Workforce 4.0: KUNGFU.AI’s Ron Green on Responsible Innovation

1 - >

Speaker 1: We are back discussing Workforce four point zero with my 2 - > friend Jerry's Vonnie. 3 - >

Speaker 2: Jerry, how are you, hey, Joe. 4 - >

Speaker 1: I'm great, I'm excellent, And you know, we all like 5 - > to open up these episodes chatting about where in the 6 - > world is Jerry's up, Vannie, So you're no longer in 7 - > New York. 8 - >

Speaker 2: Where are you right now? 9 - >

Speaker 3: Jerr? 10 - >

Speaker 4: San Francisco. 11 - >

Speaker 5: The beautiful city of San Francisco. 12 - >

Speaker 2: Now business or pleasure business? 13 - >

Speaker 4: Business? 14 - >

Speaker 2: It's all wow, you sound very serious saying that. 15 - >

Speaker 1: So he's not messing around, and we're gonna enjoin by 16 - > someone who's not messing around either. His name is Ron 17 - > Greeney's the co founder and chief technology officer at Kung 18 - > Fu Ai. 19 - >

Speaker 2: Ron, Welcome to the podcast. How are you. 20 - >

Speaker 3: I'm doing great? Thank you so much for having me on. Guys, Hi, 21 - > You're very welcome. 22 - >

Speaker 2: So I'm in Charleston, Jerry's in San Francisco. You're in Austin, Texas. 23 - >

Speaker 1: And before we started recording you, you were talking about 24 - > the fact. 25 - >

Speaker 2: How much Austin has changed since you've been there. 26 - >

Speaker 1: You're not one of these interlopers who just ended up 27 - > in Austin in twenty twenty twenty twenty one, you've been 28 - > there for a years. So to give folks a little 29 - > perspective in terms of the tech scene there, because I 30 - > think we've all heard it. Like I often joked that 31 - > when I first started doing these business pods, everyone was 32 - > in Silicon Valid. Then fast forward, everybody's in Austin. Tell 33 - > me about what that progression has been like in your city. 34 - >

Speaker 3: It's been pretty crazy. 35 - >

Speaker 4: You know. 36 - >

Speaker 6: I lived here back during the dot com bubble in 37 - > the late nineties early two thousands, and there was a 38 - > lot of growth at that time. But when the bubble popped, 39 - > you know, I really thought, oh, well that's it. You know, 40 - > that was maybe our big spike and growth. It's hilarious. 41 - > That was nothing compared to what's happened in the last ten years. 42 - > And really the biggest change is Austin's been a Austin's 43 - > had a tax scene since the early eighties. Actually, part 44 - > of what put Austin on the map was the founding 45 - > of a tech initiative that was a DoD back that 46 - > was artificial intelligence related, which we maybe can talk about 47 - > put it on the map. And you know, there were 48 - > things like Dell and things like other you know, really 49 - > major tech vendors, but it was all B to B 50 - > Everything here was just sort of B to B based, 51 - > you know, enterprise software, and it lacked diversity for the 52 - > longest time. And that's different now now in twenty twenty five, 53 - > there's everything there is. There are robotics startups, there are 54 - > consumer plays, CpG plays, et cetera. And it feels like, 55 - > I don't know, if I'm out at dinner downtown, I 56 - > feel like every other. 57 - >

Speaker 3: Person is a startup founder here in Austin. 58 - >

Speaker 2: Now that's funny. 59 - >

Speaker 1: And Jerry, obviously a lot of the work you do 60 - > is be to B with OSF, But what have you seen, 61 - > Jerry in terms of like the people you've been working with. 62 - > Obviously you're internationally based, but like, tell me about like 63 - > your connection often. 64 - >

Speaker 5: I've well, I've visited the Austin multiple times and i 65 - > know the city quite well. I've been interacting with companies 66 - > out of Austin in the past five years, I would say, 67 - > And it's it's impressive how the tech scene is has 68 - > grown in Austin and it has diversified. It's it's I 69 - > can't really relate to what Toronto was saying that just 70 - > to be more of a B two, be kind of 71 - > a setup. 72 - >

Speaker 4: It's it's uh. 73 - >

Speaker 3: I think. 74 - >

Speaker 5: That is is a good intersection between what they used 75 - > to do in B two B with all the new 76 - > technologies that came online, and that that made everything possible, 77 - > even the remote work made often possible because it was 78 - > before the pandemic, nobody really would embrace. 79 - >

Speaker 4: Remote or that well. 80 - >

Speaker 5: And then when that kicked in, Austin was like at 81 - > least a good second home, if not your headquarter. So 82 - > that that that also helped a lot. 83 - >

Speaker 1: All right, so we were we paid the Austin tourism 84 - > spot there. 85 - >

Speaker 2: But let's do one more spot. 86 - >

Speaker 1: Let's do a quick plug for kung Fu dot Ai. 87 - > Before we get into the agree, run quick plug us 88 - > about the company. 89 - >

Speaker 6: Yeah, yeah, Kung Fu Ai. We are an AI strategy 90 - > and engineering firm. We focus on AI exclusively. We have 91 - > since our founding nearly eight years ago. We help companies 92 - > developed there as strategy, we build you know, customer spoke AI. 93 - >

Speaker 3: Solutions for them. 94 - >

Speaker 6: And like I said, we've been doing AIS since long 95 - > before AI was cool. It's kind of fun to fund 96 - > to see the entire field blow up and you know, 97 - > the hype cycle just going overdrive. 98 - >

Speaker 3: It's funny. 99 - >

Speaker 1: You mentioned the fact that you even doing this, you know, 100 - > over eight years of AI and we often have guests 101 - > on the podcast who are like you who've been working 102 - > in the AI field before. You know, check BT was 103 - > a thing in twenty twenty three, right, So can you 104 - > take us back. Do you have any memories of those 105 - > early days when you're trying to explain AI to whether 106 - > it's potential investors, potential customers, Like what were those conversations like. 107 - >

Speaker 2: Now event as opposed to now? 108 - >

Speaker 6: Yeah, Oh, it's radically different. You know, I did my 109 - > masters in our official intelligence back in the late nineties, 110 - > so I've been working in AI so long that I 111 - > remember when it was you know, it was legitimately overhyped, 112 - > like we would we wanted things to work better than 113 - > they did. But you know, the technology and the data 114 - > just really won there. Back in let's say twenty eighteen, 115 - > all of our clients were early adopter, true believers. You know, 116 - > for the most part, they would reach out to us 117 - > and they would say, you know what. 118 - >

Speaker 3: How does be crazy? 119 - >

Speaker 6: Would it be possible to solve this one this one 120 - > point problem with artificial intelligence? Those days are gone. Now 121 - > we have people reach out to us and for the 122 - > most part they say, we need help with AI broadly, like, 123 - > how do we get started, how do do we develop 124 - > a strategy? How do we even know whether it's you know, 125 - > the investment is worth it on any given initiative because 126 - > it's so new to pretty much all the companies out there. 127 - >

Speaker 3: Wow. 128 - >

Speaker 1: And so now let's get to the current state of 129 - > AI and actually, you know, almost look a little bit forward. 130 - > Let's not go crazy and look like, let's say I 131 - > get to be like in ten years, because I don't 132 - > think anyone knows what it's going to like contagious. But 133 - > what we are seeing in a theme of this episode 134 - > of this podcast has been the workforce, how it's going 135 - > to immediately affect people. Workforce four point zero is what 136 - > Jerry calls it. 137 - >

Speaker 2: What are you seeing on. 138 - >

Speaker 1: The ground now in terms of how like whether it's bots, agents, 139 - > all that stuff, how does that work into fu AI's 140 - > world Right now? 141 - >

Speaker 3: More and more, you know, we're doing really there's been 142 - > an evolution. 143 - >

Speaker 6: I think we're on what I would saw call maybe 144 - > the third phase of our journey and AI as a 145 - > company coun to AI. In the beginning, it was mostly 146 - > sort of domain specific problems. We're doing computer vision solutions, 147 - > We're doing NLP solution. Things like that chat gbt K 148 - > blew the world up for a couple of years. 149 - >

Speaker 3: It was almost painful. 150 - >

Speaker 6: Everybody viewed everything as generative AI, like that's that's pretty 151 - > much the only way they could imagine interacting with it, 152 - > and they were blind in many ways to the real 153 - > difficulty of putting generative solutions into production without human oversight, 154 - > because you know, it's got that sort of jagged, jagged 155 - > edge of performance where it may may perform really well 156 - > in one use case and then fall down and you know, 157 - > be be as you know, as dumb as a kindergartener 158 - > and on the next one, and that made it really 159 - > really challenging. I'm really happy to say that things have 160 - > kind of come full circle and now clients understand that 161 - > generative AI is one tool and there are many many 162 - > more options out there, and that really the number one 163 - > thing I preach is if you have proprietary data that 164 - > will allow you to build some type of a solution 165 - > with it's predictive analytics, some new capability. It could be generative, 166 - > it could be anything, but it's domain specific. That's really 167 - > where you need to spend your time and your money. 168 - > The next wave, and I think everybody's talking about this, 169 - > and I think it's a little premature, but I think 170 - > it's going to come sooner than most people think. Is 171 - > agentic AI. The abilities of these agentic systems is growing 172 - > at a pretty amazing clip. A paper came out recently 173 - > was showing that the length of task that these systems 174 - > can complete successfully is doubling every seven months. Right, It's 175 - > been doubling for about four or five years. There's even 176 - > some evidence that it's actually four months now. So I 177 - > you know, again, we work with people that build AI 178 - > systems that go into production, so we're very cautious. We 179 - > tell them you need to have agentic systems on your radar, 180 - > and you need to be thinking about them, but you 181 - > don't need to be racing to put them at production. 182 - >

Speaker 4: Now. 183 - >

Speaker 6: They still need a lot of human oversight buying. That's 184 - > going to be the number one change to the workforce 185 - > over the next five years. 186 - >

Speaker 5: It's interesting that you mentioned domain specific, that this is 187 - > where it used to be, but it's somehow coming back 188 - > to the same area. I would say, maybe call it 189 - > a little bit differently. Domain specific for me is now 190 - > an intersection between technology and industry, like deep understanding of 191 - > your industry, deep understanding of your very specific situation, and 192 - > you can take a lot of these things that are 193 - > now being built off the shelf and you can customize 194 - > it and make it very specific for your very use 195 - > case and in your industry, with your entire set of 196 - > best practices that you've accumulated along the year. So I 197 - > think you know, like especially companies that are here for 198 - > a while and they are thinking, okay, are we going 199 - > to have a future. Well, if you treasured your best 200 - > practices and if you all the know how and everything 201 - > you created databases of knowledge, then I think you could 202 - > have a future just by reusing all of that and 203 - > repackaging it. 204 - >

Speaker 3: Yeah, totally. 205 - >

Speaker 6: And one example that I think is perfect that you 206 - > just scribed is we just built a system that were 207 - > live late last year for a publicly traded company they're 208 - > doing roughly two point five billion in loan factoring. That 209 - > system was completely manual, all human decisions. It was business 210 - > hours nine to five, five days a week, twenty four 211 - > to forty eight hour turnaround. But because they had this 212 - > giant wealth of historic data, just like you said, we 213 - > were able to train a model to make that decision 214 - > process almost automatic. It's now nine seconds worst case nine seconds. 215 - > They moved from business hours to twenty four to seven stands. 216 - > You can get loan factoring decisions on Christmas if you want. 217 - > And because they'd collected so much data, we were able 218 - > not only to automate this decisioning process, but fraud went down, 219 - > chargebacks went down. Like it was just a complete virtuous loop. 220 - > And it's an example where you can take historic data 221 - > and really leverage it to do something powerful and transformative 222 - > for your business. So they saw a quarter billion dollar 223 - > market cap bump when they announced it in their sharehold 224 - > letter that month. 225 - >

Speaker 4: Wow. 226 - >

Speaker 1: And what are you seeing in terms of surprising because 227 - > you mentioned earlier about like how general of AI is 228 - > just that it's just literally just the you know, the 229 - > chatchype's of the world. But I think AI is going 230 - > to be the future. But in terms of what you're 231 - > seeing now in the playing field of AI, is there 232 - > something that you've worked with a client that were. 233 - >

Speaker 2: Like, oh, I don't know I could do that. 234 - >

Speaker 1: You know how they always find like whether it's like 235 - > you know, like apple side of vinegar was al all 236 - > of sudden was like a weight loss here or what 237 - > after for one hundred years it was just the apple 238 - > side of vinegar. 239 - >

Speaker 2: Is there something with AI that you've noticed that's like, oh, 240 - > I didn't realize we could do that. 241 - >

Speaker 6: You know, I don't exaggerate, but I don't know if 242 - > there's a client that we've worked with in the last 243 - > eight years that wasn't surprised by the capabilities, right. 244 - >

Speaker 3: It. Yeah, there are. 245 - >

Speaker 6: Definitely some of some who over anticipate or overestimated, rather 246 - > some of the capabilities. Every now and then, once in 247 - > Blue Moon will have a client come in who will 248 - > think that it's it's literally like magic, like can you 249 - > just point can you point the AI at our data 250 - > and then just tell it to fix all the problems? 251 - > But those very very rare cases aside, most of the 252 - > clients we work with are just floored by the scope 253 - > of the capabilities now and they don't really understand that 254 - > from like a like a language understanding, speech recognition, you know, 255 - > coding capability, math, reasoning, on and on and on. It's 256 - > either at human level or seem to be at a 257 - > human level across the board. And that's just you know, 258 - > scratching a surface. 259 - >

Speaker 5: Wait until you will be able to talk to your AI. 260 - > That's gonna be quite transformative. It's like very natural way 261 - > of communicating with the AI. We've been doing some things 262 - > that in some poss and now you've seen you've seen 263 - > the models that are coming out. 264 - >

Speaker 4: Now, okay, how can we. 265 - >

Speaker 5: Bring these into some early use cases? And the results 266 - > are mind blowing. I guess, like, wow, it's because humans 267 - > would much rather communicate in a more natural way to 268 - > assist them that like the old way of typing and whatnot. 269 - >

Speaker 3: That's exactly right. 270 - >

Speaker 1: And Ron, what do you see in terms of with 271 - > this AI and the capabilities? You know, one of the 272 - > things that the Coldhart facts of it is just the 273 - > amount of jobs that may be lost by it. And 274 - > you are seeing how a lot of either tech founders 275 - > or CEOs have been very I don't want to say heartless, 276 - > but they've been almost like proud of the fact that 277 - > they're developing something that's going to. 278 - >

Speaker 2: Put a lot of people out of work. 279 - >

Speaker 1: You, as a tech founder, how do you balance that 280 - > because you as a futurist and thinking in the future, 281 - > you kind of see you read between the Tea's all 282 - > that jazz in terms of looking at the future, but 283 - > there's also the human element that you have people on 284 - > the boots on the ground now working how do you 285 - > balance that and in a way that it doesn't come 286 - > across as just being a complete a hole what potentially 287 - > a lot of these founders are looking like right now. 288 - >

Speaker 3: And they really are. It's ridiculous in my mind. You know, AI. 289 - >

Speaker 6: AI really can be and should be augmentative as much 290 - > as possible. The idea that you're just going to come 291 - > in and leverage this new technology and you know, put 292 - > a bunch of people on the street. That's not why, 293 - > that's not why I've spent my career trying to build 294 - > intelligence systems. Right now, look, I get it, that's going 295 - > to be a side effect. I like to joke with people. 296 - > They'll ask me, you know, is AI coming from my job? 297 - > And I say, look, AI is coming for all of 298 - > our jobs eventually, right like, But it's going to be 299 - > it's going to be a much slower and sort of 300 - > fractal process over time, going to replace parts of tasks, 301 - > and certainly some tasks are going to be eliminated entirely. 302 - > But you know, five years ago it was predicted that 303 - > radiologists would be completely out of work by Neil, that's 304 - > not the case. Closing that gap on that last five 305 - > percent is enormously hard from just a completely you know, 306 - > eliminating human job perspective. I'm super optimistic about the future 307 - > and I feel like, you know, I get it. You know, 308 - > change is tough. When the camera was invented, they thought 309 - > that was the end of you know, portraiture and painting. 310 - > And when you know, film, photography and video was invented, 311 - > they thought that was going to be the end of plays. 312 - >

Speaker 3: No, no, no. It just expands the landscape on what 313 - > we can do. 314 - >

Speaker 6: And so I'm extremely extremely excited about what i can do, 315 - > but I'm not blind to the consequences. And last thing 316 - > I'll say on that, I'm actually kind of happy that 317 - > probably the those disrupted field is my own. 318 - >

Speaker 3: It's it's it's computer programming. Right. 319 - >

Speaker 6: If there's any field that has been somewhat i won't 320 - > say automated, but but augmented in a maybe a negative way, 321 - > it's been software programming. And I think that that's deliciously 322 - > ironic and I love it. 323 - >

Speaker 4: Yeah, and indeed it's it's a little bit ironic. 324 - >

Speaker 5: But like up until now, I've even in in programming, 325 - > I I haven't seen something that has been done from 326 - > top to bottom by AI completely you know, with all 327 - > the corner cases, and especially all the integrations to sometimes 328 - > very legacy systems that are neither here nor there, and 329 - > you have to figure out a way around all of 330 - > these things. Yeah, it will definitely accelerate some tasks. It 331 - > will definitely accelerate some areas that even in in in programming, 332 - > it's it's now when you're looking at it's absurd why 333 - > it's taking so long. But there will be some other areas, 334 - > like think of just technical depth, technical debt is it 335 - > has been always a big issue in our industry, like 336 - > things that you would leave behind because you know. Now 337 - > you know, because you have more time, you can figure 338 - > out some of these things. I think that first we 339 - > will not see necessarily people being let go. I think 340 - > first we will see for sure an increase in quality, 341 - > an increase in speed. This is what's going to happen 342 - > the first step, and then we're gonna start shifting the 343 - > way we work. I still, you know, there is so 344 - > much work to be done. There is so much, you know, 345 - > like just your imagination stops you from actually see what 346 - > else we can do, and it's it's it's it's definitely 347 - > I don't see it as okay, this these jobs are 348 - > going to be gone. 349 - >

Speaker 4: They are going to transform for sure. 350 - >

Speaker 1: And Ron, let's get into the data piece a little bit, 351 - > because you know, I'm an old school radio guy, Ron, 352 - > and we used to have this adage that said garbage 353 - > in garbage out, Like if you recorded something poorly, there 354 - > was nothing you could do to fix it, right, And 355 - > so I think about this when it comes to building 356 - > these AI systems. If you don't have the right data 357 - > to build upon, what you're creating on the back end 358 - > is going to be craps. 359 - >

Speaker 2: It's like garbage in garbage out. 360 - >

Speaker 1: How are these companies that you're working with in terms 361 - > of like they've got data, I'm sure, but do they 362 - > have the right data? 363 - >

Speaker 2: Is it organized correctly? 364 - >

Speaker 4: Like? 365 - >

Speaker 1: Tell me about that, because that's one of the things 366 - > that people don't seem to really be discussing when it 367 - > comes to this AI revolution, Like the data has to 368 - > come from somewhere, and we've all heard the stories about, 369 - > you know, scraping of Google and Reddit and all that stuff, 370 - > But for companies that are using it for B to B, 371 - > it's like they're mostly they're internal data. So what is 372 - > it like behind the scenes, like under the hood, in 373 - > terms of the data that they're going to use for 374 - > these agentic AI projects. 375 - >

Speaker 3: Yeah, no, you couldn't. You couldn't be more right. 376 - >

Speaker 6: The there's an old joke, you know, we've been telling 377 - > this for a decade like that, the dirty secret of 378 - > artificial intelligence is it's like eighty percent data cleanup, right, 379 - > You're just you're mucking with the data. We have a really, 380 - > really sort of strong opinion about the methodology, and it's 381 - > that it has to be front loaded with two things. 382 - > One is, even before the data is you've got to 383 - > make sure that you have business ROLI associated with the 384 - > initiatives and that you can. 385 - >

Speaker 3: Articulate what the key performance metrics are. You'd be shocked 386 - > if the number. 387 - >

Speaker 6: Of systems that we are that we've seen built where 388 - > they built the system and then they had no way 389 - > to know whether it was performing well or even if 390 - > it was performing well, what impact I might have. 391 - >

Speaker 3: On the business. 392 - >

Speaker 6: But but just taking that, I think obvious blocking and 393 - > tackling for granted. The second phase is the day data 394 - > traditional software. You have an idea, you wouldn't you want 395 - > to go implement it? Great, There's there's nothing stopping you 396 - > but your own imagination, your own your own capabilities with 397 - > AI systems, and specifically AI systems that are supervised learning based, 398 - > which is most of them today. You're beholden to the data. 399 - > If you don't have the data, you can't do anything. 400 - > If you have bad data, or you have skewed data 401 - > or bias data and you train your model, the model 402 - > will suck. 403 - >

Speaker 3: Up all that bias and replicate it. So what we 404 - > do is we always fast fall on these initiatives. 405 - >

Speaker 6: If if a client of ours has an idea for 406 - > something they want to do, you have to go in 407 - > vet the data before you do anything else. 408 - >

Speaker 3: And there's really two parts of that. 409 - >

Speaker 6: One is do you have the volume, the quantity, the 410 - > quality of data that you need and can you access 411 - > it again. You'd be surprised, I'm sure you've see this 412 - > all the time how often it's like, yeah, we have 413 - > the data, but it's spread out across these ten data 414 - > stores and there's no way to get it together. 415 - >

Speaker 3: And then the second part is you've got to figure 416 - > out what the signal is in the data. You may have. 417 - >

Speaker 6: You may have relational tables with you know, thousands of rows. 418 - > We'll see this all the time, and you know eighty 419 - > percent of the signal is coming from four columns and 420 - > you need, You need to do the work. 421 - >

Speaker 3: You got to roll up your sleeves. 422 - >

Speaker 6: To suss that out. Once you've identified all of those 423 - > four things, you've got alignment on the RLI, the KPIs, 424 - > the data volume, and the data quality, then you can 425 - > go to modeling and start tackling those downstream tasks. 426 - >

Speaker 3: And too many we just see this over and over again. 427 - > Too many companies just want to skip those first four steps. 428 - >

Speaker 5: I think the data in itself, it's the structuring it 429 - > and making it, making sense of it from a business 430 - > point of your understanding how different bieceins relate to each other, 431 - > connecting it with each other, and then consuming it because 432 - > as you mentioned, the volume and then different systems can 433 - > be an issue. And then in some cases you know 434 - > there is also this synthetic data that you can actually 435 - > create if you know what you're doing and you understand 436 - > the space. That's that's i would say sixty seventy percent 437 - > of the AI work. I'm not sure what's your take 438 - > on it, but like this is what I've seen that 439 - > most of the work is being done. 440 - >

Speaker 3: Yeah, it's a tremendous amount of work. 441 - >

Speaker 6: Like you said, it really depends upon the organization and 442 - > their level of maturity in synthetic data can be a 443 - > great call in certain circumstances, especially in scenarios where you 444 - > have really sort of a skewed a circumstance, like like 445 - > fraud is a really classic example we see over and 446 - > over again where you know, luckily fraud is pretty uncommon, 447 - > but that means you are you're. 448 - >

Speaker 3: Typically dealing with a limited data source. 449 - >

Speaker 6: So if you can mimit that that data distribution through 450 - > statistical techniques and synthetic data techniques, it can really it 451 - > can really make a difference. 452 - >

Speaker 1: And what are you seeing ron in terms of the 453 - > idea that a company now has to. 454 - >

Speaker 2: Be an AI company. 455 - >

Speaker 1: For for a couple of years, Jerry and I had 456 - > this running joke where I would say, you know, people say, 457 - > if you're a business, you're also a technology business, and 458 - > I would rile Jerry because he hated he hated that, 459 - > like like he is going to make no sense, Like 460 - > if you're a butcher, you're not a technology company. But 461 - > now all of a sudden, you're seeing these companies who 462 - > weren't AI companies are now AI companies. And we've all 463 - > seen it where when when we when you go to 464 - > a VC or you go to Angel, They're not looking 465 - > at you unless you have like AI in that first. 466 - >

Speaker 2: Sentence, and obviously you know we did. You've been doing 467 - > this for a while. 468 - >

Speaker 1: What's it like now seeing all these people saying they 469 - > are now AI companies? 470 - >

Speaker 2: Does that kind of rankle some feathers there for you? 471 - >

Speaker 5: Run? 472 - >

Speaker 3: No, that's such a great question. 473 - >

Speaker 6: No it doesn't at all, because you know, if you've 474 - > been doing this as long as I have, the dream 475 - > was to get AI to work right. So for us 476 - > to be in this state where where people are jumping 477 - > on the AI hype train, that means we finally figured 478 - > it out. 479 - >

Speaker 3: We're doing something right. 480 - >

Speaker 6: And you know, back during the Internet dot com bubble, 481 - > everybody overnight was was an Internet company. These are these 482 - > are normal, I think, sort of business hype cycle things. 483 - > The difference is that, you know, and I granted, I 484 - > give that I will I will admit that this is 485 - > a very self serving thing I'm about to say, But 486 - > AI is really different than these other cycles. 487 - >

Speaker 3: If you look at the PC revolution or. 488 - >

Speaker 6: The Internet or mobile, all of those different technical ways, 489 - > there was a component on which the next generation could 490 - > build upon the first, and there was this sort of 491 - > positive feedback loop. AI is different, and it's really over the 492 - > last year where we started, we're starting to see the. 493 - >

Speaker 3: Next generation models being trained by the previous generation models. 494 - >

Speaker 6: So there's this feedback loop specifically around this sort of 495 - > technique called reinforcement learning with verifiable rewards that we're going 496 - > to get into this positive flywheel where we're going to 497 - > look back in a few years and we're going to say, 498 - > my god, that went fast, because this is the first 499 - > time in our history as a species where we have 500 - > a technology that can increasingly be used almost autonomously to 501 - > train the next generation of technology. 502 - >

Speaker 5: I'm maintaining my viewers, just because you have an AI 503 - > chad bought on on your website somewhere, that doesn't. 504 - >

Speaker 4: Mean you are an AI company. 505 - >

Speaker 5: And even more so, you should make sure that you 506 - > don't lose focus from your core business. Understanding your core business, 507 - > understanding the core industry. Yeah, you can build AI on 508 - > top of it. You can harvest every thing that you're 509 - > having your organization and put AI to kind of make 510 - > it better and faster. 511 - >

Speaker 4: But yeah, I mean, it. 512 - >

Speaker 5: Doesn't transform you into an AI company, just very west 513 - > features and we'll function now that. 514 - >

Speaker 6: This I could agree more really well said, I got 515 - > to tell you a funny joke. When GIP came out, 516 - > I'd say the first half of twenty twenty three, twenty 517 - > twenty three, we literally had potential clients calling us up 518 - > and say, hey, can we get some AI? And we 519 - > would say, well, what do you need? What you know, 520 - > what do you need? And they would say, I do 521 - > not care. I just need some AI or the border 522 - > is going to kill me. 523 - >

Speaker 5: Like that's kind of how you know, I tell you 524 - > one pound of AI, I. 525 - >

Speaker 3: Wish one pound of AI. 526 - >

Speaker 1: Ron to the question I was about to ask you 527 - > is like when is AI not necessary? 528 - >

Speaker 2: Like is AI for the sake of AI? 529 - >

Speaker 1: And you kind of lose an example there, But like 530 - > to me, I see, I'm seeing it how it works out, 531 - > like the fact that like Meta incorporates AI into Facebook 532 - > and Instagram, Like what's what is this doing here? 533 - >

Speaker 2: Like why why do I need that here? 534 - >

Speaker 1: Like Snapchat has like an AI bot that you can 535 - > talk to, so like where does it get to the point? 536 - > And I know, obviously when you're making a limit off 537 - > of this rodzof that's telling you to lose money here. 538 - > But but in terms of like the idea of using 539 - > AI just for the sake of AI, that seems crazy. 540 - >

Speaker 3: And it's a it's a waste of time and it's 541 - > a waste of money. 542 - >

Speaker 6: We'll see we we frequently more frequently than then you 543 - > would imagine, we'll work with our clients and stop them 544 - > from investing money and an AI initiative that's either something 545 - > that they could buy off the shelf or that there's 546 - > really not going to be there's not going to be 547 - > a return on the investment. It just looks cool, it 548 - > just feels cool, and maybe there was some excitement internally 549 - > from the company, but it's just an absolute complete waste 550 - > of time and money. 551 - >

Speaker 3: Happens all time. 552 - >

Speaker 5: I would I would think of a few cases and 553 - > where AI is totally not necessary. 554 - >

Speaker 4: But I especially when when we are dealing about. 555 - >

Speaker 5: Analysis on very specific data sets that that are limited 556 - > are limited in in in that and are limited in quantity. 557 - > You mentioned fraud. With fraud, you can get around on 558 - > some on some areas with with generating data synthetically, but 559 - > in some in some areas you just simply cannot, and 560 - > you need the human being to be there to actually 561 - > look at it and analyze it. For AI, you need 562 - > you need volumes of data. You need data that you 563 - > can kind of learn from and that's not possible in 564 - > all cases. Like take an example in aviation, if you 565 - > look at the accidents and if you would want to 566 - > kind of AI and see how you can build a pattern, 567 - > that's that's a difficult, if not impossible task. 568 - >

Speaker 4: So I would definitely not use. 569 - >

Speaker 5: AI to try to learn from it and try to 570 - > make the process of designing airplanes better. 571 - >

Speaker 4: You just don't have the data. 572 - >

Speaker 1: And ron I want to go through rail here of 573 - > the world regulation, It's been highly debated since twenty twenty 574 - > three how much the government should be involved because I 575 - > think obviously people are probably more concerned about the AI 576 - > apocalypse and Terminator and Judgment Day and all that, so 577 - > I know that's like the extreme of it. 578 - >

Speaker 2: But in terms of building these models, should there be 579 - > someone either just keeping an eye on it, like a 580 - > you know, a third party the independent to be like 581 - > you know, yay and nay things. 582 - >

Speaker 1: Because I think right now where it's so nascent that 583 - > our magic is just gonna run wild, like Okay, the 584 - > robots are gonna. 585 - >

Speaker 4: Run the world. 586 - >

Speaker 1: But obviously that's not the case in the short term. 587 - > But what are your thoughts on regulation? And you're in Texas, 588 - > we're not not big fans of regulation. So what what's 589 - > your thoughts on that? 590 - >

Speaker 3: Yeah, yeah, well I I am. 591 - >

Speaker 6: I am in Austin, Texas, which is like this little 592 - > this little blue racist in this the overad. 593 - >

Speaker 3: The It's man. 594 - >

Speaker 6: That is such a tough question because I'm I'm like, 595 - > I'm conflicted as much as anybody. I do think we're 596 - > gonna have a systems capable of super human intelligence. 597 - >

Speaker 3: I really do, you know how soon? Hard to say. 598 - > I think it's probably gonna I. 599 - >

Speaker 6: Think I think we'll have uh, I think we'll have 600 - > a g I artificial general intelligence. It'll be here before 601 - > we actually realize it, because it's going to be jagged, 602 - > you know. You hear this more and more where it's 603 - > not like we're gonna wake up one day and somebody 604 - > has flipped a switch and that there's this one AI 605 - > model and it has every superhuman ability in it. It's 606 - > gonna be jagged and we're gonna wait, We're just gonna 607 - > we're gonna realize, maybe in hindsight, that we crossed the threshold. 608 - > But I do think we're going to cross that threshold. Uh, 609 - > certainly within the next ten years, and probably sooner. On 610 - > the regulatory front. You know, it's so it's so complicated 611 - > because you've got the China issue, You've got you've got 612 - > competitive issues, You've got the downside that comes with any 613 - > type of technology being mandated by some you know board. 614 - > You know that old joke about you know, you know, 615 - > a horse designed by a committee is how you get 616 - > a camel. You know, if you get it up in 617 - > the you know, the worst of all possible worlds that net. 618 - > I think we're largely doing it right, which is we're exploring, 619 - > for the most part cautiously open AI might be opening, 620 - > I mean, might be the exception of the rule, and 621 - > that as long as we are open minded and collaborative, 622 - > and as we start getting closer and closer to these 623 - > these danger points, that the regulation comes top down in 624 - > a in a broad collaborative way, I think will be okay. 625 - > But I also genuinely worry about climping too hard on 626 - > regulation and losing the race to aid a g I, 627 - > you know, from a sort of a geopolitical perspective, I 628 - > don't want again, I don't want to over indexe on 629 - > that too much, but it. 630 - >

Speaker 3: Is I think a real concern. 631 - >

Speaker 5: My view on this is that governments are like so behind, 632 - > and many they have no clue what they're talking about. 633 - > And and it's it's casing point. AI could be used 634 - > greatly to filter out content on social media that is 635 - > just not true. 636 - >

Speaker 4: It's it's I think. 637 - >

Speaker 5: We we have the capability, but are we doing anything 638 - > about it? I don't think so. And this is influencing elections, 639 - > this is influencing societies, this is influencing the brain of 640 - > our kids. And governments are asleep at the will not 641 - > doing anything. And and tech companies are not really self 642 - > regulating the way they could. 643 - >

Speaker 4: It's there is just look at them social media. 644 - >

Speaker 5: So much crap out there that could be made better 645 - > with with AI. 646 - >

Speaker 4: And we are not. 647 - >

Speaker 5: We are instead focusing on how do we increase engagement, 648 - > which engagement means. 649 - >

Speaker 4: Lets let me just. 650 - >

Speaker 5: Push out the most ridiculous thing that I can and 651 - > that will bring me attention, that will create a lot 652 - > of feedback and whatnot. But that's where we are. And 653 - > that's one typical problem. And don't get me started if 654 - > we are going into defense, because defense in itself, like 655 - > all the look at the war in Ukraine. I bet 656 - > Talenteer is making a lot of money there with the 657 - > drones and everything, and that should be a lot ditriculated 658 - > in my opinion, at least having an intelligent conversation about it, 659 - > like having let's have a conversation about what. 660 - >

Speaker 4: The regulation would mean. 661 - >

Speaker 5: And I'm not seeing that that happening. And at least 662 - > the governments are kind of a slip at the will 663 - > in my opinion. 664 - >

Speaker 1: All right, let's end up in a positive note. Thank 665 - > you Jerry, bringing your Eastern European energy the conversation. But one, 666 - > what does a responsible AI powered business look like in 667 - > the near future to you? 668 - >

Speaker 6: Yeah, the the things that businesses can do is the 669 - > average business can't do anything with respect to AGI or 670 - > ASI or things like that. Right, those are those are 671 - > Those are models that can only be built by the 672 - > frontier labs who are putting literally tens of billions of 673 - > dollars into their development. Every day businesses, and I'm talking 674 - > about you know, even Fortune five hundred businesses. 675 - >

Speaker 3: The number one thing they can. 676 - >

Speaker 6: Do to be responsible from an AI perspective is look 677 - > at look at the data, and look at the bias. 678 - > And the example I always us is we've built systems 679 - > that do auto valuation, auto house valuation, and so we 680 - > had to go look at tax records and sales records, 681 - > going back data going back decades, and we found like 682 - > the most unbelievably clear racist like policies like embedded in 683 - > the data. And had we just trained that model on 684 - > that historic data, it would replicated those biases. It would 685 - > have been as racist is anybody you know from the 686 - > fifties or sixties. And here's the thing is, if we'd 687 - > have employed that model, a lot of people would have said, well. 688 - >

Speaker 3: You were turned down. You were turned down for that loan. 689 - >

Speaker 6: The AI did it, and the AI is unbiased and 690 - > it's super intelligent, and so it's objective. 691 - >

Speaker 3: No, that's not true at all. 692 - >

Speaker 6: You train the model to reflect the bias and the 693 - > historic data because you were lazy and you didn't take 694 - > the time to do the hard work to identify the 695 - > bias and remove it. And so the number one things 696 - > that businesses can do to be responsible is make sure 697 - > that they're thinking really really carefully about the data and 698 - > the modeling techniques and they are not perpetuating any type 699 - > of biases or any other negative And there's more that 700 - > we could go into negative traits like that, and then 701 - > throwing your hands up and acting like you've built some 702 - > omniscient objective system. 703 - >

Speaker 1: Nice Jerry, give me the final word on responsible AI 704 - > power businesses. 705 - >

Speaker 4: Mhm. 706 - >

Speaker 5: Remember what you said with garbage in garbage out. That 707 - > is so much what you said, Ron, about how you 708 - > you you put in bias into your everyday life, and 709 - > the AI. 710 - >

Speaker 4: Is it's I think. 711 - >

Speaker 5: Doing it responsibly and looking at not replicating our own 712 - > biases and understanding what you're doing. I think that's that's 713 - > that's something that we should do. Not sure if we 714 - > are doing it. I see, I see that there is 715 - > a there is a concern out there, and people are 716 - > thinking about this more thoughtfully. 717 - >

Speaker 4: Than in a couple of years ago. 718 - >

Speaker 5: But you gave an example chudgipiity Okay, three days ago, 719 - > I asked Judge a pity one simple question, and I 720 - > asked it three times. I got three different answers. I 721 - > don't know what they are fitting into, chad Zipti, but 722 - > sometimes I'm wondering and I'm scratching my head. I think 723 - > we should be more responsible in the way we set 724 - > it up, and we should pay a little bit more 725 - > attention to this. 726 - >

Speaker 4: So yeah, awesome, that's my. 727 - >

Speaker 2: Thing, Gerry. 728 - >

Speaker 1: All right, Ron, let's wrap things up by giving a 729 - > little plugski here to kung Fu AI. If people want 730 - > to find out more about you and your company, where 731 - > should they start? 732 - >

Speaker 6: And please check out kung Fu dot ais. That's our 733 - > web domain. And we've got a podcast called Hidden Layers. 734 - > It's a little bit more of a technical divell we'll 735 - > cover the latest developments in a pretty pretty deep technical 736 - > way every month, and check that out if. 737 - >

Speaker 3: You want to learn a little bit more about artificial intelligence. 738 - >

Speaker 2: Awesome, Ron, thanks so much of time, good luck. 739 - >

Speaker 3: Thank you so much. Guys appreciate it. 740 - >

Speaker 4: Thank you.