The Unseen Side Of AI

SPEAKER_00 0:00

Welcome back to Inspire AI, the podcast where we explore how leaders, builders, and organizations can navigate an AI-accelerated world with clarity, confidence, and intention. Today's episode is one I've been especially excited about because we're diving into a side of AI that most, almost nobody talks about publicly. The legal, operational, and governance infrastructure that actually determines whether AI gets adopted inside enterprises. Most AI conversations focused on models, demos, productivity gains, and the latest breakthroughs. But behind every enterprise AI deployment are much larger questions. Who owns the data? Who carries the liability? Can organizations trust these systems? To help us unpack all of that, I'm joined today by Ben Hawkins. Ben's background is a fascinating intersection of technology law, enterprise software, intellectual property, and AI commercialization. His experience-bansed technology transactions, sales agreements, IP licensing, and governance frameworks shaping how organizations evaluate and operationalize AI systems. What makes this conversation especially valuable is that Ben brings visibility into the invisible shaping the next generation of AI adoption. Contracts, governance models, data rights, and the risk frameworks that most technical teams rarely see, which ultimately determine which AI systems make it into production. Welcome to Inspire AI, Ben. Hey, great to be here. Awesome. Why don't you tell the audience a little bit about yourself and what brings you here today?

Meet Ben Hawkins And His Lens

SPEAKER_01 1:50

I am from Richmond, Virginia. I made connections with some folks involved with Inspire AI and wanted to come today to talk about my experiences working at an AI company, working as a technology transactions lawyer, uh Wilson Sini, a big law firm, and my more recent experience working in kind of the solutions for solving the overeyeing of the internet. And um how do we know when we we are interacting with AI versus a versus humans? And it's not that we want to get rid of AI, but we need to know the difference, I think. So yeah, I'm excited to be here. Wow.

SPEAKER_00 2:33

Uh I I want to know more about that. All right. Let's

The Over-AI Internet And Human Value

SPEAKER_00 2:37

talk about the over-AIN of the internet. What do you see, the biggest trends, the biggest concerns, the things that our audience here are going to be most interested in learning about? Because I think that with all of the content out there, people are using their intuition to identify when the content is AI generated and they're getting sick of it, right? Yeah. So what do you see from your perspective?

SPEAKER_01 3:03

Yeah, I mean, I think my biggest perspective is maybe not as a lawyer, it's just as a enjoyer of the internet, a consumer. And um it's kind of discouraging. It's not as interesting to me to go on you know, Substack and read a bunch of things that someone prompted ChatGPT. So I think that there's a there's a the thing that I'm excited about is the value of human, real human input and real human, like you said, intuition and ideas. I think that's where the premium is going to be on the internet. And um I think you know, there's definitely a leveraging of AI that we want to continue. And I think we need to know the difference, especially so that's in your kind of personal life, but in an enterprise and a business, it's a little bit of a different issue. We're talking when you talk about agents, we really want to talk about like the legal concept of agency and keeping these agents within the principal's control, I think is something we really need to start thinking about on an enterprise level. So we you we want to be able to set agents out there on the internet to do work for us and to really expand our productivity, but there needs to be an understanding on both sides of what that that agent is permitted to do, and there needs to be some way to prove that an agent is there on the principal's behalf. And so I think that's something that's super interesting and a little bit scary to think about. Uh, but there's solutions there, and it's something that an enterprise level just needs to be under control for security and um even thinking about for our customers.

SPEAKER_00 4:40

Yeah, what what are some of the more generic controls that you can speak about that help us identify the AI-generated content and whatnot?

SPEAKER_01 4:50

So I think it comes down to really, like you said, in human intuition at some level, but as AI becomes more indistinguishable from humans, uh, I at some point we do need technical, technological solutions. And that's where you know where I work now, World Foundation and Tools for Humanity has been working on that for a few years. Just to contrast that against some of my other experiences working for AI company that's doing customer service, where you might have more of a direct control over, and I'm not you know speaking for anyone, obviously, but there there is an aspect of of consent and uh a knowledge about what you're like when a person needs to know or should know. I think if you're calling a service, you're going out there and asking for something, you know, an agent can do that. But I guess there is a situation where what is really I and I think this is where the enterprise concern comes in. Like it is an AI genuinely and correctly representing the information that that the enterprise wants it to represent. And so then I think you're talking about hallucination issues. World Foundation AI is reaching a point where AI can can fake, it can take out watermarks, it can oh boy, reproducing, it can go around all these kind of more primitive controls. So that's where World Foundation is trying to solve and is we need to have some sort of you basically have to go in person, and it's a very technical, very complicated technical solution because AI is now able to fake a lot of the things that we have been relying on for so long. Again, like you know, this this year, the next 12 months, like that stuff is gonna be just not really gonna work all the way. I think there's gonna be certain things where it's fine. Um, but you know, when you get into the more critical applications like on the internet, financial health, privacy, that kind of stuff, we're not gonna be able to rely on just kind of CAPTCHA. Like it's gonna, it's the AI is gonna be able to beat that now.

SPEAKER_00 6:55

For sure. Yeah. I don't even know why that's still out there anymore, but anyway, I guess it's something. It's something. Um, so

Technical Proof Versus Intuition Online

SPEAKER_00 7:04

to tell me a little bit about your thoughts on legal and governance issues in AI. Are enterprises underestimating things? And if so, what are they?

SPEAKER_01 7:13

Yeah, actually, I I think that a lot of the bigger concerns, the harder concerns for me are where there's not a regulatory framework in place. So privacy and especially in data security, I actually think is quite robust. And a lot of the highest tier top shelf AI vendors out there are gonna have have that buttoned up. That's to me not so much a concern. I think proprietary data, proprietary uses of models, I think is a tougher governance question, especially when you're talking about becoming reliant on these models. And I am a proponent, I'm kind of my belief is that we should create these kind of smaller scale models uh that are more tailored to specific use cases. I think that actually does relate to the legal and governance where we're it's kind of a dealing with a smaller, a smaller world when we have these kind of we let companies train on some of the enterprise models is something that I really believe in. I think the risk is a little bit is not as much of a problem for kind of the proprietary IP side. Once you have the security, like taking for granted that security and privacy are buttoned up. You know, I want to set that aside and talk about what we're doing with the kind of specific use case learnings, is what I'm most excited about thinking about from governance and legal issues.

SPEAKER_00 8:37

The specific use case learnings.

SPEAKER_01 8:40

Yeah.

SPEAKER_00 8:41

Can you say more about that?

SPEAKER_01 8:42

Yeah, so for example, like okay, it's great that we I think there's a concept out there that these general purpose models are more powerful. So they they're always gonna be foundation models you're referring to. Foundation, yeah. Foundation models are always gonna be the best. But there's two things going on there's cost of the foundation models, and there's also what is going into those and how specific they are to a particular use case that I think the foundation models are are not addressing. And then overall, longer term is it's diversifying, but it's kind of the fundamental capitalist thing that there's other solutions, there's competition. So I think the more that we can create more that we can foster going outside of the foundation models, you we don't if you're you have a very specific business use case, you don't need the the foundation models to necessarily do that efficiently. You know, you might have a marginal, marginal gains in certain ways, but I mean it's using a giant lifted truck to go down to the grocery store. Like you just maybe you just need a little electric golf cart. You know, I think that that's kind of the overall thing. And I think that applies to governance as well, where you're you're dealing with these huge ecosystems. Why not have a tailored tool? Uh that's kind of easier to manage in my mind from a legal perspective.

SPEAKER_00 10:02

That makes a lot of sense. I haven't seen that applied, but I definitely understand that smaller models, more fine-tuned to specific use cases, which should be easier to control because the outcomes are designed for your specific use case and it can't generally go outside of those guardrails that you're applying to it. Yeah. Yeah. So yeah. All right, let's roll into some of your experience.

Why Enterprise AI Adoption Drags

SPEAKER_00 10:32

Why would you say the enterprise AI adoption is slower than the public narrative suggests?

SPEAKER_01 10:38

Yeah, I think in my experience, there is definitely a who takes on the risk question. I mean, obviously, there's a large risk question involved with AI. In my opinion, in my experience, for more of the embedded larger businesses, enterprises, there's a notion that these AI startups basically should take 100% of the risk. And that, like, if you want, if a AI startup, if you want to come and we're gonna pay you money, like we're gonna have zero risk, by the way. I think that is not really a workable approach. Large company says, hey, AI startup, if you want this ARR from us, we're not taking on any risk. And then the AI startup has to say, like, of course we need that. But I think there should be more of a shared risk, a shared responsibility for developing these things, especially going back to my previous point of these AI startups are an opportunity to create a tailored use case, a smaller, more efficient model that reduces reliance on foundation models. So going into the negotiations, especially with AI startups, you know, I'm not talking about the foundation models that are these massive hyperscalers. I'm talking about more tailored startups. Hey, let's be more of a partnership, a collaboration where the risk is shared. Going into these, you know, what really slows down enterprise AI adoption is there are large risks for if you're sharing, especially when you're talking about PII involved, which can be as small as someone's name, but there are very large statutory damages that are definitely gonna hurt. But if you want to put data into these new systems that are so powerful, there is some risk. And I think that sharing that risk is something that I would ask the large companies to consider taking more of.

SPEAKER_00 12:35

That makes sense. That resonates because when I think about when I went through a migration to the cloud many years ago, the cloud was an open playground of getting infrastructure shifted and applications and software. And when you move to the cloud, if you don't know what the shared responsibility model looks like, you're leaving yourself vulnerable, right? It's great that you don't have you no longer have to manage your own servers, but you also need to lock down your data, right? So there is that. So that I think as time goes by, people wise up to what the actual risks are that they're dealing with through trial and error, sometimes painfully trials and errors, but yeah, but they they realize what it is, and then and then they go back to the negotiating table and say to the Googles and the Microsofts of the world, hey, I don't like this shared responsibility model the way it is. So can you take on more risk? If so, we can we can have better dealings in the future and you can have more of my money. But until then, I'm not taking I'm not moving any more of my infrastructure. Things like that, I think, need to strike a balance in the in in the AI era as well. But yeah. So Ben, you know, thinking about the risk and these big companies not having taking much responsibility for it, especially since there's hardly any federal compliance or AI risk adoption. What do you think the AI vendor in enterprise trust is going here, Ben?

SPEAKER_01 14:21

Yeah, I think that I think that it's I do believe that we can rely pretty heavily on your privacy and uh security standards. Those are based on heavily regulated rules. And I personally believe that if you hit these, you know, industry standards that are well established, and that those are we can we can trust that. And you know, I don't think we need to put these burdensome impossible uh liability numbers on them, basic insurance coverage. I think there just needs to be basically industry standard trust. If if and I think there's also how long these companies have been operating, I think that that's something we that we can really rely on.

SPEAKER_00 15:08

Yeah. So tell me your take on this, Ben. Are companies exposing themselves every time employees paste proprietary information into the AI systems? Yes.

SPEAKER_01 15:18

But yeah, it's straightforward because I think that this is a perfect question because it I'm there are differences, there are key differences in what AI can do with that information. Ultimately, I'm not sure that it's more than having a Gmail account. I mean, I think that there's definitely it can go far. There's it's more of a target for hackers, for example. But I think that this is kind of getting back to the shared risk that I was talking about earlier. That yes, this is a reality. You're when you put things into the internet. If you want zero risk, then write it down on a piece of paper and put it in a safe. But if you want the the benefits of technology, yeah, there's some risk that you put it out in the internet. It can be found theoretically, however hard that might be. But yeah, there's definitely a risk. And I think that risk is something that enterprises uh should be more willing to take on. That's kind of my personal belief. I, you know, there's that's kind of an oversimplification, obviously. But that's kind of where what I want to get on my soapbox about is that, yeah, there is risk, and I think it's worth taking.

SPEAKER_00 16:28

Yeah, and do you think that employees really understand that? Do you think they they need more education?

SPEAKER_01 16:35

Well, yeah, I definitely have experience from this pre, you know, pre-AI uh, and when I was at a tech working with tech companies at a law firm, employees don't really know how to use social media correctly. So yeah, I think there definitely needs to be education. I think you're gonna run into the same kind of human error, human judgment error problems. So, you know, and there's a there's a fear that error can is exponential in what it can cause. But I definitely think there's a responsibility for employees to to be careful, you can put in systems, but yeah, at the end of the day, we are still humans working. We're not completely turned into agents yet. So I think there's always going to be some error and some risk. And um yeah, I think we should it's definitely worth mitigating and doing education and training. And um, but I think we can get there.

SPEAKER_00 17:31

Yeah. So

Contracts Leverage And Non-Negotiables

SPEAKER_00 17:32

I want to circle back to something you said a few moments ago uh about l leveraging the legal system to ensure that the policies are in place and we can like have faith that that there's good structure around the governance and protections and privacy and all of that. But what do you say when people don't read the fine print? Like uh a Google or uh or an Amazon puts a contract in front of you that's five pages at six font, and they look at that and they're like, uh, forget this. I'm just gonna agree to it because everybody else is.

SPEAKER_01 18:12

Well, if it was only five pages, I would say that's pretty short, actually. Um I have no idea how many pages they are here, Ben. Yeah, I think that's hard because I think you're getting kind of hinting towards the problem of leverage. You know, what actual leverage do different parties have in the negotiation? Basically, if you're going against a huge hyperscaler like Amazon, you're essentially not going to be able to negotiate those terms unless you're another hyperscaler at the same kind of scale. Often, though, that comes down to the leverage between the two companies involved. So, as far as what what needs to be in there, there's obviously going to need to be some sort of risk shifting of indemnification and limitation of liability. Ultimately, that is dictated by insurance. Although, so there there often are demands that there's a certain level of insurance. Those demands can be a little unreasonable and possible to meet. So I think there's sometimes there's a depending on the level of sophistication and the experience of the companies with tech generally. I think those negotiations can be a little bit silly, to be honest. It's for a hundred thousand dollar contract, you're saying you want a hundred million dollar insurance policy. That's silly. That that can happen. I think ultimately we get there, how much pain there is and how many emails there are can vary. But yeah, certain things are non are non-negotiable. I think that going again, going back to going back to your question about when uh do your employees paste proprietary information D AI. You know, something that I do think should be essentially non-negotiable is if there's uh breach of confidential confidentiality, your confidential information needs to be secure. If there's any any breach of that, they share your information with competitors or something, that's gonna be a breach. So that that's something that should just be covered. It's basic. So there's basic confidentiality terms that are gonna go into these contracts every time, you know, you can't disclose or share confidential proprietary information beyond what you agree to do, and that's usually only for internal purposes. So there's this notion that if an AI company, for example, trains their proprietary model using your a company's proprietary information, that proprietary information is some have or queryable, like I can say your competitor can go in and say, What is company A's price list? And that's a misunderstanding of how AI works, from what I've seen. That's the AI model is learning general, general lessons, general perspectives. It's not gonna be able to, you're not gonna be able to query that and find out the specific almost every AI system that's used in an enterprise level is gonna work like that. So that that's where if we talk about what's non-negotiable, yeah, confidentiality, I think that should be covered. If that happens and your confidential information is shared with a competitor, that's gonna be covered by basic contract terms. Um there's willful misconduct, fraud, and gross negligence are not waivable for liability. So there's also a notion if an AI company goes in and does something that's just purposefully sharing your proprietary information. Basic contract terms are gonna that's gonna be unlimited. You don't need to, we don't need to like fight over insurance amounts for that. It's not waivable. So that's just kind of something I go into these negotiations with. Yeah, I do think maybe I'm in. Created myself, but I believe in like legal principles are pretty good for this kind of thing. Even though AI is a novel technology, you know, at the end of the day, it's the legal principles still cover it. It's like just because there's certain things that are on the edge, you know, copyright, I think, is getting a little bit confusing and weird, but you know, for basic enterprise application of tech, standard contracts are covering this kind of thing. There's just a little bit more uncertainty because it hasn't happened that much yet.

SPEAKER_00 22:32

So right. All right, let's pivot here for a second, Ben.

Agentic Procurement And Human Approval

SPEAKER_00 22:36

How does procurement change when software becomes agentic?

SPEAKER_01 22:41

Ooh, so this is super complicated and getting into a real kind of sci-fi world. But I think what we're starting to talk about are agents going out into the internet and procuring software that they need to do the jobs that that we've set them out to do.

SPEAKER_00 23:04

So for example, if we say Enterprise says go out and secure a bunch of uh your homegrown app to make monthly repurchases for you and go and find various vendors online that whatever the cheapest price is, just buy that.

SPEAKER_01 23:27

Yes. And to do that though, you need to have this other little software or API that allows the agent to do that. So then I I guess expanding that's a very simple example, but expanding that out to an enterprise scale, uh, you're looking at agents making pretty substantial procurement decisions. And that that's where you know, solution like world ID and agent kit, some of the solutions that that we are working on are going to become pretty essential. I think that you're always already seeing that with uh some applications on like Twitter, something called BankerBot, which is you just basically tweet at a bot and a bunch of times until it goes out and figures out how to do the thing you're asking it to do. I obviously the the technical function of that is super complicated, but just kind of like as a preview, that is something that's gonna start happening and probably is gonna start becoming the norm where you set these agents out and they are permitted to do a lot on your behalf. And so harnessing that and keeping that under control is gonna be maybe the the kind of like central question here soon.

SPEAKER_00 24:41

Yes, indeed. I feel like we we should talk a little bit more about this because this is a hot topic.

unknown 24:46

Yeah.

SPEAKER_00 24:47

So one of your examples of agentic software creating or wreaking havoc in in society, uh I think makes me think about open claw. Yeah. Right now, you can hold it to a virtual machine and you can give it permissions to send messages and things like that. But I feel like with the what's the name of that agent social media club? What is it called? Multbook. Are you familiar with that?

SPEAKER_01 25:22

Multbook. Yeah.

SPEAKER_00 25:24

Yeah. So okay. So when I think about that example and open claw and how Moltbook was created to give agents agency to have conversations with themselves and concerning discussions, if you will. You release something like that into the wild and let people do those things themselves. What happens? Like, how do we put the cat back in the bag or the genie back in the bottle here?

SPEAKER_01 25:51

Um so obviously, very technical. World ID has an as agent kit, and so how do we harness it and keep those things on track? I think there's building systems that require uh certain level of human approval. So human in the loop kind of solutions are just gonna be essential. I think that there are potential solutions where we let, you know, maybe there's a maybe there's a continuum where the agents are permitted to go certain level in certain areas. But I think for the foreseeable future, there's going to be spaces that are protected from really too much agency, talking about agents. Um, bring back again where there's financial data involved, basically PHI, protected health information, other kinds of these sacred spaces. But I think that's something we need to decide as a society and enterprises that are deciding this. You know, I think we're talking about a little bit of a different question, which is what how what do we want humans to be uh influencing in our business? Yeah, I think that I think that there's certain fundamental things that we should always that we're gonna want to always keep, you know, a human-centric experience. And I think we should actively avoid scenario where yeah, like Mote Book, where their agents are just kind of running rampant and humans no longer even know what's going on. I I I think that that where that line is is kind of uh it's easy to say we definitely don't want it to go to the extreme, but where that line ultimately is is something that's actively evolving right now. And that way, bringing it back to kind of what what why enterprise AI adoption is slower, I think in in that way is an example of maybe it's good to be a little slower. I mean, I think that you know we should make sure that these systems are under control. I think using AI in enterprise is a lot of a different question than how much do we let agents just take over everything? I uh you know, I think that line is what we're balancing at again, going back to intuition. I think that's where it does make sense to kind of avoid the side the sci-fi nightmare scenario.

SPEAKER_00 28:13

Yeah.

Proprietary Data As A Competitive Moat

SPEAKER_00 28:14

Let's pivot again about proprietary data. Do you think that there's a competitive moat that needs to be considered? And what does that look like for companies leveraging their proprietary data to protect themselves?

SPEAKER_01 28:27

Yes, I think that I want to shift it a little bit to talking about how for AI companies, AI startups, again, not not really talking about the foundation models or hyperscalers, but I think that the the this is another reason I want to talk about enterprises who are bringing in AI vendors, sharing training data or giving rights, data rights to do model training for the AI startups benefits everybody. It's a shared, it's a shared benefit because the AI companies, yeah, their models get better. But that just means that the models perform better for the enterprise that is their customer. So that's that does become an important part of the moat that AI companies have, like how much of the proprietary data they've been able to train on. There is value in that in the negotiation when an enterprise is talking to an AI company. Yes, there's value in that, that I think should be part of the negotiation. But I think that it needs to be more understood that it's a mutual benefit. The the model is probably fine. Foundation models will work fine, they're less efficient, more expensive. So the more that we can build up these more tailored models based on proprietary data sharing, I think it's better for everybody. And it does give you a kind of indirect moat for when you don't have to rely on the foundation models for everything you're doing with AI. Yeah. So the future competitive mode, you know, maybe that's as we were talking about earlier. Maybe agents uh going out into the world creates a new moat that we don't even really know how to talk about yet. But yeah, uh, I think for the for now, yeah, proprietary data is an important moat, and building up proprietary models is gonna be an important uh part to be resilient in this space so we don't have to be so reliant on the same resources as everyone else.

SPEAKER_00 30:26

Yeah. For a second there, I had a glimpse of not all agents are bad on the internet, right? They're not yeah, they're not going out to try to attack and steal your information. What if we had good agents that are like battling the bad agents on the internet and you know they're they're detecting when the bad agents are are in the the vicinity and they're putting up a force field, right? I I don't know, shooting them with lasers and of digital data and saying, die, bad agent. That's that's pretty crazy, isn't it?

SPEAKER_01 30:59

Yeah, yeah. It's like the arms race kind of getting into like white blood cell territory. There you go.

SPEAKER_00 31:04

Yeah, exactly.

What Executives Must Decide

SPEAKER_00 31:05

So what do you think executives need to understand before approving the organization-wide AI deployments?

SPEAKER_01 31:12

I think there needs to be going again. I I definitely need to be basic industry standard, however that's defined, security and privacy infrastructure from these startups. I think that's 100% valid concern. Luckily, it is again a pretty robust regulatory framework, and the technical solutions there are well established. So definitely need to have that covered. But getting into kind of the the more esoteric what is the value there, I think that looking at what is the potential ROI is definitely important. There's certain applications that have clear ROI, and there's others that don't. And maybe you don't need that to have a clear ROI. For example, I don't think that uh there's a clear ROI to in my job, oh, I need to read 50 research papers on indemnification. I don't know if that has an ROI, but it's something I couldn't do without AI. So maybe it's there are more situations where it's kind of like experimental, kind of like almost like a why not YOLO kind of vibe on ROI. There's other applications that do. So I think that's a key. What are you actually trying to accomplish? Then if we look at kind of the IP questions, though, it's there's a balance between the risk of sharing proprietary information, confidential information, which we've brought up, and the value of sharing proprietary confidential information. I think that's a we need to have a real practical perspective on that. There are risks, yes. We're not gonna get to zero risk. Zero risk is nothing happens. So you wanna you want something to happen, there's gonna be some risk. I think the value of sharing the proprietary information and confidential information and business information and trade secrets and everything into these models, uh I think should be seen as more of a mutual journey. And so executives though should understand if sharing the proprietary information does apply. Is there a specific use case here? Is there the use of AI something that a foundation model is correct for? Then maybe sharing proprietary information is not necessary. And so when you get a you know, Chat GPT or Claude or whatever enterprise license, I'm pretty sure generally there's no training of involved there. But when you're talking about these smaller, more tailored models, that is important for trading. That's what we've been talking about. That in it's more efficient, it can be more specific to your business use case. So I think like, yeah, what's the goal here? Are we trying to do some kind of like YOLO AI? Let's see what happens, or is there a specific use case in which case sharing propriety information makes a lot of sense? And I think it should be kind of a more generally accepted approach.

SPEAKER_00 34:11

Yeah, I read a lot of medium articles on this subject, and YOLO doesn't work out very well for the companies that try it. I'll just be honest. Anyway, they end up shutting that down pretty quick. Yeah.

Cautious Optimism About 2026 And Beyond

SPEAKER_00 34:26

So my final question for you, Ben, is let's think uh a few years from now, what do you think the organizations will realize they completely misunderstood about AI adoption in 2026?

SPEAKER_01 34:38

Okay. I think that my personal opinion is that the risks were this is I'm probably I probably get roasted if anyone hears me say this, but I think that the Yeah, I think that the risk is overflown. I don't uh again, this is not legal advice, but yeah, in my in my opinion, I think that is kind of a social, I always think of the um the there's kind of like a social network of this, of that we all want it to work out. I don't believe that the robots are gonna take over. And I think that it's going to end up being much more under our control. I don't, I think that the fears that are out there are very scary. And I think there's some really smart technical people that definitely say that the risks are real. I guess I'm very optimistic that these systems are gonna work out for our benefit, and a cautious adoption is definitely the way to go. I think avoiding this because of, again, like we said, oh, there's some risk, that means that there's something to gain. Yes, sir. Well, thank you so much for it. I have no idea what's gonna happen.

SPEAKER_00 35:53

I don't think anybody does, but it's good advice. Cautious optimism, as they say. Don't yellow your business into your business. Yeah, don't do that. Make smart decisions, test and learn. Fail quickly, then learn from your mistakes, right?

SPEAKER_01 36:11

Yes, yes, and uh, I think let's all work together for mutual benefit. Let some of these AI startups take take some ground from the foundation models, it makes everyone everyone better.

SPEAKER_00 36:25

Indeed, it does. Alright, and thank you.

SPEAKER_01 36:29

Alright, awesome.

SPEAKER_00 36:30

Yeah, that was fun.

SPEAKER_01 36:31

Thank you. I wanted to do it again, I might already want to do another one.