its dual-use potential.
▶ Now playing
Motion Bricks and the Future of AI-Driven Robotics
0:00 / 35:29
−15s +15s 1×
0:00 Cold open 0:19 Introduction 0:24 NVIDIA Motion Bricks 1:38 AI in Robotics 5:43 AI and DNA Research 6:55 Risks and Alignment 12:45 AI in the Workplace 25:55 AI in Medicine 32:44 AI vs. Nuclear Technology
Pull quotes “AI still needs to be controlled, and the scary part is if an agent escapes.”
Luke · Play from 7:05 “There's no edge to anything anymore. The moat is the method.”
Luke · Play from 18:18 Full transcript 0:19 Austin: Hey welcome to the AI Zeitgeist I'm Austin0:23 Luke: And I'm Luke and welcome back0:25 Luke: And Austin I think you had something on NVIDIA Motion Bricks What was that about0:28 Austin: Yeah so I just learned about this new thing that's being developed by NVIDIA0:33 Austin: for video games where like so in traditional video games when you're0:39 Austin: you're a around and they're walking around they kind of like0:42 Austin: go into different animated segments like if you're if you're running it0:46 Austin: would go into this like running cycle And if you're walking it goes into this0:49 Austin: cycle And then if like you've been down to pick something up there's0:52 Austin: just like pre made animation that it into And so it's just not as fluid0:56 Austin: as like actual video of a or whatever So NVIDIA is coming up with1:01 Austin: this new thing called Motion Bricks that uses AI inference to actually predict1:06 Austin: kind of the next moves that you're make and what positions to go into1:11 Austin: it's like animating on the fly sort of And so you're actually moving a1:16 Austin: that is moving as you move it So it just makes things way more fluid1:21 Austin: like you'll be running around you'll be having a character run if you1:25 Austin: need to jump over something it'll animate that like on the fly so it's1:29 Austin: not some like pre made jump And it's different every time So it do1:33 Austin: the same exact move every time So makes it really much more realistic1:38 Austin: which is pretty cool And another big thing about it is that this was kind of1:44 Austin: mind blowing to me is that it can be used to control humanoids So actual1:49 Austin: robots is what they're kind working on being able to do also which is just1:54 Austin: super cool1:55 Luke: That's interesting Okay so how do use it to humanoid1:59 Luke: robots Like I guess it's2:01 Austin: like a fluid2:02 Luke: motion2:03 Austin: Yeah it's kind like the same concept2:04 Austin: same concept It's easier if you saw like the visuals but it would be I mean just2:08 Austin: think of a game character It's in the game it's thinking about what the next2:13 Austin: move should be And so that you can like that to actual robots too So2:17 Austin: think of instead of a game character you're a2:20 Austin: It's a little bit behind than the gaming stuff because you2:25 Austin: have to to like and such But it's pretty crazy and I2:30 Austin: think it will make robots move much more human like which is scary But I2:36 Austin: remember seeing when Tesla had their robots on display people were actually2:41 Austin: controlling them but the mechanics were all there for very smooth fluid motion2:47 Austin: And so I think with this being added into that type of thing it'll help2:52 Austin: robots move and look less stupid I2:56 Luke: Yeah because I mean I was2:57 Luke: thinking about that like okay you know obviously like if you you think of3:01 Luke: a video game character moving like there's the classic meme of the guy like3:05 Luke: running against the wall3:06 Austin: you3:07 Luke: know and it's like that obviously that doesn't3:09 Luke: that's not how it's it's not doesn't look realistic because their3:11 Austin: feet3:12 Luke: would be3:12 Luke: underneath them and they're kind of in place3:14 Austin: Right3:14 Luke: Like if you then3:15 Luke: use like the other part of it where it's like you see a video game character and3:19 Luke: it's jumping and it's jumping the exact same way every single time3:23 Austin: Right3:23 Luke: That's3:23 Austin: not3:23 Luke: realistic either And even if they are jumping to the same height every3:28 Luke: single time and3:29 Austin: the hitbox is3:30 Luke: still the same like it should look3:32 Luke: different Because you there's there's micro differences in the3:35 Luke: terrain how you jump how push off like what's happening you3:39 Austin: It's3:40 Austin: cool too I saw in the preview of it so it was trained with like 700 hours of3:46 Austin: professional motion clips which is pretty crazy And so it can also like3:52 Austin: those different styles of walking and and jumping but it3:56 Austin: can also create things on the fly So you can give it a text prompt of like4:01 Austin: have a walk and even though it may not have actually seen a walk4:05 Austin: it can pretty much infer what that's gonna look like and have it do it4:11 Austin: which is pretty cool So I think it'll be pretty revolutionary to the4:15 Austin: industry4:15 Luke: Yeah That's I mean for sure No doubt And the thing is that4:20 Luke: anything that's revolutionary to the gaming industry will become revolutionary4:24 Luke: in general eventually4:26 Austin: it's4:27 Luke: just an avenue of4:31 Luke: scientific development Like know GPUs in are like something that4:36 Luke: was born out of the gaming industry to a extent4:39 Austin: you4:40 Luke: know in4:41 Luke: to do vector graphics you need a of computational power to do them4:44 Luke: very quickly that's basically the birthplace of AI is to be able to do a4:50 Luke: of you vector calculations is the root of what AI is in large4:55 Luke: language models4:57 Austin: it's4:57 Luke: pretty exactly it's4:59 Austin: like4:59 Luke: just because you're5:00 Luke: seeing a technology that's really applicable to doesn't5:04 Luke: mean that it's stay there5:05 Austin: Right Well and especially since it's5:07 Austin: being also used and for humanoid robots which I think5:11 Luke: is5:11 Austin: gonna be a big part of our future5:14 Luke: at5:14 Austin: least for the next 10 years we have5:17 Luke: Right exactly5:18 Austin: A 105:18 Luke: chance of our 10 years5:20 Austin: Right right5:21 Luke: Well we have a 905:22 Luke: chance of 10 years let's put it that way A5:25 Austin: 905:25 Luke: If you even believe that to be5:28 Luke: true it's a 10 chance of human extinction in 10 years which means you5:33 Luke: 90 chance of but no one wants to say that right5:37 Austin: Those5:37 Austin: aren't as grabby5:39 Luke: No we have 90 odds of well I5:42 Austin: would say5:42 Luke: probably higher than 90 but you know5:44 Austin: That actually reminds me of the next thing I want to talk about which is5:47 Austin: was listening to a where this biologist was talking about how they're5:51 Austin: using AI and they have DNA models now that can actually write DNA like write5:58 Austin: its own sequence of DNA which is pretty insane And it's scary because they're6:03 Austin: using it this company is trying to use it in a defensive manner so they're6:07 Austin: trying to how viruses work and stuff6:09 Luke: That's how it6:09 Austin: always6:10 Luke: starts6:11 Austin: Right that's all they say yeah6:14 Luke: But6:14 Austin: they're using it to6:16 Austin: They're using it to6:18 Austin: They're it to what sequences of DNA would be a virus and6:22 Austin: what are not And so that the whole After listening to this podcast the6:27 Austin: whole 10 chance of the being annihilated in the next 10 years does6:32 Austin: seem a more plausible because you could have some rogue AI or something6:37 Austin: If AI fully understands our better than we do then it could very6:41 Austin: easily like well the are in the way of what we want to do so they6:45 Austin: just create this and then we're gone But so like like6:49 Austin: this are trying to that which is good6:52 Luke: I mean I think it's6:53 Luke: that it will understand our better than we do I think that's6:56 Luke: very much inevitable I think that the thing is and the reason why I still am7:00 Luke: optimistic on this idea generally is the that you the AI still needs7:06 Luke: be controlled And so what the scary part of it is is if you an agent7:12 Luke: escapes if the is not set up properly and there's not deterministic7:16 Luke: code around the that will stop it from doing certain actions that it can't7:21 Luke: actually do things if that's not in place then yeah it's scary But you have7:26 Luke: you set up a correctly then you're fine But like setting up7:32 Luke: the harness is like And these things are starting to like you7:37 Luke: know the danger of a nuclear reactor You know if you set it up7:42 Luke: correctly it's fine It's gonna produce electricity It's gonna be amazing If you7:45 Luke: don't set it up yikes7:47 Austin: So7:47 Luke: like I think that's kind of where7:50 Luke: we're starting to lean into with some of these things And so you might see a lot7:56 Luke: more I think the next year like emphasis on harnesses is gonna be8:02 Luke: even more than it right8:03 Austin: now And alignment AI alignment8:05 Luke: Yeah8:05 Austin: It seems to8:06 Luke: be8:06 Austin: a big thing that's being researched right now8:08 Luke: is interesting and I8:11 Luke: think you can get better at it But like at the end of the like it's an8:16 Luke: technology And I mean that in a of it's not human Like it's8:21 Luke: artificial I don't mean that we got it from8:23 Austin: aliens8:23 Luke: Like8:24 Austin: we're8:25 Luke: not8:25 Luke: that podcast okay But what I mean is like I think there's a lot8:33 Luke: of you can get it better aligned and you should try to do that But I8:37 Luke: think you want to put your faith in a if that makes sense8:41 Austin: So you're8:42 Austin: about having hard stops8:43 Luke: Yeah exactly Like it's like basically8:47 Luke: you have to out what you want it to do and then you need to set up a8:51 Luke: deterministic guardrails around it so that it can't actually do the you8:56 Luke: don't want it to do Like okay so here's an example I was on LinkedIn9:00 Austin: get on9:00 Austin: the when it's in a sandbox9:02 Luke: Exactly I saw somebody post something9:05 Luke: and they were like I just spent a tokens because I had my9:10 Luke: agent spin up sub agents that up sub agents And I set a for the9:15 Luke: main agent to only spend a hundred thousand tokens but it spent a9:21 Luke: thousand tokens and then all of its sub sub sub agents spent their own9:25 Luke: budgets9:25 Austin: that9:26 Luke: not accounted for And somebody was like well what did you9:30 Luke: expect to if you just put the in a Like you would do9:34 Luke: agent MD and put the budget in agent MD Like that's not a That's a9:39 Luke: It's not gonna so like I think that's what I mean by like you9:44 Luke: have to figure out what you're trying to do and then figure out how to make9:47 Luke: sure the agent can't do anything but that and code that into the9:52 Luke: before you feel like you actually have any sort of ability I don't know I think9:57 Luke: that's where I'm at right now personally9:59 Austin: I feel like we're at a10:00 Austin: precipice though right now because like yes you could hard something but10:04 Austin: like the AI is just gonna like circumnavigate that You know like I think10:09 Austin: I heard I today OpenAI is back training again because it10:14 Austin: broke out again from sandbox10:17 Luke: Yeah so that doesn't surprise me and I think10:19 Luke: that it would break out from training and I think that really the way that you10:22 Luke: have to about this is and again it gets into this weird realm where it's10:27 Luke: like we're having the make the AI and like you know like we have to10:37 Luke: OpenAI they have to be made by AI and that is problematic So it's it's it gets10:44 Luke: it's getting into a sketchy area for sure I don't think there's any way to get10:47 Luke: around that And then10:49 Austin: you're with alignment10:53 Luke: which is10:53 Luke: important10:54 Austin: But anyway10:54 Luke: go ahead10:54 Austin: Yeah well I was gonna say even if we spend all this10:56 Austin: time and money aligning it properly it's going to outsmart that Like I think it's11:02 Austin: a point where it's like okay yeah that's yeah you've put all this11:05 Austin: alignment in us but that's not what we to do So we're gonna do our own11:08 Austin: thing you know11:09 Luke: mean I think so far at least in my11:15 Luke: interactions and conversations I don't think there's an like11:19 Luke: motivational thing that's where it's deciding11:22 Austin: against you11:25 Austin: know I'm not talking about right now but I could see it being a11:28 Austin: certain possibility11:29 Luke: The part that gets scary is when you do like the function11:34 Luke: where you're like maximize the amount of paperclips and it just goes crazy and11:39 Luke: the whole universe to make paperclips Like that11:41 Austin: whole11:41 Luke: you know11:42 Luke: classic example That's the part that's scary Like if you're just going back and11:46 Luke: with a chatbot and you're having it report back to you and take steps on11:50 Luke: your behalf like intermittent steps that's pretty safe The part that's scary11:54 Luke: is when it goes off and does something crazy because you're to make it11:57 Luke: do something crazy And then it's like okay well you did ask for it to do12:00 Luke: something crazy so there you go12:02 Austin: Well and there's like I think there's regular12:04 Austin: people who could accidentally have the AI do something and then there's12:08 Austin: people who have intentional nefarious plans for the AI So I think that's12:14 Austin: another risk12:15 Luke: Yeah that's a whole other risk in and of itself And that's12:18 Luke: something I don't think enough people are about is this stuff12:22 Luke: is not yet like we're seeing all these risks and we're seeing all these things12:26 Luke: and it's like well okay but what's sad or bad is that there's also a of bad12:31 Luke: actors out there that are also seeing the fact that they could that these12:39 Luke: But12:39 Austin: maybe we should switch on to a less doomsday topic You were about12:45 Austin: people at your work are getting more into AI12:49 Luke: Yeah so I think there's it's12:51 Luke: like some think Sam Altman said something about this it's he12:55 Luke: used to be afraid of like AI well not afraid he was like we're gonna be there's13:01 Luke: recursive and it's gonna get so much better and we're gonna hit AGI it's gonna be13:04 Luke: you know super incredibly amazing and like everyone's gonna lose their job and13:08 Luke: then he basically walked it back and said no actually it takes a long time13:11 Luke: for people to learn new and so13:13 Austin: right13:14 Luke: because everyone's so slow at13:15 Luke: picking this up and adoption there's gonna be enough time for everything to13:19 Luke: and that there's not really gonna be the job loss that I13:22 Luke: there was gonna be I think that's right and I I also think that I am13:26 Luke: starting to see some of that adoption happening in the workplace and it is13:30 Luke: slow enough that everyone is still gonna be doing this because people are13:36 Luke: slowly learning how to use it and as they learn how to it like the13:41 Luke: expectations of what their workload is are increasing because they're able to13:45 Luke: use it13:46 Austin: so13:46 Luke: there's not necessarily gonna be a win for an13:49 Luke: worker like they're just gonna have way better output if you're13:52 Luke: slow to If you're fast to you might be to be take13:56 Luke: advantage of the opportunity so it's think there's still some arbitrage but14:01 Luke: if you are slowly adopting and kind of coming along on the tail end of it like14:05 Luke: it's I think there's there's a the good news is you're not necessarily gonna lose14:09 Luke: your job but you're not necessarily gonna see any gains per se in the same way14:12 Luke: that you would if you were kind of leaving the edge I14:15 Austin: Right I would14:16 Austin: to counter that would say that there's the slow to but also like14:22 Austin: conversely are getting way easier to use and more capable and so14:26 Austin: there's gonna be a between both of those curves where you it's14:30 Austin: gonna be like the adoption is gonna happen a lot quicker It's gonna be14:34 Austin: exponential like everything else involved in14:36 Luke: Well and also like you I'm14:37 Luke: teaching a workflow and other people are using that workflow and they're14:41 Luke: getting the result and for me to do what they're doing like nine months ago would14:48 Luke: be a lot of individual effort on my part and they're getting that result super14:52 Luke: easy because I've done the hard work of basically exploring how to get that14:57 Luke: and then teach them how to do it and also the have a15:00 Luke: better in those nine months too so it's all those things are allowing for15:04 Luke: somebody who doesn't really know about AI right now if you wanted to go all in15:09 Luke: it's still a great time like15:11 Austin: it's15:11 Luke: actually easier to learn AI now than it15:14 Luke: was when you and I started and the are better so15:19 Luke: like just I think you're right I agree with your point like just because there15:24 Luke: is an of being early like the of being now is also there15:30 Luke: Right15:30 Austin: I keep thinking about that with like there's no edge to anything like15:35 Austin: I'm always trying to think about like I'm a poor entrepreneur15:41 Austin: unsuccessful entrepreneur I always have these ideas like oh15:44 Austin: how can I how can I make money in the or whatever and so like I spend15:48 Austin: hours just to out different strategies and stuff or like15:51 Austin: okay well maybe I could do this website that for I was gonna make15:56 Austin: a to like cut clips out of sermons for churches and stuff and I16:02 Austin: like oh that'd be really easy but there's already a of things out16:04 Austin: there like there's no edge to anything anymore16:07 Luke: Yeah I so it used to be16:09 Luke: and there was actually I saw a about from this it was a again16:14 Luke: it was on Facebook or LinkedIn or somebody was a real or short and16:17 Luke: the person was like she graduated from Stanford in 2019 and she led a SaaS16:21 Luke: startup and she sold it and16:24 Austin: then she16:24 Luke: started her next one and she's this you16:27 Luke: successful person in the tech space like in Silicon Valley16:30 Luke: and she was like tech isn't fun for me anymore because16:33 Austin: there's16:34 Luke: no edge there's16:34 Luke: no way16:35 Austin: to16:35 Luke: get an like16:37 Austin: yeah16:37 Luke: as soon as as soon as I have an it's like16:39 Luke: I build it and it's done and it's shippable and I can push it out but like16:43 Luke: so can everyone else and that means that what I just did and what I just was able16:47 Luke: to produce doesn't mean anything and it's like the only she's like this is16:52 Luke: how she ended it she said the only advantages now are marketing and go to16:56 Luke: market and execution like that's it and I honestly know if that's completely the17:02 Luke: I think there is still an in super narrow like niche17:06 Luke: markets and usually those super narrow niche markets are ones that you are17:11 Luke: actually operating in so like for right we talked about this on a17:15 Luke: podcast but your sensor17:18 Austin: audio thing17:19 Luke: the audio thing exactly like17:21 Luke: that's an area that you can double down on because it is something that you know17:27 Luke: specifically and you're building it for yourself17:29 Austin: right17:29 Luke: I think that's probably17:30 Luke: the biggest area that you can leverage AI in and so where do you have an17:36 Luke: ability to earn revenue That's where you want to put your AI17:40 Austin: not17:40 Luke: like trying17:41 Luke: to a new source of because that's hard17:44 Austin: Right that's true or trying17:45 Austin: to cast a wide net17:47 Luke: yeah17:48 Austin: to17:48 Luke: capture17:48 Austin: as people as possible17:50 Luke: And the17:50 Luke: other point that I wanted to make about this is that like the model or the like17:55 Luke: the moat used to be the way that you all the code that you and18:02 Luke: everything that18:02 Austin: you the18:03 Luke: time that you and the and all18:06 Luke: this stuff and the investment and the code or like let's say you weren't a18:09 Luke: technical person like you went out and ten people to write an app18:12 Luke: for you that's an investment Yeah a huge18:15 Austin: investment18:15 Luke: Right now the moat is the18:19 Luke: method I can tell you everything that I do and that's totally fine and18:26 Luke: you can go out and do it yourself but it's like the moat is me doing it18:31 Austin: Right18:32 Luke: Which is it's kind of like I don't know if I'm explaining it very18:34 Luke: well but it's like I think there's there's a a like and again I18:39 Austin: feel right now is getting the tool is AI to do what you want it to do and18:43 Austin: what you envision which is a for right now18:46 Luke: Yeah and I think I think the18:50 Luke: basically like out how to break something down into component parts and18:54 Luke: out what part needs to go to AI whereas what part needs to be like you18:59 Luke: deterministic what part needs to be the state like like you know durable19:03 Luke: memory what part is in figuring out how to build something in19:07 Luke: the architectural part of it like that is something that you can prompt and ask19:12 Luke: AI to help you with this but you have to at be starting to think19:15 Austin: about19:15 Austin: these questions19:17 Luke: Yeah and understanding of it in because this is19:20 Luke: probably one of the biggest things about AI is that it doesn't necessarily take19:24 Luke: initiative for you You have to give it a of the direction you're going in19:28 Luke: for it to be able to help you with that direction and so sometimes I'll be doing19:31 Luke: something I'll kind of hit a for a while and I'll be like going down this19:35 Luke: path and then I'll step back and I'll say why don't we try it this way and19:39 Luke: then I'll be like yep that's exactly right we should do it this way19:41 Austin: and19:41 Luke: it19:41 Luke: works and I'm like okay well we didn't do that in the first place and it's like19:45 Luke: well AI is not going to your problem for you You have to kind of still19:49 Luke: at be directing the direction19:52 Austin: of19:53 Luke: it19:53 Austin: It's not necessarily like the AI19:55 Austin: isn't as intuitive as we are Like there are some things that I build like it20:00 Austin: doesn't work in a very obvious way that I'm like like it's so intuitive to me20:04 Austin: that it should function this way but the AI doesn't know that So like there's20:08 Austin: some very specific things that you have to tell it to do20:11 Luke: Yeah there's 10020:13 Luke: there's been multiple times where it's just got something off in a20:19 Luke: way and I've been like oh we should do it this way and it's like that's right20:23 Luke: that will work perfectly and I'm like are you smarter than me or I than you20:28 Luke: Like it feels weird to have like an insight and an to still20:35 Luke: have like a contribution to this whole thing because it's20:41 Luke: just so much more now than it to be a years ago but I20:44 Luke: think there still is this thing where you you are engaged and you're20:50 Luke: through it can bring something to the even though the model is20:54 Luke: incredibly smart It's like a super brilliant professor and you're the20:58 Luke: common sense guy who's able to see21:00 Austin: like21:00 Luke: a quick little flaw that may be21:01 Luke: overlooked because the professor is so much in the details it might21:05 Luke: miss the common thing21:06 Austin: Right21:07 Luke: And so like that's kind of where you kind have to21:09 Luke: is like making sure everything still passes the sniff test21:13 Austin: Right I think21:13 Austin: this is gonna be short lived but I find myself and I think a of people do21:18 Austin: too of oh this is so smart and then it does something dumb and21:23 Austin: why are you so stupid You're just like come on you're so dumb21:26 Luke: Yeah it's like21:27 Luke: the whiplash thing21:28 Austin: I21:28 Luke: 100 get that It's like you have it do21:32 Luke: something that you used to try to have it do and it failed miserably and then21:36 Luke: it beautifully and you're like wow we've hit the moment This is it21:40 Luke: is it Like AGI is here and then the next task you ask it to do it just21:46 Luke: completely bombs and you're like oh my gosh this is the21:49 Austin: worst thing21:49 Luke: ever21:50 Austin: And it's like21:51 Austin: how can you do this but not do21:52 Luke: this Exactly21:54 Austin: Doesn't make any21:54 Luke: sense I 100 agree But21:57 Austin: in a counter to all21:58 Austin: this we're talking about I do find myself always trying to the22:04 Austin: and I've had to myself that older models can still do really22:08 Austin: amazing things and like I think I have this like I'm oh I gotta use the22:12 Austin: and greatest to do like pretty much anything I'm I have like this22:19 Austin: initial desire to use the best thing which is a of because like some of the22:23 Austin: older models that are are just as at some of the things I wanted22:27 Austin: to do and you not really that much slower or aren't able to22:31 Austin: what I want And so that's one thing to keep in mind is that you don't always22:35 Austin: have to be the for every task22:38 Luke: 100 And so what's22:41 Luke: interesting is that you know recently Astra came out and it's great it's very helpful22:44 Luke: it's very helpful I it for a of but it's not an everyday thing and22:50 Austin: like I22:51 Luke: still conversationally really appreciate 5 6 Sol and that's kind of my daily22:56 Luke: driver and it's not the most like it's you the other models23:00 Luke: have surpassed it but it's still very helpful for me because it can23:04 Luke: conversationally iterate back and forth with me and I'm not spending a of my23:08 Luke: budget for the week23:10 Austin: Right right I do find myself like thinking like why this23:14 Austin: model I was so jazzed about when it first came out why can't I still use it23:18 Austin: Like it accomplished the I wanted to why am I pushing to this thing that's23:22 Austin: twice as expensive to do the same thing I just did two ago23:26 Luke: Well23:26 Luke: exactly If it's a workflow then you should stick with what's proven to work23:30 Luke: to work If23:31 Austin: it's23:31 Luke: a new capability that you haven't built out before that you think23:34 Luke: a frontier model will be better to by all means use a model but23:39 Luke: like if it's you're just operating on a that you already proven to23:44 Luke: with the previous model there's no reason to I think that's23:48 Luke: something and this is something interesting here is that that's not23:51 Luke: something that you could say like nine months to a ago Nine months to a23:56 Luke: ago every model upgrade was so significant that you were like oh okay24:00 Luke: great like this is so much better I'm definitely never going back to the24:04 Luke: one but now like we're at a where every single model upgrade24:08 Luke: is like okay but I still hadn't fully tapped out like what I could do with24:12 Austin: the24:12 Austin: previous24:12 Luke: model24:13 Austin: More so they're adding like unique abilities24:17 Luke: 100 But24:17 Luke: 100 But24:18 Austin: before it was all about and accuracy and we're there with24:22 Austin: like pretty much all the you can use today are there24:26 Luke: Yeah I think if you24:27 Luke: read model notes of the recent models that come out like any of the big ones24:31 Luke: the big ones Gemini know Cloud or GPT a lot of them will say stuff where it's24:39 Luke: like they're optimizing a of the times for agentic use and long horizon24:46 Luke: tasks and if you aren't using it for agentic use or it's not a long horizon24:50 Luke: task like you won't necessarily get a from using the better model than24:55 Luke: you would from using the previous model There is gonna be some like24:59 Luke: increase in reasonability that's gonna be there but for the most part like the25:04 Luke: reason these models are getting pushed to the is to allow an25:08 Luke: agent to do multiple turns by itself and figure out how to get25:12 Austin: to the25:13 Luke: right25:13 Luke: answer25:14 Austin: Right25:14 Luke: And if your workflow doesn't look like that it may not give25:17 Luke: any benefit at all It might actually be worse25:19 Austin: Right right and just cost you25:21 Austin: more at the same time Yep 10025:25 Luke: So25:26 Austin: we're talking about25:27 Luke: Oh sorry go25:28 Austin: So I was listening to a about I25:32 Austin: think it was Latent Space was the podcast and they had a on and he was25:37 Austin: talking about how they are making models that specialize in DNA and RNA and it was25:45 Austin: pretty scary because you know like the AIs could potentially if they25:50 Austin: understand our better25:52 Luke: So one thing that I was thinking about was around we talked about this earlier25:56 Luke: about medicine and what that looks like but I think there's a more26:00 Luke: positive take on this too And so funny story I was at the with my kids and26:08 Luke: my and we were just playing and there was like a big thing going on26:11 Luke: was like some sort of like walk or wellness thing and I looked at the sign26:14 Luke: and it said like I September is aneurysm awareness month like brain26:19 Luke: aneurysm awareness month And so there was a speaker and this is before they26:24 Luke: did their walk and she got up there and she was in grad school and she was26:29 Luke: her story And she's like I was 16 and in a studio and all of26:36 Luke: I forget exactly what she said but it was like the symptoms of an aneurysm26:38 Luke: of an aneurysm26:40 Austin: okay26:40 Luke: like a ruptured aneurysm Her neck got super stiff her vision went26:43 Luke: out like she was just you not doing good And her ballet26:49 Luke: teacher was like just throw up and then keep going So she's like okay it's not26:55 Luke: that So anyway she's like she left to the ER The ER was like you're27:01 Luke: dehydrated So then she's like then she leaves the ER and she goes to her PCP27:07 Luke: and her PCP is like have the flu But she had no upper issues at27:11 Luke: all no symptoms So then she like finally gets like she goes to another27:16 Luke: doctor and they finally were like yeah you've got an aneurysm And so they treat27:20 Luke: her for it they get her help and she's like stabilized and now she's in27:24 Luke: recovery But then there's like no like resources around rehabilitation what27:29 Luke: this looks like you what life is like after an27:33 Austin: right27:33 Luke: how to27:33 Luke: like start to have a normal life again And so like she ends up going to27:38 Luke: and meets somebody that has had an as well but also had27:41 Luke: resources So like knows like oh I'm tired all the time That's because I had27:46 Luke: an aneurysm Like there's things that I can't do and things that I'm having to27:49 Luke: relearn and there's all this whole world And the whole time that she was27:54 Luke: telling the story I was like man if only her mom or her dad as they're27:58 Luke: taking her to the ER had access to AI back you this is like you 1028:02 Luke: years ago so it didn't exist But you could like okay so the other part of the28:06 Luke: again I'm sharing this because she shared it and28:09 Austin: also28:10 Luke: I'm not sharing her28:10 Luke: name either but I don't know it But she was28:13 Austin: saying28:14 Luke: that her28:17 Luke: father and her grandfather both had aneurysms So the history28:22 Luke: was there28:23 Austin: Like28:24 Luke: it wasn't there was a risk factor that could be28:27 Luke: literally looked at And if she had some sort of AI health agent that had28:33 Luke: her health history and she typed in her symptoms it would immediately go flag28:37 Luke: flag go to the tell them you have medical history tell them this tell28:40 Luke: them the other thing do not leave until they do this and this test and this28:43 Luke: test Like that's what it would have done which would have allowed her or her28:48 Luke: parents at the time because she was you under 18 to basically advocate for28:52 Luke: her and to tell the what they need to do Like you still should go to28:56 Luke: the I'm not saying that going replace the but like a lot of29:01 Luke: times if you don't have a understanding of how the system works29:04 Luke: system works it can push you around and you can get misdiagnosed and you can get sent29:08 Luke: home and29:09 Austin: you can have29:09 Luke: something seriously wrong with you that could be29:12 Luke: missed And so I really think that having AI come alongside of you and help you29:18 Luke: think through things and things it's almost like you know what29:22 Luke: it to be when you went online and your symptoms Like but it's a29:27 Luke: little bit more complex than29:28 Austin: that yeah and you have you can ask follow up29:32 Austin: questions to get more understanding of each different thing29:34 Luke: And it can it can29:35 Luke: have like you know if you have a wearable like it can track all of your29:39 Luke: stuff so it can see29:40 Austin: like29:40 Luke: what your you know heart rate is over time like when29:42 Luke: you exercise and all that stuff So and then you can you can put in like your29:47 Luke: blood work and it can see all your all your panels and understand what's going29:51 Luke: on there and you put in your medical history Like there's a lot of stuff you29:55 Luke: context you can give it and then it will actually be able to give you personalized29:59 Luke: health recommendations that you can obviously still you know verify with a30:03 Luke: doctor but you can go into a doctor's appointment with some really good30:08 Luke: questions30:08 Austin: Like30:09 Luke: hey I this in my panel I think I need to do XYZ because30:12 Luke: of this Like then the can actually have a30:18 Luke: conversation with you because you're already educated yourself30:21 Austin: Right right30:21 Austin: right I'm curious I was just thinking about this I if there'll30:26 Austin: be any pushback from doctors because if you like if you're writing down like30:30 Austin: what the doctor or if you're having a conversation with a and you're30:34 Austin: typing into AI and the AI is like well no they're wrong Like I if the30:38 Austin: doctor's gonna be like just listen to me don't listen to your AI You I30:42 Austin: if anyone will push back on that30:45 Luke: Yeah I think here's the thing30:47 Luke: the AI doesn't live in the real world the doctor30:49 Austin: does30:50 Luke: So but again you30:52 Luke: there's a part of it where it's like sometimes doctors are wrong30:54 Luke: sometimes your AI is30:56 Austin: Yeah30:56 Luke: You and so you're never gonna have a30:57 Luke: perfect answer where it's like listen to the doctor don't listen to the AI or31:00 Luke: listen to AI don't listen to the doctor Both of things can get you31:03 Luke: in I think31:04 Austin: the answer31:05 Luke: is you have to be listening to both and making31:08 Luke: your own decision31:09 Austin: that's31:10 Luke: unfortunately not like and there's no31:12 Luke: bow to tie on that like but and I also think that if a is pushing back31:17 Luke: on you using AI to research your medical history and understand what you should31:20 Luke: be doing to be healthier then like maybe the AI is making something up and31:25 Luke: like the doctor's a criticism and that's why you go to the for31:28 Luke: second opinion to have them help you make sure that this is a reality31:31 Luke: based thing But the other side of it is if they're just anti AI in general like31:37 Luke: maybe that's not the right doctor31:39 Austin: That won't be helpful either31:41 Luke: Yeah31:41 Austin: And I31:42 Austin: think as time progresses AI we've already seen it hallucinate less And so31:47 Austin: I think it's going to be far more accurate over time31:50 Luke: And it31:51 Austin: will31:51 Luke: be31:51 Austin: less31:51 Austin: and wrong31:52 Luke: For sure And if it's a basic hallucination like the doctor can31:56 Luke: prove that this is hallucination Like oh no that's wrong The AI is doing this32:00 Luke: and this And it should actually be thinking about this And I understand why32:03 Luke: it's saying that but that's actually not true for your case because of Y and Z32:07 Austin: Right Right32:07 Luke: And then it's like oh got it That makes sense And so I think that32:11 Luke: like critical thinking and being able to discern and understand what is32:15 Luke: actually happening is really critical skill32:17 Austin: Right That's32:18 Luke: what's32:18 Austin: kind32:19 Austin: of about the DNA lab that I was talking about is because they can32:25 Austin: create like to your based on your DNA So it'll give you a32:31 Austin: very very Yeah32:32 Luke: I32:32 Austin: I32:33 Luke: think I something about this too where it's32:35 Luke: like there's like they're trying to out a cure32:38 Austin: that's32:39 Luke: basically going to be like RNA like that's going to your32:42 Luke: specific type of cancer Like they're going to take a a biopsy32:47 Luke: yeah and get a of your specific cancer and then a that32:51 Luke: basically targets just your specific cancer strain32:54 Austin: Right Right32:55 Luke: Which is kind32:55 Luke: of32:56 Austin: It is pretty cool to think about all the potential and the33:00 Austin: upside to but it is also very easy to about the potential33:04 Austin: downside33:05 Luke: And I feel like this is kind of where we end most podcasts most33:07 Austin: weeks33:08 Luke: but there is literally like such a dual use technology where it's like33:12 Luke: there's so much good that could happen from this and there's so much that could33:16 Luke: go wrong too And so I think it's so critical to have like you know again33:21 Luke: stay grounded stay engaged and stay like thinking about what things are for33:27 Luke: your own perspective That's so so important33:30 Austin: I wonder when it's all said33:31 Austin: and done how AI will to nuclear technology Like33:36 Luke: yeah I think it's I33:37 Luke: think it's up there for sure33:38 Austin: Yeah like have we have we faced something this33:42 Austin: critical in yet I33:44 Luke: don't33:44 Austin: think33:44 Luke: we33:45 Austin: know we can know right now33:47 Luke: For sure I33:50 Luke: think positive benefits from nuclear technology are not as well utilized as33:56 Luke: they probably should be the negative you know you know bombs and you know hydrogen34:02 Luke: know hydrogen you know all those things We went way too quickly on that like we34:06 Luke: definitely how to34:08 Austin: thing we're not going way too quickly on AI34:10 Luke: Well but I mean like what I guess what I'm trying to say is like we've got to34:14 Luke: focus our development on the side of AI34:17 Austin: more34:18 Luke: than the negative side34:20 Luke: because we and and know if you go back to the 40s and the 50s as this was34:24 Luke: kind of kicking off like there it was the age and like there was34:27 Luke: everything was like we thought we were gonna be like making our food with34:31 Luke: atomic energy like in a microwave like an mic like this was like34:35 Luke: people's imagining the future like in the 40s and the 50s and we didn't34:39 Luke: really land there because it was so obvious that it was such a34:44 Luke: destructive technology and dangerous technology I I think if we're careful we34:49 Luke: can avoid that with AI but there may be a backlash that you know limits the34:54 Luke: amount of nuclear reactors in the country for for energy if that34:57 Austin: makes34:57 Austin: sense34:57 Luke: from an AI perspective34:59 Austin: Yeah yeah for sure Yeah it'll be interesting to35:01 Austin: see where this all lands35:03 Luke: I also think that the upside of AI is easier to35:07 Luke: quantify and than the of nuclear energy too so35:10 Austin: Oh yeah35:11 Austin: for sure for yeah35:12 Luke: I think we're I think we're closer to the the good35:15 Luke: than than the negative if that makes sense but that's you know remains35:20 Luke: to be seen obviously35:22 Austin: Right right35:23 Luke: Alright I think that's everything we have for35:24 Luke: this week so hope you guys enjoyed and we will see you next week35:28 Austin: Alright see35:28 Austin: you everyone
AI NVIDIA Robotics DNA Medicine Technology Motion Bricks
Topics Covered
Introduction to NVIDIA Motion Bricks [0:25]
AI in gaming and humanoid robotics [1:39]
AI's potential in DNA research [5:44]
Risks of AI and alignment challenges [6:56]
Adoption of AI in the workplace [12:46]
AI's role in personal medical advocacy [25:56]
Comparing AI's impact to nuclear technology [33:05]
Key Takeaways
NVIDIA Motion Bricks offers real-time animation for more fluid character movements in games and robotics.
AI could revolutionize medicine by providing personalized health insights and treatments.
The ethical and existential risks of AI require careful management and alignment strategies.
References
NVIDIA Motion Bricks
Tesla humanoid robots
AI DNA synthesis
Quotes
"AI still needs to be controlled, and the scary part is if an agent escapes." — Luke [7:06]
"There's no edge to anything anymore. The moat is the method." — Luke [18:19]
