Artificial Intelligence Livestream - Bilawal Sidhu - AI & Creativity

Bilawal Sidhu: Welcome back everybody to the Artificial Intelligence live stream.

Jon Radoff: I am Jon Radoff. I'm the CEO of a company called Beamable. We make live services infrastructure for game companies. I'm accompanied here by Oscar who keeps us on track through every one of these episodes and I'm so excited today because I have one of my favorite people in the entire artificial intelligence landscape and that is Bilawal Sidhu. Bilawal is hot off talking at Harvard XR. He was at TED interviewing Eric Schmidt. He is one of the people who I really started watching early on in the formation of generative AI over the last few years because he did so many interesting things actually applying these technologies and not just in a simplistic way actually chaining together the different technologies to do really interesting work. So Bilawal thanks for joining me on this once again. We talked I think a couple

Bilawal Sidhu: years ago back when I was a YouTuber and I didn't do things live. Now we do things live

Jon Radoff: so that everybody can actually ask gotcha questions along the way from the audience.

Bilawal Sidhu: Thank you so much for having me. Thank you so much for having me. Thank you for having me. Thank you for the generous introduction and yeah it's great to be here. Yeah two years ago I think right when this generative stuff was like popping off and hitting the mainstream and guess what that train's been just it's going choo choo choo and seeming to proceed without

Jon Radoff: any signs of slowing down. So it's great to be here. So it wasn't hype. It seems like it's lasted a few years now or if it's still hype it's the hype train that won't stop I guess. But I don't I use chat GPT pretty often. I don't know if it's because I'm a fan of chat GPT. I don't know if it's because I'm a fan of chat GPT. I don't know if it's because I'm a fan of chat GPT. I don't know if it's

Bilawal Sidhu: every day. I mean I'd say at this point probably 10 to 20 times a day and I and it's not even the

Jon Radoff: only one I use. I use and I use generative art programs for doing just day-to-day work. So like generative AI has certainly impacted my life. I'm one of the guys who says thank you to chat GPT apparently. Me too. Costing millions of dollars to. Hey it does increase model performance right.

Bilawal Sidhu: So like I think those extra tokens are worth it.

Bilawal Sidhu: Just start kind of diving right in in terms of like creative workflows and how generative AI is

Jon Radoff: impacting it. But maybe use that as your opportunity to reintroduce yourself to everyone here too. And because that's where you got started. That's how I learned about you is like the shift from some of the tick tock work you were doing and then starting to use generative AI. So talk about that what you've learned along the way like and what's changed in the last couple of years since we did a YouTube conversation.

Bilawal Sidhu: I love it. This is the perfect tia. And since I'm hot off of this talk I gave at Harvard XR I got some slides for you. Visual umami for your audience. So let me just go ahead and share my screen here. There we go. And here we are. I mean so yeah to give a bit of context to your viewers here is like you know I got into visual effects early on as a kid. You know I was 11. I was into flash animation. People might remember you know goofy vector animation on the web. And you know I saw this show called Mega Movie Magic where basically they were showing digital compositing and the earliest instantiations of computer graphics. And I realized that same computer I was using for very cheesy you know cartoony stuff could create visuals indistinguishable from reality. And of course at that age it meant very cheesy you know fan films like this filmed with my camcorder and my homies in the backyard. But that is basically the theme that has kind of stayed with me throughout my career. So you know I've spent a decade in tech. Predominantly at Google doing the utilitarian version of what I just talked about. I would call it blending the physical and digital world. Working on everything from VR to 3D mapping. And yeah I'm a creatorpreneur now and entertain on YouTube and TikTok and educate on LinkedIn, X and a bunch of

Bilawal Sidhu: other platforms. But really to focus on sort of creation workflows I think we're seeing this sort

Bilawal Sidhu: of you know continued democratization that's become a huge buzzword. But I think it's absolutely right. But at the same time

Bilawal Sidhu: I don't think it's like you know I like to say this has never happened before yet has happened

Bilawal Sidhu: time and time again. And you see sort of the essence of the core idea repeat itself. So maybe it was manual rotoscoping back in the days of Alice in Wonderland and it's now generative AI video to video workflows with stuff that Corridor Digital is doing. But it's this like kind of trend

Bilawal Sidhu: towards simplicity and ease of use right. And so what was a complicated multi-step workflow in After Effects is a couple clicks and new tools like Pika. To bring it back for

Bilawal Sidhu: a full circle to your question. I mean it's really awesome to see how much more I can get done as like a solo creator. Like on the left you see sort of my old waterfall workflow for you know creating visual effects. You had to use a bunch of specialized tools and go through this sort of like you know serial like workflow. You know this waterfall process to get stuff done. And a bunch of that's collapsed into a handful of tools. And that's been very exciting for me because you know contrary to what. Immediate intuition. Would tell people that hey like suddenly I'll have all this extra free time. That's been the opposite of what I've experienced is like suddenly you can just do more. So I'm doing more on more platforms. And so like a great example of this is this like video that I made last year that went stupid viral. And it got like 30 something million views right around this like Miami mall incident near the new year that happened. Where people thought aliens were overrunning Miami mall. Basically is mid journey and runway to make this video. And there was a moment in the zeitgeist that it made sense to make this video. If this would have taken me a week to make it. I just wouldn't have made it. And so here's just a couple of other examples of stuff that I've done that would not be possible or I certainly would not have made without AI creation tools. And I think you're seeing that on the high end too where indies like me can start doing really cool stuff. But then like studios can set whole new standards all together.

Jon Radoff: So we're not freed to just go live in the utopian society quite yet with AIs just running everything for us. But the value of our time has increased. That's sort of the thing that really I took away from this conversation I had with the founder of Perplexity. And when he explained it to me he was like you know economically speaking what this is really going to do is just increase the value of all of our time so that everything we do is much more valuable and interesting as a result of it. We'll see in 10 or 20 years whether AI takes over completely I guess. But yeah. And by the way that's

Bilawal Sidhu: a really good point. I would like to circle back to maybe we end on a more philosophical or kind of future looking note because I think you know it is an interesting question when most economically productive tasks are increasingly done by machines and maybe a small number of people. What does that mean right? Like yeah there's like yeah this person at a big lab I was talking to was basically like yeah I think like in 10 years there'll be like a thousand PMs and like 5,000 designers and maybe 10,000 engineers and that'll be the tech industry. I'm like wow. Holy shit what is everyone else going to do if this is like the future that we're going towards. So yeah just to maybe throw a little teaser out there for. Yeah no let's definitely

Jon Radoff: come back to that because I think the economics of it are interesting. I'm not sure I buy that argument but I'm going to let it process for a little bit. Let it percolate. Yeah. So sitting here in 2025 what do you think these technologies mean to everyday creators? Like not the people who like you doing millions of millions of audience members looking at content like just the everyday person who needs to do a PowerPoint deck or in our case we have a lot of people in the game industry you know so I can tell you the answer there is like production workflows are getting a lot faster especially in the prototyping phase. 3D graphics it's still really really hard to do at scale. Yeah. Production wise in like AAA games but it is speeding up the processes quite a bit. Some people tell me that having AI is like having a lot of really talented interns that you can like shape their behavior in interesting ways and get useful results out of it much faster than the alternative which is human interns a lot of the time. What do you what do you think this means to to most folks out there. I mean I I think you nailed

Bilawal Sidhu: it with the intern analogy. It's like uh you know I like this meme that was going around and I made variation. of it because uh it was like nick fury standing and having that like video chat conference with like a bunch of like you know ominously lit personalities on the screen and appended on top of it just composite on top was like the various vibe coding tools and so like you know the way the message i'd give to you know everyone from the indie creator to you know high-end creators is just like you gotta lean into the tools first off and they do increasingly feel like i'm standing over the shoulder of a human and telling them what to do and they're doing them for me and i think a great example of this is like the exceedingly viral multimodal image generator update to chat gpt 4.0 and then also the same thing with gemini you know gemini was touted by

Bilawal Sidhu: google to be this natively multimodal model that can do multiple modalities in text image you know audio and then do the same thing out output as well and that's cool for me because that collapsing

Bilawal Sidhu: trend is sort of continuing like like even a year ago or even right now for maybe certain types of production workflows you might have to learn comfy ui where you're wrangling a bunch of these like purpose built models to kind of do something as simple as like style transfer and now i just like you know say uh here if we could pull this image up very quickly um oscar like now i just say hey turn us into roblox or gta3 or minecraft or whatever and the fact that it just works is kind of like wild to me and i think like people don't appreciate outside of ai just how hard something like this was to get right at a good hit rate in comfy um and so like these tools are more accessible than ever before and i think the other beauty of this is like with that collapsing you don't have to spend the time that a bunch of other technical artists spent figuring out how to tune these knobs and dials and control net and taking custom lauras and like pulling that all together and figuring out the right weighting and then doing some compositing to get the final result you just describe it like you would as a creative director to a human and get some very very cool results and i think that's um very very uh exciting here let me just figure out how to use this and so that is kind of cool right and so people should be playing with these tools the other thing i'd say is if you're like a novice creator and you've been very intimidated by learning some of the more advanced tools i mean i would be remiss if i don't bring up like vibe coding and mcp right like it's like kind of wild that you've got these llms that can now interface with these existing classical applications and you tell it what you want like you provided an image reference of hey this is the kind of scene that i'm trying to build or hey this is the kind

Jon Radoff: maybe just double click on mcp a little bit because i think it's still a new concept to some people explain what that means and how it's different than just like iteratively

Bilawal Sidhu: chatting yeah yeah totally think of it like this it's like you know anthropic describes the

Bilawal Sidhu: marketing language as like it's the usbc connector for ai and i just view it as an interface for you

Bilawal Sidhu: know your large language model of choice uh claude obviously came up with a standard but google's you know kind of adopting it and you can use it through cursor and then use a bunch of other llms it's a way for your large language model to basically use tools like blender for example

Jon Radoff: and so it's

Bilawal Sidhu: you can also use this aten統 or secretary BRA up on any other platform

Bilawal Sidhu: but you can obviously use a lab in the cluster however if you like that ldvc thing is gonna take you to the virtual world and you can use baby okay all right well a quick question we ran into something when you were talking earlier ldva and it was trying to make it abysmal but um kind of in logarithmic ways if i were trying to like like do a likeBIART program uh for example if I went to the sport program plus image references, for example, these systems can now control Blender for you. So to make it concrete, the example you're seeing on the right is basically, you know, a tripo has this amazing text and image to 3D, you know, API that you can use, but, you know, kind of collapsing those steps even further, you now just tell Claude, hey, here's the type of scene I'm trying to build. And then it decides, it decomposes your image reference into, hey, I need these five different kinds of assets. I'm gonna kick off jobs to like generate using text to 3D each of those individual assets, and then iteratively start refining and placing those objects in a 3D scene until you end up with something like this. Now think of that, right? Like how powerful that is as a new creator, that what would have taken you two or three weeks to learn all of these intricate steps, you hit your time to first compelling render way faster. And then you've got a starting point based on Eurovision, not some canned tutorial to start figuring out what do the dials do, or again, just keep using Claude or your LLM of choice as a way to interface with these existing tools. So that's remarkable. You're seeing this for Blender. I don't think we're that far off from like a cursor for video editing too, where like these LLMs will just be able to do the edit for you in Resolve or Premiere or whatever. And that's super exciting.

Jon Radoff: I'm at the point where when I use a software application and it has a feature that's just buried in the internet, it's like, oh, I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do this. I'm gonna do that. I'm gonna do this. I'm gonna do that. I'm gonna do this. I'm gonna do that. I'm gonna do this. I'm gonna do that. I'm gonna do that. And then when I see a sheet in the UI somewhere, I'm like, can they just do an LLM interface to this thing? Because it's just annoying to use typical, after you're used to chat sessions or vibe coding as a way to get things done,

Bilawal Sidhu: everything else is just fucking annoying.

Jon Radoff: I can't stand like searching through and what do I end up doing anyway? I go to an LLM and ask it how to do it. Because it's the fastest way to figure out how to use the software anyway.

Bilawal Sidhu: I couldn't agree more. It's like it is like, like, hey, LLM, tell me the, you know, like the five steps I need to do in the F-35 cockpit to do what I'm trying to do. Why not just have it do that for you and abstract out all that complexity? And a lot of people might criticize this saying, well, you're not learning like, you know, the skills or whatever. And it's like, I would argue that like, I don't think those are actually skills. Those are just the limitations of the technology that we had. The real skill would be like how to do scene layout, how to do really good lighting, how to figure out like immaculate, immaculate post capture, like our post render compositing. And those are the higher order skills that you can still build out without needing to learn how to, you know, do all this stuff manually. You know,

Bilawal Sidhu: that of course you and I had to grow up doing. It's no different than when compilers first came

Jon Radoff: around, frankly. And there were a bunch of people who are like, all these programmers are never going to have to understand like what a register on the CPU is like. They just, they'd, they can just code it and it works. So, it seems like LLMs are going to be the connective tissue that allow almost all software to interconnect and cooperate with each other, including, like so a couple of years ago, myself and our CTO of Beamable, we went to this A16Z hackathon, which we ended up winning. We created this Dungeon Master and we had this concept that I think you would almost consider it to be MCP now, but it was very, very old. I'm not sure. very preliminary. We had this programming model, which was, well, when you're building a game and you don't know all the functionality that you're going to need yet, rather than code any part of that, why not just use an LLM to implement the equivalent of functions for you, and then just use something like XML to exchange data, shuttle data back and forth between the main program, which I think in principle is kind of what happens with MCP, even though it's for controlling software as opposed to adding functional modules. But LLMs, I mean, what are your thoughts on that? Like LLMs seem like the interface that we all wish we had for computers all along. It's what

Bilawal Sidhu: was in Star Trek, right? It seems like it, and especially this push on like, if you look at models like O3 and Gemini 2.5 Pro, where you've got like, you know, there's sort of this push towards agentic AI, which has sort of become the new buzzword. And, you know, whether that's just like very basic chain of thought thinking or like, you know, static AI, which is sort of the new buzzword, and, you know, whether that's just like very basic stacking things on top of that, this idea of sort of LLMs, or even, you know, LLMs is such a limited term, given they're multimodal now. Multimodal large models, like, I don't know what the right, even like, let's just say LLMs for now, are like sort of the conductors of the orchestra, or sort of like the, you know, kind of they're the conductor of the orchestra, and you're kind of telling the conductor exactly what to do. And I think that's super, super cool and exciting, right? And I think you're spot on. It's also connecting not just software, but also making sort of content fungible in a sense that, like, I really started thinking about a lot of things as like content in, content out, and not thinking of it as like text to something. Which means like, yeah, you know, you have this conversation. Yeah, one simplistic way is like, let's just transcribe it, right? But some of the ways I've started using these, you know, LLMs is like, man, I'll record like four or five different takes for my like, YouTube channel. And I'll be like, you know, you are the creator of this thing. You are the creative director, tell me which was the most energetic and, you know, take and which had the best delivery and which one's like, you know, the best for, you know, YouTube retention, and pull from the principles of like Daryl Eves and Patti Galloway, or some, you know, insert YouTube expert that's prolifically written about this topic. And it's kind of wild that that works just as well. And it's the same thing around like taming the chaos of just knowledge work too. It's like, you know, you if you're collaborating with folks, like, you know, I'm like doing a sponsored podcast. I'm like, I'm like, I'm like, I'm like, I'm like, I'm like, I'm like, I'm like, for working on sponsored content for Adobe right now. And there's like, you know, notes that you get from them notes that you get back from your team, your own notes, and some of them are like frame IO on annotations, and some of them are images that are annotated, and some of them are video calls. And I just love dumping all of that into AI studio, and having just like make sense

Bilawal Sidhu: for me and kind of reference things back. And so it's like this general purpose glue connector or

Bilawal Sidhu: the bailing wire duct tape with intelligence baked into it. And yeah, so it's like the conductor is right, but like a really good way to do it. And I think that's a really good way to do it. And I think that's a really good way to do it. And I think that's a really good way to do it. smart conductor that knows how to play a bunch of instruments too.

Jon Radoff: The killer app for this live stream would be the day that we could feed the whole recording of the video into something and say, find the best 30 seconds of that, extract it out or do some kind of super cut that's amazing so that we can just post the one minute version of the live stream. We've tried to do that. I mean, there's a whole bunch of things that purport to do this. Yeah, we've tried it. Yeah, it hasn't, it hasn't really figured out what's most interesting in the conversation. Sometimes it figures out what the most energetic exchanges are, but I'll let you think about that for a moment. Cause I also want to just take a moment here for a little station identification. So we've got a few hundred people who are watching live right now. Thank you so much for tuning in and watching this conversation that I'm having with Belaval. There's a reason we do this conversation live now instead of a recorded YouTube video, which is it allows you as the audience to actually be part of the conversation. There's two ways you can do that. One is just post a comment. We're on LinkedIn, Facebook, YouTube, Twitch, X, all of those. We're monitoring all of them and Oscar will pop your questions right onto the screen. So if you have any questions about generative AI, creative workflows, becoming a creator preneur. I think I said, pronounced that correctly. You should ask that because you're here in the room with, with one of the true world experts on that. And it's a, it's a rare treat to be able to talk to someone like Belaval. Hey, if you want to jump in live, you got a camera. We are risk takers on this stream. We'll invite you right in Oscar. We'll give you a stream yard link. You can, you can join us. It hasn't gone sideways yet, so we're going to keep doing it. But we do keep the link secret. So you have to DM us. At least. So let us know if you want to participate, but please do take part in this conversation. All right. So back to sort of the editing process.

Bilawal Sidhu: I mean, my, my thinking is, is a video.

Jon Radoff: There's probably a lot of other cases where we can just say to generative AI or an LM. Give me the good version or the good parts. Have you seen anything like this? How close are we to that?

Bilawal Sidhu: I mean, it all comes down to how good, like the multimodal understanding and kind of. Yeah. Yeah. Yeah. Yeah. Kind of scene understanding too. Um, I think the best one I've used thus far is opus. Um, so it's like a, I think opus.pro and it it's after they moved to Gemini for multimodal reasoning, because like a lot of the initial instantiation of these sort of find the best, most compelling segment was basically like take video and audio file, like transcribe the audio and then use some LLM to sort of rank and grade, you know, the clip worthiness of the various segments. And as a consequence, you know, there are two issues there. One is like your timestamps on the, the, the transcript don't always perfectly correspond to the moments in the clip itself. So it's not very good at taking, you know, like the best clips editors.

Bilawal Sidhu: If you work with them in real life, they'll like pull out a lot of dead air in the middle

Bilawal Sidhu: and know just how to take this little bit over here and that, and sort of how it flows together. Models are getting better at that with sort of this, like, I believe they moved to two Dotto flash. I don't know what they're using now. Um, but there's an interesting blog post on it that they jointly did where now the reasoning about the contents of the imagery too, right? Like, and so I think the more signals that these models can tap into, it'll, it will get better. Um, but it's clearly not there yet. Right. And it's funny, like my buddy, Riley Brown, who's obviously like one of the, I would say preeminent vibe coders, um, on X is, you know, has been thinking about it and like just his like background is interesting too, right. To go from like marketing and doing like B2B marketing to Tik TOK to now being a vibe code.

Bilawal Sidhu: Um, this has been really interesting and engaging to me is he said, if you address

Bilawal Sidhu: your page, you can move or do something with Folio mode in Y Prix mode. Yes, we paste and input fic down the edges of a theme. Um, and he writes, and I couple notes, right. Because sometimes in logic binary, which I don't believe, but maybe in logic, inferences for, for video, it's easy to domon, um,都닥려 Electśnieко. He has a Navy word, which is aело. So he can. You can make a dumbец, play a moose, let's say. And just like, in the video game is totally absorbed into letters like and repeat get allHAy. It just makes them 먼저. No. Me. be like, hey, find all the most, you know, kind of juicy existential clips from this conversation I had with John. And I get this like stack rank list of three or four of them. But the closest I've seen is Opus and I suspect it'll only keep getting better.

Jon Radoff: Let's zoom the camera lens out a little bit and maybe this will put us back on the path to that original question about the future trajectory of the economy with all of this stuff. But you had a really unique, enviable opportunity to talk to some of the greatest minds and leaders in this entire industry. Like you spoke with Eric Schmidt recently at TED. So we won't do a spoiler on like whatever that conversation was, cause we're all looking forward to watching it ourselves. But you've talked to him and other people like that. Are you seeing any common themes in what you're hearing from people about what they expect from the future? Or what are the areas of controversy if not?

Bilawal Sidhu: I think it's everyone has their perception of the possible futures, right? And it's interesting and it keeps flip-flopping. And you could say that like, so I hosted the TED AI show last year and did 25 interviews around that. And it was similar thing. Like some of the conversations are very focused on like experts in a field. And like what the future of like, I don't know, like mental health might look like or Hollywood might look like, or defense tech might look like. And it's interesting. All of these professionals that are at the forefront

Bilawal Sidhu: of all of these vertical industries have their sort of,

Bilawal Sidhu: here's what happens if things go well, and here's what happens when things go badly. And some reality is typically some combination of the two playing out simultaneously. And so like, I think what's exciting is, we're clearly on this trend towards, again, more and more of sort of things, tasks that you can do on a computer are gonna become, like, you know, you're gonna be able to do a lot more of these things that you can do on a computer, and you're gonna be able to do a lot more

Bilawal Sidhu: and more commodity, right?

Bilawal Sidhu: And I don't see that stopping. Like, you know, when there were two moments for me, you know, coming, like having learned visual effects

Bilawal Sidhu: as a kid, like, man, I spent years trying to learn

Bilawal Sidhu: some of these skillsets. And even when image generation was popping off, and that was probably right around the time we had a convo. This was like right after I think Control Net had come out and you'd come across my like reskinning my parents' living room video. And man, that was like, you know, and then after that I made like a video to video, like a hacky video to video pipeline, with, you know, like, you know, like, you know, like, you know,

Bilawal Sidhu: like a hacky video to video pipeline, with Control Net

Bilawal Sidhu: and Evisynth at the time. And, you know, then Kyber productized something like that and a bunch of other folks did, and that felt state of the art. And I felt good at that moment in early 2023, where I was like, oh man, I got some time before video and 3Ds figured out, like I'm chilling. And then, you know, when Sora and then, you know, specifically VO2 by Google dropped, it really rocked me pretty significantly because my timeline shifted. And so the other challenge with this is like all the bright minds are constantly recalibrating what the future looks like. And so, you know, you'll see this certainly in the Eric Schmidt convo when it comes out mid next month. And you'll see this in the Ted AI show conversations too, is like, as the season starts to where the season ends, people's predictions kind of change because you update your priors so much constantly. So the way I personally stay sane is I'm like, I got a decent idea where things are gonna be like maybe six to 12 months out from now, beyond like, you know,

Bilawal Sidhu: I'm like, I got a decent idea where things are gonna be like,

Bilawal Sidhu: here's what we're gonna do next. And then beyond that, the compounding, like stacking of a bunch of these crazy unlocks happening is unclear to me, right? And sometimes it's not even technological. It is something like just the industry converging on a standard like MCP that unlocks a bunch of interop between a bunch of different tools, like that it's so hard to predict. So I don't have a pithy answer. What I do know is tasks that you can do on a computer will increasingly be automated. So I've been thinking about what can you do in that world? And I've certainly got a pithy challenge take on, you know, four things that I encourage like young professionals and young students and even like, you know, mid or late career professionals to do. And we can get into that

Bilawal Sidhu: if that's of interest. Let's hear the four. So, uh, of course I've got a slide for it because

Bilawal Sidhu: this was like the, the, the fricking thing that came up the most, um, at, at Harvard. This is, so this was at the graduate school of design. And these are the four things that I bring up, which is like, number one is like a lot of people will just put their head in the sand. And, you know, uh, you know, some of my creator friends kind of, and even entrepreneurs that, that I know write this off as like, Oh, that's the Luddite crew. And I have deep empathy for

Bilawal Sidhu: folks that are feeling like they're like, I don't know, like the thing that you grew up learning is

Bilawal Sidhu: suddenly super commodity. It feels weird. I think like image artists got it first writers got it second. And now, you know, VFX and 3d folks are getting it. Maybe they're starting to get it. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't know. I don't time left for 3d and certainly a bunch of time left for XR and spatial computing. So maybe I feel good about it, but you got to play with the tools is point number one. And if you don't, you're like, you are absolutely missing out. This is where I try to create some anxiety for people, a positive anxiety to get you to do this thing. But the way you should play it is I, I've kind of reduced it to when you're playing with these new tools, you should think of it like

Bilawal Sidhu: you're sort of in sandbox mode or architect mode and sandbox mode is like, you know, you're

Bilawal Sidhu: building castles in the middle of the city. You're building castles in the middle of the city. You're building castles in the middle of the city. You're building castles in the middle of the city. You're the sand and the waves are going to come in and they're going to, you know, like, you know,

Bilawal Sidhu: kind of wash everything away and you're okay with that. But you also allow serendipity,

Bilawal Sidhu: happy accidents to happen. And you sort of like learn cool new things that you should then put in your like library of like things to try and apply that towards a concrete end and more of an architect mode. We're trying to build a piece of content, you know, uh, you know, a SAS product, whatever it is, a product experience or content. That's something that you should focus on and apply that towards a concrete end and more of an architect mode. And I think if you just do those two things, your level of anxiety around AI will reduce too, because you'll learn the one skill that seems to be the most important one, which is the ability to adapt as these tools keep coming out. And you want to build this muscle because the rate of change is only going to get crazier, right? Like, you know, whether people can debate the scaling laws and no pre-training is dead. And now it's about like post-training or no, it's about like, you know, test time, compete, whatever. It doesn't matter. This stuff,

Bilawal Sidhu: we'll keep get, there'll be a lot happening as various industries adopt the stuff. And so

Bilawal Sidhu: building that muscle of upskilling and making it fun for yourself, I think is super key. The third piece that I'd say to a lot of folks, especially folks that are hesitant to be creators, funny is like one student pops into my head who was like a, uh, UPenn DMD digital media design student who graduated and is like going and doing a product job at JP Morgan. And it's funny. It was like, she was super passionate about like XR and spatial and all this stuff. And she was like, yeah, you know, I want to do exactly what you're doing, but first I'm going to go do the product thing. It was like, oh, what are you going to do? Oh, I'm going to go work at JP Morgan. And I was like, you know, no hates at JP Morgan, but I was like, you just did digital media design and now you're going to go, you know, like work at a bank, like, you know, maybe they're doing cool metaverse things or whatever, not to fully fully write them off, but it didn't feel like the highest, it didn't feel like the highest contribution based on the skillset that you spent the last like four to eight years of your career doing this master's degree, you know, like in bachelor's degree building. But the one takeaway I gave that student and many others is to start building your island of influence, right? Like, I think this is the time, you know, it's very cringe now. I think a lot of people that go, I'm not going to build in public that's cringe. There's like move in silence and success speaks for itself and people will bring out all these tropes. And I think that's like completely outdated advice. Like unless you're working on some like deep defense tech thing, where you explicitly don't want to, you know, draw attention to it, or you've truly got some novel IP.

Bilawal Sidhu: I think

Bilawal Sidhu: most people, they should be sharing their interests, interestingly, and building out some island of influence. And that doesn't have to be like millions of followers, it can be like 10,000 folks on LinkedIn could be your goal. And if you're in that niche community of like, I know so many GIS geospatial information systems, professionals who've got that 10,000 15,000 20,000,

Bilawal Sidhu: you know, follower group, and they've got the heavy hitters, you know, at the big companies

Bilawal Sidhu: at the startups, you know, at on the customer and, you know,

Bilawal Sidhu: on the platform and, and it's unlocked doors for them that you cannot imagine.

Bilawal Sidhu: And the second reason I would say you need to do this now is, as the barriers to creation keep dropping, as we covered in the earlier part of the stream, I think this content is just going to be flooded, right? Like a lot of it will be, you know, automated, you know, faceless bot content, and then eventually will be embodied, but content that will be indistinguishable from other stuff.

Bilawal Sidhu: So if you got this window now, to basically become a voice,

Bilawal Sidhu: that people look to and trust, I think you should be putting your interests out there and doing that in an interesting fashion, not trying to reverse engineer the algorithm. That's the third thing I'll say is like, a lot of people get into the mindset. Well, like, let's go see what's like, what's caught on Google Trends, or what's on like the first page of YouTube. It's like, no, no, no, this will not be fun for you, unless you're sharing the things that you care about. And it's fun for you to make that video. And so maybe it's not even a video. The actual last thing I'll say is, find a medium that you do your best work in. And that could be a substack, it could be tweets, like some people are so good at vague posting on x, and that's their thing. But then point a camera

Bilawal Sidhu: at them, and they are like, they lock up, or you meet them in real life, and they can't even make

Bilawal Sidhu: eye contact. It's kind of wild, especially in the SF AI community, but oh my god, they thrive on this short form canvas, right? And so find whatever medium your most, you know, where you do your best thinking. And use that as sort of your hub and just start there. Like, it doesn't have to be complicated. It could be like one LinkedIn post a day, three LinkedIn posts a week, whatever it is for you, just start somewhere.

Jon Radoff: To just spend a moment more on this, like you, you gave a lot of sort of the marketing reasons why it could be worth and finding your niche and the particular platform that works for you, there's a lot of advantages to building that community around yourself. But it's not only marketing, it's also feedback and information. Like when you put your information out there, and I think that's also what's scary for a lot of people is that once you put yourself out there, you also make yourself vulnerable to critique, being wrong about stuff. I'm wrong about stuff all the time, by the way. And the only way I get better is I hear that feedback from people or I have a really productive conversation with people in which we're both partly correct and partly wrong and sometimes you get to amazing meeting of the minds where you can actually improve things. And I think this can exist almost anywhere but politics. Politics, it doesn't work. But everywhere else, people who are sincerely dedicated to craft and improving things and learning how to be better at things, just putting your stuff out there is part of how you get great at something. And in games,

Bilawal Sidhu: which is the media that I'm most familiar with, the only way you can build a great game is, A,

Jon Radoff: a lot of velocity in your process, because if you're too slow, you're going to run out of runway. The market will change too much by the time you're done, things like that. So velocity is important. But what you do inside that velocity is equally important. It's having a lot of shots on goal where you try things and you find out what people like or don't. And you incorporate that feedback rapidly into your development process. So that's one of the biggest advantages of building in public. And of course, sure, if you do that well and people love what you're doing, you can end up having a marketing advantage as well. But I just want everyone to really think about the learning advantages and the product improvement advantages of putting things out there and hearing real feedback. And by the way, not many people want to just rip off an idea. Really, you have the holistic understanding of an idea in your head. You're going to carry it out. You're the one who is going to be sincere about it and see it to the end. I guess there's copycats out there, but it's not the biggest thing to be worried about. The biggest thing to worry about is someone going to just do something faster and more meaningful than you.

Bilawal Sidhu: I think that's on point. And even if you have copycats, right, John? It's a good thing, right? It's like, holy crap, your thing is worth emulating, right? It has some memetic potential to it that it is something that people are replicating. And I would view that like imitation is the highest form of flattery. I've always thought that. And yeah, I always take it positively. And I honestly don't even worry about the credit thing too much, right? Because yeah, you're totally right. In the book writing space, this has come up a bunch, right? There's a lot of these folks that are sort of doing their book chapter by chapter on Substack now instead of sort of signing the trad two to three year deal. With a publisher and where, you know, to your point, your idea waits so long to incubate before it hits reality. Whereas if you could kind of iteratively refine your idea and put it out into the public and then you package it up, I think there's a ton of potential there. So I think that's very, very cool. Absolutely. All right. So I stopped you at part three or four just around. Yeah. I mean, the last part is very related to what you're saying too, is like, you know, a lot of folks ask me, he's like, well, I love all my AI influencer friends, but I think one of the things is like, you know, they say, don't hate the player, hate the game. And the game on social media incentivizes the Excel is cooked. Hollywood is cooked. Everyone's cooked every single day, all the time. And all these old tools are dead. And, you know, you know, MCP is a great example of like, you know, LLM sort of breathing new life. I wish I actually shouldn't say new life, but like a way to like, you know, essentially expand the set of like tool use capabilities. These systems have a being, like the industry standard tools. But the last piece of advice I give people is like, even though you're preparing for the future, I think there is tremendous alpha and learning how to use Photoshop today. I think there's tremendous alpha and learning after effects in Maya blender in unreal engine, unity, whatever the arc GIS for crying out, like whatever the incumbent tool is in your industry. I think there is so much value in a figuring out how the status quo, the right way to do things in quote quotes are. So you can figure out what to discard and throw away. And B at some point, you may just be telling an AI agent to do those things for you. And it's helpful to know, like, you know, like, I'm working with an editor right now is doing a bunch of after effects work for me for this Adobe thing. It is helpful to know what is and isn't possible. So you can articulate your requests and turn your vision into reality in a more concrete way that it is in line with what the capabilities of those tools are. So those are my four pieces. And yeah, I mean, like, the last thing I maybe would say there is like this thing, like, it feels like again, because of this sort of incentive structure on social media, particularly X and LinkedIn, even YouTube, to an extent, even threads in Instagram, like, all social media, I guess has this where you stop the scroll by having some provocative claim. I would tell people that it is still early. I think these Lego bricks that are already public and out there, they still have not been put together in the most obvious combinations,

Bilawal Sidhu: let alone.

Bilawal Sidhu: The insights that you are going to unlock when you apply them to your specific creation or business problems in your specific verticals, through the lens, the lens through which you see the world. And so like, I would just say, like, gosh, start playing with this stuff. Because so many people feel like, oh, it's it's the ship has sailed. And I've seen so many creators and entrepreneurs pop

Bilawal Sidhu: off very recently, because they've managed to put things together in a way that the rest of the

Bilawal Sidhu: world just wasn't because they have this unique worldview.

Bilawal Sidhu: Insight or idea. And so it's like, you know, it's like, just because we're all working with the same

Bilawal Sidhu: concrete doesn't mean like all the buildings in the world have been built already.

Jon Radoff: Well, one of the interesting things about a lot of these generative technologies is they do allow you to mash up things and interesting combinations that haven't, in fact been done before. At the same time, I think it's really useful when they do mash things up in a way that has been done a bunch of times before, because you should know that.

Bilawal Sidhu: Yeah, it's like the boilerplate stuff, right?

Jon Radoff: Yeah, totally. 100% agree. I remember there was this one a couple of years back where people were someone who was not a game design. I forget now the name of the game, but he, he created a game through what would now be called vibe coding a couple of years. But he ended up rediscovering a game that had already been known. And then of course, a bunch of game design. He's like, hey, look at this novel game design and a bunch of people piled on.

Bilawal Sidhu: Well, actually, this has been done like blah, blah, blah. And I mean, that's the that is it's worth double clicking on the point that you brought up of like, you need to grow some your your skin needs to get thicker. If you are going to be on social media, especially if you're going to put your real identity out there. I think a lot of folks will do the anon thing for this reason, because they want that layer of separation, which works really well, to be honest, and what better time to go be like, you could do an anon podcast now, like Dworkesh did that podcast with one of these like prolific, like, sub stack writers. Like blog writers in the AI space, Gwerne, who didn't disclose his identity, and they did like Unreal MetaHuman clone, which I thought was really dope. But you could do a whole pod like that, right? And there's a bunch of folks in the virtual production field that are have started like kids channels like this, where they're like puppeteering, you know, avatars and doing amazing content that way. But you do have to build a little bit of thick skin. And that is to get critique online or offline. It's just that online people, the people will, you know, they're going to be like, oh, I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this. I'm going to do this sometimes be a little less nice in the text based comment medium than they would be face to face to you. Let's just put it that way, but to put it very lightly. But that's a part and parcel of it to your earlier point, the feedback that you get. And, you know, sometimes you'll get really good ideas like, oh, this is cool. Have you tried this other thing? And so like, especially I feel like in the like sort of the coding communities and the generative AI communities or the creative tech community, if I had to put a bucket on it, there's so much like, oh, that's cool. Plus one, why don't you do this thing? That like the community sort of building workflows on top of each other. And I think there's something very exciting at a very fun conversation with the Civitai founders on the TED AI show that people might enjoy where they saw that community of like folks that were not from an ML background starting to build models. And then other folks figuring out various ways to put that together. And that compounding effect that happened where you're, you know, like ideas met reality and kind of the iteration that happens from there on out.

Jon Radoff: So I'd love to talk about right now the tools you like a bit earlier. You also mentioned you had a fairly clear idea for yourself of what you thought the next six to 12 months look like. And I'd love to hear that. But let's start from like the right now. So you mentioned a couple of tools around along the way that you like that. Like what what are you finding you're using the most like time wise?

Bilawal Sidhu: Yeah, I mean, like, let me pull up this LinkedIn post and just DM it to you because you can include it. Because I answered exactly this question. And this one person was like, very kind enough to summarize it all in a LinkedIn post with links. And again, this is also the other cool thing will happen is like people create derived content from stuff you deliver. And then that content can go basically like, you know, take on a life of its own, and it links back to you and all that other stuff too. And so if I may just find that over here, but as I'm pulling this up, I'll say it's like, like you have to do it. I use all the LLMs. And I've sort of kind of started viewing them as like, you know, kind of like members of the team, if you will. And right now, it's definitely a lot of chat GPT. That's the mobile app I use probably the most. And then here we go found blank. And let me copy paste it for you here.

Jon Radoff: And I'll reveal my answer in a moment to

Bilawal Sidhu: here, if you could just cast this out to cast this up, Oscar, and I'll paste the link here. Yeah, these are sort of the tools that I end up spending the most time with, like, it's a mix of like commercial tools, and, you know, models and like hardware, even, I'm not going to go through all of them. But what I'll tell you is like, if you're not using LLMs, you have to. And so for me, it's GPT-4.0, for sure. But it's also heavily, heavily AI studio. And I've very recently started using Gemini, the pro subscription, as well. I'm predominantly for deep research. But 2.4, 5 pro is amazing. So between 2.5 Pro and 03. Those are probably the two models and spending the most amount of time in. And it's kind of funny, because that was like Claude 3.5, for me for the longest time, but I'm waiting to see what response and Thropic has in the space. Because right now, I don't know. And I kind of feel like that about perplexity, too, is like I was using perplexity pretty heavily, you know, two or three months ago. And then, you know, first with just search getting better on chat GPT. And same thing with Gemini. And then like, oh, three, especially, you know, like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, two, one, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, three, and then like, oh, I would say the ultimate mid journey stand. When his posts become disproportionately about some other model, that's when you know we need to move off of mid journey. And I think we've reached that tipping point where he's writing a lot more about other image and video generators.

Bilawal Sidhu: And gosh, like GPT-4.0,

Bilawal Sidhu: I talked about this already in the Gemini Imagen, absolutely amazing, use that heavily. I use Flux a lot. I found myself gravitating more towards things like krea.ai, K-R-E-A.ai. And also like to a lesser extent, Freepik. But you know, I know some of the folks behind frigging, gosh, what was that super resolution tool? I'm blanking on his name right now. But the super res tool that's in Freepik, you know, I've used that heavily. And so they've been encouraging me to use it, but I like krea.ai for the UI. You can access all the other models there too. And the video model I use the most is still VO2. It's just, if I'm doing text to video, I am using VO2 by Google inside of video effects. It's available for GA in most countries now, or a lot of countries, I should say. And when it comes to image to video, it keeps changing a lot, man. I think Kling is still pretty fucking good. Sorry, I don't know if we can curse on the channel, but it's pretty fricking good. It's pretty fricking good. And especially the recent update that they came out with. So it's always, it feels like this like funny race where like the various like competitors, one juts out ahead and is out there momentarily. So the other piece of advice that I give people is like, have a flexible workflow, which is why I think like thinking like a creative director or like thinking like, you know, the conductor of the orchestra is like, yeah, you know, the instruments are gonna keep changing, but you know how to put it all together.

Bilawal Sidhu: And so that's on the video side.

Bilawal Sidhu: I would say for all the Gaussian splatting stuff we talked about when we spoke like two years ago, I would say for all the Gaussian splatting stuff we talked about when we spoke like two years ago,

Bilawal Sidhu: I would say for all the Gaussian splatting stuff we talked about when we spoke like two years ago,

Bilawal Sidhu: just as you know, the nerf to, you know, explicit implicit radiance fields to explicit radiance fields sort of transition was happening. I still love, love, love, love, love using Polycam. I think that's a really cool app. And I use that for their like photo realistic capture capabilities, but also creating these like CAD like abstracted models. And what I mean by that is this, if we wanna just pull back the slide here very quickly. So I add some visuals to the mix here, stuff like this. Stuff like this can be very easily done inside of, there we go. Like stuff like this, if you pop that back on. Oh, sorry. I need to share the other tab. There we go. Like stuff like this, you can do it inside of Polycam very easily. I love using Scannerverse. That's become my go-to splatting app predominantly because it trains all the splats locally on your phone. And so just think about like how wild it is, like how far things are going to go. I understand that this like皆さん has probably dishes spits of this thing,

Bilawal Sidhu: even on your phone already

Bilawal Sidhu: because we are test content maker tigers. I'd like to hear what you guys feel. So I think it was only just about the number of slots we had, we had this in game that we could just like warehouse up maybeka and keep going minimal progress. I don't really want it.

Jon Radoff: So that's my available mail. I used two or more of them. opposed to Gaussian splats.

Bilawal Sidhu: Yeah, I mean like honestly, I don't like, I like Meshi a lot speaking of that. Like Meshi is really cool. Trippo is pretty cool too. Like no complaints there. I haven't seen anything that's sort of like text or image to scene in a largely automated fashion. You see the earliest instantiations of that with some of the MCP demos, but that's still like those demos are pretty cherry picked. If you go use, you know, set up one of these MCP servers, you'll find that the human still needs to do a lot of the massaging to get a good output. And so I haven't seen that yet. And it's interesting, maybe that's an interesting point to talk about, which is sort of like you're seeing, where is it? Yeah, here, I'll bring my slides back. I'm sorry, ask her, I just stopped presenting there, but let me, let me re-present. Is, there we go. It's like, I think we're headed to this world of like dynamic disposable content. And this whole Jensen quote, which I had a chance to ask Jensen at the last GTC, about every pixel is generated, not rendered. That would make the 3D native thing moot in a sense, right? And if you'd asked me that like pre VO2 and pre like Sora, I was in the camp that I think you're in, but feel free to update my priors on where you're at now, since we last spoke is like, I thought you needed an explicit 3D scene graph representation, like that you need something like, where the hell is it?

Bilawal Sidhu: And you need something like this.

Bilawal Sidhu: Like you need to define what is in the scene and like what the, you know, like you need explicit 3D control. Now, after seeing how far video diffusion takes you, I think Jensen's actually making a conservative estimate of this being like five to 10 years out. And the other perhaps indicator. Sandbag him again, just like he sandbags his profits. Okay. I mean, yeah, like, and this is the reason to like take him, take him not seriously. Like, I mean, of course, Jensen, you think that everything's gonna be generated, not rendered because you're gonna sell all the GPUs and you know, the infrastructure to go do that. But I think when you do look at stuff like, I think I have, what is it? Like Genie 2 here by Google DeepMind.

Bilawal Sidhu: Yeah.

Bilawal Sidhu: The example of like, given the massive context windows you have, you can go from like one frame to like a 60 second playable world that's largely coherent. And that's where we are now. If you just compound the progress that's happened in image and video gen, I think like interactive, playable, interactive, playable worlds that are direct to pixels. And you know, you're just like passing on the input and the model is giving it to you. Microsoft has done similar research. All these companies are doing similar research. Man, I don't know if this is just gonna bypass a bunch of 3D and the real, like, I would say, I wanna say the nail in the coffin, but yet another straw that might break the camel's back is, you know, like of course, Luma AI, right?

Bilawal Sidhu: And so Luma is the, the, the, the, the, the, the, the, the,

Bilawal Sidhu: the first one that sort of transitioned pretty heavily from, you know, there were a 3D first AI startup and they moved from trying to do like nerfs

Bilawal Sidhu: to Gaussian splats, to let's take on Matterport,

Bilawal Sidhu: to let's do text to 3D, to fuck it. We're just gonna invest in Ray and we're gonna go all in on video diffusion. And if you see Amit's recent posts and podcasts, he talks about sort of, this is the world engine. This is the path to AGI and having this world engine. And you'll see folks like Jan LeCun, who's obviously very smart and way more than me, you know, poke holes in this. And of course the homie, Gary Marcus will throw pokes, holds in this, like how smart are these models really, et cetera, that's all fine and good. But another straw that got added to this camel's back that seems to be very fatigued at this point is Odyssey ML. And so Odyssey ML, I love what these cats are doing. Like they're obviously like, I can pull up a funny tweet here cause the CEO responded to it. Is so Odyssey ML is a bunch of X crews and they're all like, they're like super smart and they're like, they're like, they're like, they're like, they're like, they're like, they're like, they're like, they're like, you know, autonomous driving folks who are very used to this world of high-end sensors and like reality capture and sort of like, you know, creating this explicit, you know, representation of the world and how to semantically decompose it and do all of these other things. And yeah, I don't know. It feels like, here's the tab I'll pull up is like, if you could just flash this Oscar,

Bilawal Sidhu: it would be super cool.

Bilawal Sidhu: Yeah, I've heard this coming into Egypt,

Bilawal Sidhu: I heard he's a super God on out button general.

Bilawal Sidhu: They've got this amazingDB Дав correct. Like good strategy. He's crazy. These guys both are Iya, which makes it funny, which makes me wonder, once again, this list of three people that are not really character Arti models that are enters is basically 3D ready. It means it's 3D steady! It means you can just wherever for whatever reason whatever projections I could use well, but the video depth and stuff here while, the incumbent market, maybe you go down that path. But hey, don't get me like the unsugarcoated version of what Oliver is saying is there is no market for 3D assets today because it's such a small TAM, but the market for images and videos only growing. And so if you can bypass a lot of that legacy pipeline that you need to get there, I mean, I think that's like the path to invest in. Now I will call out other startups like, Bharat and his team, this amazing ex-Unity founder, co-founders from ex-Unity that are trying to do that sort of hybrid approach that we talked about two years ago. That's super exciting and cool for sure. But man, gosh, it just feels like if you take a company like Microsoft that owns Xbox and a bunch of AAA studios and Google has YouTube and a couple of these other companies will create this data or acquired in interesting ways. It feels like you're just, 3D becomes a means to an end. So if we think content to content, if you can go from the idea for the game or the detailed game design document or whatever, with very detailed specifications and storyboards or even final artwork to like a playable experience, does it have to be 3D? Or can we just bypass it altogether? So those are some of the things I'm going through. We had this interesting conversation

Jon Radoff: just a couple of weeks back with Tejas Kulakarni, who's the CEO of this cool company.

Bilawal Sidhu: Oh yeah, love what they're doing, yeah.

Jon Radoff: Yeah, 3D foundation models. So he had an interesting take, which is he, if I understand what he said correctly, I'll try to reproduce it. He said that he does think it'll eventually be kind of direct to video presentation from the model. But if you look at what those models are really doing, 3D is such a great optimization that they're really gonna optimize around something that's not too dissimilar from 3D, kind of, maybe not that much throwing everything in hand. So that's kind of ourBaubler amongst that. So I had you looking at some of the models and trying to figure out maybe which model you, not exactly in this case, but maybe just in the suspension of this thing of 세상 afect 2D versus really, you know, if any of the models were good you would know and I,

Bilawal Sidhu: yeah, so we hope that it'll take a little while and that people actually find out

Bilawal Sidhu: when it gets even better. So I hope we can get more like Arg Alliance combined that this kind of like aY Argentine model that you could do, it get to that log intake, something. Just help me. He says that's our reason to work, right? Yeah, I think it's a personal choice too. So I'd like an ITSAN model, for game dev or visual effects or, you know, even things like, you know, product, you know, rendering and photography use cases and do some sort of a hybrid workflow. Yeah, but I don't know if that's, to the limit, I don't know what the right answer is, right? This is where it's way beyond certainly that fog that's in my mind past 12 months.

Jon Radoff: So Marty in the audience here on LinkedIn asked a couple of interesting questions. He first asked what's up with organic and synthetic people, which is another kind of AI model. And then he's asking this question here about robots feeding into generative video models. I have no idea on this. Really interesting question, Marty.

Bilawal Sidhu: Yeah, I've got tons of thoughts. I mean, so, you know, I had the One X CEO in my session at TED. It was amazing. Like Neo is trying to get robots into the home to collect the type of data that you can only collect in a home, like versus a sterilized sort of factory setting and hoping that the model learns how to operate in all these edge cases and scenarios. And so capturing training data in these chaotic scenarios is very interesting. Meta has an interesting program there

Bilawal Sidhu: where they've got operators and people they hire

Bilawal Sidhu: to collect this type of egocentric data from the perspective of a humanoid, that is us, wearing these sort of glasses and you have external LIDAR, other external sensor systems. And you get this like everything from like an Indian mom going through like the fricking spice rack, you know, like, and all of these kinds of edge case scenarios. And data sets is like, I think very, very interesting. And so you have a huge amount of that data on platforms like YouTube, but not a lot of it is, you know, like, yes, you've got crazy walking tours through Tokyo and San Francisco, but do you have all the scenarios that you might see? So I think, you know, just like a Tesla car is capturing all these edge case scenarios in the wild compared to Waymo that has to have a smaller number of cars that are doing high def mapping. I think there's gonna be a place for both of these approaches of like really high quality, sensor systems capturing the best quality data and the noisy sensors, you know, capturing all this long tail data, but doing it at scale. And so some combination of that's gonna be pretty magical, not just for robots, but for sort of the, you know, AI assistant that's gonna be the Jarvis in our glasses or in our homes, right?

Jon Radoff: Balava, this has been an amazing conversation over the last 58 minutes. Somehow in the last couple of minutes, I got a cram into it. What is your vision of what's gonna happen?

Bilawal Sidhu: Oh Lord. I think, I don't know what my vision is. I'm figuring it out, but I think the lines between the physical and digital are certainly gonna continue to blur. And you know, the way I view what I'm doing is sort of mapping the frontier of this convergence that's happening and building simplified and detailed maps of sort of this uncharted territory. So I would say that like, you know, if people are excited about this convergence and there is the promise in peril of what's gonna happen there, right? Like it was interesting to see the reaction to Sharam Azadi's Google XR, a glasses demo. There's half the people that are like, oh my God, I can't wait for this. Give me, give this to me now. And the other half are like, what? I'm putting a ring camera on my face.

Bilawal Sidhu: Like what?

Bilawal Sidhu: Like this is like the death of privacy, et cetera, et cetera. And as always, the answer is somewhere weirdly in the middle. And you know, so if these kinds of maps of the frontier of, you know, technologies that are connecting bits and atoms are of interesting to you, I would just say like, check out my Twitter, my X, my LinkedIn, my social media platforms. They're all linked at Balava.com. I'm also on my Twitter, balava.ai. And that's exactly what I'm trying to do. As like I build, like I'm slam mapping the space, right? Like simultaneous localization and mapping. And as I'm updating my map, I'm sharing that with you so you can follow along for the ride and figure out what chart you want to navigate through in this future that is fast and perpetually approaching us.

Jon Radoff: Well, there's your action item. Go to balava.ai, just like it is on the screen right there. You'll get that list. We'll also post a link over to that. X post where a gentleman was kind enough to list out all the tools you use. Use that, you've got an immediate roadmap to learn a shit ton about this space. And then you'll be all ready to continue learning along with the rest of us and trying to keep up like the rest of us, which is really, really hard. But this has been a delight as usual, Balava, happy to have you back any time whatsoever. So thank you for taking an hour of your really super valuable time and joining us on the artificial, intelligent live stream.

Bilawal Sidhu: Always a pleasure.

Bilawal Sidhu: Thank you for having me.

Jon Radoff: All right, Balava, until next time. And thank you everybody who sat through this and were in the live audience. Really appreciate your time as well. And if you were watching this in a recording, come on live next time, ask a question like Marty did just a moment ago, and you can be part of the conversation as well. So until next time, take care of everybody. This has been a blast. Thanks everybody.