A (3:34)

Great. I know you know a little thing about natural language processing there. Great, Richard.

C (3:40)

All right. Hey everyone. I'm really excited to be here in Austin. I run u.com y o u.com it's essentially a productivity engine that gives you answers. Agents and AGI. What do I mean by this? You can think of it as a mixture of Google and ChatGPT, but for your enterprise productivity. So you have a question about sales, service marketing, you ask it to you dot com. It will do the analytics for you. It'll go on the web. You can think of it as almost like a consultant that will do research for you, merge it with 400 different sources or more, give you citations that are accurate. You can actually rely on it in your business to be smarter about what you want to do. And the AGI bit, it's not quite here yet, but we're sort of guiding corporations. Every LLM that's out there, Deepseek, it's all available within you.com and we basically enable your company and your employees to become managers of AI with role specific training and just do the whole thing so you can transform your business.

A (4:43)

Awesome.

E (4:44)

Hi everybody. I'm the founder of a company called Moon Valley. So basically we're working on foundational video models. So it's AI models that allow you to create scenes essentially so create video from scratch. You've probably seen a couple of models that are out there that do video. The differences on our side are essentially twofold. So the first piece is that we're focusing on production and enterprise grade video models. So what are the video models that you'll use to do films and cinema? And on the flip side, on the brand side is what are the models are you going to use to make, for instance, ads and commercials and things like that? The other piece that's kind of differentiated about us is that we're approaching building models a little bit differently from everybody else. All of the data that's gone into our models have been sourced directly from creators, so we don't do any kind of scraping or anything like that. And the goal is to essentially bring in models that on one hand are actually in the service of creators and content creators. And then on the flip side, you also avoid all the kind of copyright issues and indemnity issues that come with these models. So yeah, that's kind of us in a nutshell.

A (5:53)

Great. Well, okay, so in advance of this conversation today, I went to you.com and I asked the productivity engine, as you call it, I asked you.com what questions I should ask Richard Socher today. And let me give you the first one that popped up. What do you believe were the most significant advancements in AI over the past year and how do you see them shaping the future of technology? I think it's a good question, Richard.

C (6:24)

The only problem is I could talk about it for an hour, but there's a lot of stuff happening in AI. But I'll keep it short. I think in 2023 I predicted that open source models will be as good as GPT4. And a lot of folks were like, no, you're definitely wrong. OpenAI no one can catch up. And I think I was right. So that was a fun one last year. I think now seeing the transition kind of from POC land, everyone wants to try out something, but also companies buying like a thousand seat licenses on ChatGPT and then realizing only 5% of their people are actually using it every week and then, you know, they're sitting on 95% of unpaid for but unused licenses. So I think we're realizing now, yes, the technology can be disruptive, but AI might move very quickly, but organizations and people still need some handholding to actually use it and deploy it productively to see the advances really for their business. So I think that is kind of an exciting moment now in that thanks to Deepseek and others, companies like you.com that kind of provide the layer on top of all these models and actually help you use them, like build agents that automate a specific workflow in marketing or for journalists, that's really been really exciting. Hearing journalists say work that used to take me three days now takes me three hours. It's just like very exciting to see.

A (7:45)

They're phoning it in. Is that what's happening? No, I'm just kidding. They're using AI. I mean, it's a technology that's here. Let's figure it out, let's try to adapt to it or adopt it if that may be the case. What's something new in AI that you're excited about? Naeem, you want to go first?

E (8:00)

Yeah, I mean, sorry about that. In my world, it's all about content creation. And so I think that the piece that's compelling for us is that suddenly anybody with taste and anybody with ideas will the ability to create production grade things. And so from an enterprise or a business perspective, essentially any small brand or mom and pop shop will now have the ability to make super bowl commercials, which has never happened before. And so you can essentially be on an even plane where you're no longer, you don't have these enormity of these massive brands just out publishing you and creating these things. And I think that that's true with language models as well, is that you have this ability now to scale impact in a way that you couldn't before. And I also don't think that this is happening. I don't think people realize how much is going to come in the next year across all of these industries. And I think we're seeing it at the research level. It's really in the next 12, 14, 16 months that you're going to start to see it kind of really impact at the enterprise. But yeah, thematically that's what's interesting is now any business on earth will be able to punch way up. And I think that that's only a good thing.

C (9:16)

I actually love this because it'll show you that everyone here is going to become a manager of AI, but you still need to have the right ideas, right? The storytelling the AI isn't going to do for you, but he's going to help you execute. And I think we see that across a lot of different roles. Everyone is going to become a manager of AI and start to delegate their tasks and that will be a big mindset shift for a lot of people.

A (9:39)

Right, but we'll still need people, right? Because who's going to tell the AI to do stuff? Salmia, what are you excited about?

D (9:46)

Of course, I think without talking about agents, this would not be complete. I'm very excited. I think it's beyond hype at this point right now, but echoing the sentiment that we have here, it's about how do you use it effectively for a use case and how do you use it with all the guardrails that need to be put around it? And also, where does human creativity and human judgment, decision making come into the whole picture along with the agents? I'm very excited for that part.

A (10:15)

Yeah. So in advance of this conversation, we had a prep call and in the prep call we were kind of talking about this idea that AI, to all of us regular people, this kind of translates to just to automation. And it's great to. So I don't need to email a thousand people. I could just automate the email or whatever, whatever it ends up being automation. Is that, is that like, is that a misnomer or is that improper way to think about it or is that really where we're at? Naeem, you want to go first?

E (10:47)

Yeah, I think that that's probably accurate, but I think it's, it's like the compelling thing with AI is you can, you can automate on unstructured data and that's what's ultimately the power of it. So it's like you can actually make agents that can go out and do tasks that they weren't pre programmed to do. And that's what's ultimately really compelling is that you can start giving instructions that are fairly vague, which is like, hey, this is my agent that does SEO, let's say, and I have some vague idea of what that means or I want to get on the top of these keywords or whatever. You now suddenly have the ability to create things. I can go out there and just do that and also get better at that and then take feedback from you that it's getting better at that. And so I think that's what's ultimately compelling. It is automation, but it's like, it's automation that can learn and that can get better and that can iterate, which is very different from static animation or automation, which is kind of in the norm to date.

A (11:40)

Right, right. So basically the AI will understand that I like my coffee in such a way and that you should order it for me every day and I want it to arrive by 9:00am Is that.

E (11:52)

Yeah, theoretically. Yeah.

C (11:54)

So there's also like, there's a little bit of hype here. Right. Like a while back last year there was this device that came out and people are like, you know, in the demo, like when they announced it, they're like, oh, I want to book a trip with my four kids to London and 1, 2, 3, and it's done. And I'm like, that was BS. No way. No one will book that long of a Trip with four kids in three seconds because there's so much subtlety. And so there's actually a big business opportunity right now to help go through those that last mile and make it from like that's a cute pocket, but you can really use the device or the product to. This actually works because you thought through all the steps around like, you know, just think of flying. Right. Like throughout my lifetime I went from like being willing to pay $50 less and have like an extra 5 hour layover to like paying $1000 more just to have a direct flight and not having to do a layover because like, you know, my time becomes more and more valuable from being a like poor grad student to like being a CEO. And so it's like all these subtleties like the AI doesn't yet know about people.

A (13:00)

Well, I mean isn't it presumably learning about you and knows that you want to sit with your family or that you prefer direct flights?

C (13:07)

Isn't that over time but like right now no company has yet really honed in on that. It's definitely going to come.

A (13:13)

Yeah.

D (13:14)

But again, not every problem is an AI problem as well. Like I think that's another thing. AI is an enabler for us. We need to make sure that we're using it for the right problem to solve. Right. And some deterministic problems do not require AI at all. You could just solve it by connecting with other software. So it's important to understand what use case and what use are we putting this to.

A (13:37)

Yeah. Or just you could solve it by thinking. Yeah. So this, I had this. So Salmia, your company does basically like AI assisted search. I had this scenario recently where I was looking for a bunk bed for my kids. I have two kids. One is a twin bed over a full size bed and the bed could be no more than 55 inches high. And this, the bottom bunk would have to be detachable from the loft frame. And I searched for that thinking like, oh my God, what am I going to come up with? And actually came up with like a fair amount of reasonable results. And that's. That seems like a change. I feel like search is getting better because of companies like yours. Am I wrong?

D (14:15)

No, it's definitely. I think one of the things that is going to change is more natural language and more conversational search and more descriptive way of looking for things. And if we are just looking at the technical specs that a retailer might have, that will definitely not be sufficient to pull this data through. And that's where companies like Lilly what we do is we try to understand the consumer language and try to weave that into the products so that they are much more discoverable wherever the consumer is searching for. And so just to give you an example for address, we have like 2000 different commercially viable attributes that could be added. And this is just growing because the user searches are changing over days and the trends are changing. So these are just dynamically growing how users search for things.

A (15:03)

Right? And it goes from buying the dress to renting the dress. I mean, like, oh my God, the world of possibilities open up in front of you. It's great. I'm kind of really excited about this part of what's Next. So speaking, we're having a conversation about AI. Let's talk about headwinds. Moon Valley is basically trying to disrupt the film industry. That's wild. You.com, you want to take on Google, right? I mean, that's what you're doing. But this is actually hard. Anyone with a pay per click business model right now is literally freaking out because they're losing out on search entirely right now. Or the creators in the world who actually produce, who make the super bowl commercials are freaking out. So you're talking about building a product that's very disruptive for pretty significant industries. So how do you sell a product that is maybe difficult for society to kind of accept and handle?

E (16:00)

I think that's probably the biggest mistake that most AI companies have done so far. I think that you've got technology companies that are sitting in San Francisco making these things for people that they don't understand. And then they're coming and just kind of deploying it and they're saying things like, well, maybe that job didn't need to exist in the first place. And it's just like a very tone deaf and ridiculous way of approaching it. And I think that's why you're not seeing AI really take a lot of mainstream industries by storm. And more importantly, I think that's why there's a lot of antagonism towards AI. And so we're hoping that we're doing things a little bit differently. For us, we have a movie studio in la, we have half of our team are filmmakers, and as I said, we've paid artists for all the data that our model is consuming. And our model, as we think about it, as generative videography rather than generative video. I think that the problem is, and again, you see tech companies come in and they have this notion that, well, you'll just write a few words and then a movie will show up or write a few words and a Super bowl commercial will show up. And that's what somebody who doesn't have taste says. The people who have taste know that that's nonsense. And like, nobody here want to watch a movie that I made, right? Like that is objectively going to be shit, sorry, I would watch your movie. But what I would love to do is enable the people who have these incredible ideas to actually do that. And so with the super bowl commercial example, I think ultimately what's exciting for us is the fact that everybody will be able to make super bowl commercials doesn't mean that everybody's going to watch super bowl commercials. However, now it does mean that the most creative people and the people with the best ideas will have their content surface to the top rather than whoever can pay the most money to kind of get there. And we've all seen the commercials. We all see probably like 70% of them. And we think that we can do better, right? And it's always people are talking about that top 10%. And I can guarantee you there's at least several people in this room that will be able to create if they had an entire production studio at their command, they could make something that was just infinitely better than most of what you see out there. So I think that it has to be creator led and I think it has to be industry led. You have to do it with the industry. You have to listen to what the people want and then build for that rather than just coming in. And just like we're the technologists, we know what you know we do better than you. And I'm hoping that more AI companies actually start to follow suit and do that. But you know, that's how it is today.

B (18:38)

We're going to take a quick break, but when we're back, Diana asks Richard about how his AI chat search assistant, U.com has the lofty goal of taking on Google.

A (18:51)

This message comes from Square. You probably know Square from your favorite local spots, but you might not know there's a lot more to Square than meets the eye. What started as a little card reader is now being used to rapidly scale, build loyal followings, cover cash flow gaps, and expand to new locations. Wherever your business is growing, Square meets you there. Go to square.comgoinc to learn more. Okay, you want to talk about taking on Google?

C (19:26)

So actually what we found is that Google gets like majority of very simple questions. Like people ask like, who's the president of France? What's the weather tomorrow? Like, how old is Obama? Like none of those questions you can do much to be 10x better than Google. It's just like you get the answer within one second. There's an upper bound of how amazing that answer can be about the weather. So we actually focus much more on enterprise productivity now. And that is kind of the killer use case, making people really productive when they would have to spend hours and hours doing something. I think the impact on jobs is actually really tricky. 150 years ago, over 90% of the entire world's population worked in agriculture. If you now ask like, who here in the room wants to work with their hands again in the field in the summer, in the heat and the cold and the rain, no one would say, yeah, I'd rather do that than a tractor. Like, you know, everyone loves a tractor, but no one loved the tractor and it took their job 100, you know, years ago. Like, and over the last couple of decades when efficiency kept going bigger and like more and more. And so I think, you know, when, when you look at certain jobs, just like in, you know, most enterprises have sales and service right now if you make service people twice as efficient with AI, you may actually let go half of your service folks. But if you make sales twice as efficient, you're just selling more. You probably will hire more people because now you have to do customer enablement and customer success and all of that, right? So just, you know, it's a very tricky thing. Like illustrators, for instance. The value of an illustration used to be several hundred dollars, now it's a few cents and that is just hard to go back to. And so we will see multiple sort of ways of creative destruction in some industries. Now the nice thing is about search. No librarian could have ever searched the Internet for you anyway. So it unlocked a brand new kind of use case. And we'll all work on higher levels of abstraction where you just have an idea. Now you can have a movie that used to be like $1,000 minimum project if you want a really crappy one, and millions of dollars if you wanted to have a decent one. And it allows us all to be more creative. So if your goal is the output, you're very excited about AI. If your goal is to get paid by the hour, you will hate AI. And so there are only two areas where people love the output and not the hours work, and that's medicine. No one wants to have more jobs in medicine. Everyone just wants healthier people as cost efficiently and nicely as possible, like scientific research, fusion, new batteries, like new medicines, new drugs and things like that. Like that. Those are sort of two areas where everyone understands. Yeah, it's about the output of an industry, not how many hours of work. But most industries are very different. Yeah.

A (22:10)

Do you have any thoughts on how to make this transition more palatable?

C (22:14)

I have lots of thoughts. I think, like, in a weird way, Europe is actually quite well positioned to make this whole transition more palatable, because in Europe, you have health insurance that's independent of your employer. It's just a civil rights, you know, and you call an ambulance the same way as calling the police or the fire trucks. It's just like, you know, it's a service that the government provides. Here it's not. And so then you have, you know, unemployment benefits. If AI takes your job, you get free education all the way to the PhD level. You have education, high schools that are paid for and organized by the state, not by the zip code. So if you have an overeager parent that wants to make the education of their kids better, education for all the kids in the state gets better. There are so many small things. The problem is that Europe has all these protections from the downsides of AI, but then somehow the whole culture is not set up to benefit from the upsides of increased productivity, because you also can't fire people. You say AI automated half of your entire factory. You now have 5,000 people idling. You're not allowed to fire them and use that productivity increase. And so it's a tricky balance to have the best of both worlds. That is somewhere in between the United States and Europe.

A (23:27)

So it sounds like a policy fix regardless, whichever side works or wins or whatever. So we're talking about future proofing a business today. I'd love to have each of you weigh in on this. What is one thing our audience of business owners and entrepreneurs can walk away from today and deploy at their companies? Like, what is one thing you suggest they do, like, immediately? And why? Naeem, why don't you go first?

E (23:54)

I think that AI is infinitely more capable today than you probably think it is. And I think that that can be kind of a challenge to unlock because you have to sift through a lot of nonsense to get there. But I think the most important thing is to digitize as much as you can and then to go and find AI that works on specific parts of your data and then build up from there. Like, if you're very intentional about it, there's things that you can do today which are unbelievable. You can build things custom for your exact industry and your business and things and we see that day to day. Like we, you know, in a past life I used to work at Zapier, which is an automation tech company that a lot of people here probably work at. And like, that's where you can. People that knew how to use that really well would just get superpowers. And so I think that it's one of those things that you can't just like sample simple. You have to just carve time out and just throw yourself into it a little while because the capabilities are immense. You just have to sort of build it. That was a very non answer to your question.

A (25:00)

But yeah, no, I mean what you said is just basically just do it, right? I mean, am I wrong? Okay, no, that's fair. It's fair. Just get started.

E (25:09)

Happy to help.

A (25:11)

That's good. Samiya, why don't you go. And then we'll end with Richard.

D (25:15)

Yeah. I would say to founders, understand what your core values are for your product. What is the value you're bringing to your customer? Because it's important to stick to that and drive more value to it. Because AI today, where it is today, is going to be much different tomorrow. So trying to optimize for that is probably not the right direction. I would say, like focus on the core value that you bring to the customer and how can you make it. For example, for Lilly, we said that we are going to stick to accuracy, relevance and customization as the core value that we bring. Then how do we use AI to accelerate our journey towards that? And so that's what I'd recommend.

A (25:53)

Great. Love it. Richard.

C (25:54)

Yeah. Think about every intellectual task that you're doing that is somewhat repetitive, that you could explain in reasonable English language and then just use a tool, you, dot com, chatgpt, whatever, and like try to explain that to the tool and then see if it can actually automate it. And if it's like, good but not great, try it again every two or three months and then try to get all the people in your company because it changes the new models that come out every couple months and then ensure that all the people inside your organization get leveled up and get certified to do the same.

A (26:33)

Okay, we have time for questions from the audience. Anyone have a question for these fine folks up here? Yes, sir.

E (26:39)

Creating a funnel with AI tools. How would you go about doing that? Say for if you are networking to go to many different business locations and have a coach, but then you want to set up a funnel and use the employees at the staffing agencies that you're pitching to, coming from A background in martech. I would say that the, the superpower that you probably want to unlock is data enrichment. There's tools out there that obviously I'm sure people use clay and similar things that do data enrichment, but if you do it yourself and you take it in your own hands, you can do just pretty incredible. And this is true broadly for prospect research and stuff like that too. But knowing your exact list and your targets and things like that, you can build custom prospectors just using ChatGPT or something similar that can basically go out and find out like exactly who you're looking to talk to and the org chart and things like that. So I think that data layer and if you use tools like clay, they'll give you something generic that everybody can use. Whereas if you do it for yourself and for your own industry, you can pinpoint the exact types of data that you need. I know a company that I helped with a while ago that they're focused heavily on small businesses in their local region and they built a very custom kind of thing that would go out and essentially look at like reviews that were coming in about that from in like completely different places. And based on new reviews, they would get a prompt to go and actually do something about it. So it'd be like, oh, this is a business that has recently gotten reviews that says, you know, you haven't. They don't, they don't offer delivery. That's a good chance for you to go in and kind of deploy that. So I think like, and you can't really do that with, with just like using off the shelf stuff. But if you do it custom with ChatGPT or whatever, you can get just like incredibly powerful, like a data layer that's differentiated.

A (28:47)

We did have another question. I think the lady in white over here.

D (28:51)

So my name is Jahari Sword. I have a question really for the entire panel.

A (28:55)

What would you say are some of.

D (28:57)

The fundamental things as founders, small business owners need to know around, not just.

A (29:03)

Know around AI, but tools that they.

D (29:05)

Should be using specifically around the responsible AI framework and the policies also around.

A (29:14)

Productivity tools that would be essential for.

D (29:16)

Them right now to see the maximum efficiency in the organization.

C (29:20)

I love that the question was like tools that organizations should adopt now to really make use of it. I'd be a bad CEO if I didn't say you.com? happy to talk to you after. But we do provide these role based training programs as part of a team subscription, for instance, and we found that that makes the difference between 3 to 5% adoption and 50 to 80% adoption inside an organization. It is really that role specific training. And we're working with this company called Inversity on these trainings. And it turns out of course they're also using LMS so that the training actually is like, like the role specific. So you say, oh, I'm a marketer in Austin for retail. Like, and then it just actually like gives you example prompts and like gives like asks you questions and then evaluates you based on your answers. Like that is very specific to that kind of role. And I think that's, that's really important to sort of combine that with the previous question to 11x is like an interesting tool. You too, because they will, will do the actual outreach. They will send emails for you and find folks on LinkedIn that actually like, you know, yeah, might be interested in your product. And then they give you a nice sort of, oh, they clicked on the email, they opened the email and things like that. So there's a lot of like automation. And then you can do research questions too. If you say, oh, I want my salespeople to be more efficient, you know, you just ask like something like you.com or chatgpt and it will, it will tell you all those answers. I'm, I'm surprised still how many times even the most expert users don't understand the full power. I give you an example. We have a PhD level person who loves us for research, hardcore biotech researcher. And he's like, I love your tool but you know what'd be really cool? If there was a toggle in the interface where I could say, only search over medical research journals, like academic journals, because I want a really legit output. And I'm like, I just took his question and I added in natural language. Now only search over medical journals from legit research outlets. And it did it. And he's like, oh, wow, I guess I could have just said it. And so it's like even expert users don't fully appreciate it. You got to really get into different mindset.

A (31:33)

Well, you've given us a lot to think about today and clearly we've barely scratched the surface. So thank you though, all the same. Salmia, Richard and Naim, I appreciate it.

C (31:41)

It. Thanks everyone.

E (31:42)

Thank you everyone.

B (31:45)

That's all for this episode of from the Ground Up. Our producers are Blake Odom and Avery Miles with help from Sam Gabauer and Hawa Otori. Editing by Matt Toder. Mix and sound design by Nicholas Torres. If you haven't already subscribe to All Ink podcasts on Apple Podcasts, Spotify or wherever you listen.

A (32:06)

Panoply.