Breaking down the 2026 Stanford AI Index Report

1 00:00:01,760 - > 00:00:05,440 Narrator: Welcome to the Practical AI Podcast, where we 2 00:00:05,440 - > 00:00:08,160 break down the real world applications of artificial 3 00:00:08,160 - > 00:00:11,840 intelligence and how it's shaping the way we live, work, 4 00:00:11,840 - > 00:00:16,325 and create. Our goal is to help make AI technology practical, 5 00:00:16,325 - > 00:00:19,365 productive, and accessible to everyone. Whether you're a 6 00:00:19,365 - > 00:00:22,325 developer, business leader, or just curious about the tech 7 00:00:22,325 - > 00:00:25,685 behind the buzz, you're in the right place. Be sure to connect 8 00:00:25,685 - > 00:00:29,050 with us on LinkedIn, X, or Blue Sky to stay up to date with 9 00:00:29,050 - > 00:00:33,050 episode drops, behind the scenes content, and AI insights. You 10 00:00:33,050 - > 00:00:35,770 can learn more at practicalai.fm. 11 00:00:35,770 - > 00:00:37,450 Now onto the show. 12 00:00:41,530 - > 00:00:44,975 Dan: Well, welcome to another episode of the Practical AI 13 00:00:44,975 - > 00:00:50,095 Podcast. Today, it's just Chris and I, my my cohost and I, in 14 00:00:50,095 - > 00:00:54,255 what we call a fully connected episode where we try to keep you 15 00:00:54,255 - > 00:00:57,510 updated with some of the things that are happening in the AI 16 00:00:57,510 - > 00:01:03,110 news and maybe, share some practical practical information 17 00:01:03,110 - > 00:01:06,310 that'll help you level up your AI and machine learning game. 18 00:01:06,310 - > 00:01:10,550 I'm Daniel Whitenack. I'm CEO at Prediction Guard, and I'm joined 19 00:01:10,550 - > 00:01:14,925 as always by my cohost, Chris Benson, who is a principal AI 20 00:01:14,925 - > 00:01:17,885 and autonomy research engineer. How you doing, Chris? 21 00:01:17,885 - > 00:01:21,325 Chris: I'm doing good. I'm excited. This is we we're doing 22 00:01:21,485 - > 00:01:24,125 the episode we're doing today, we've done a number of times 23 00:01:24,125 - > 00:01:27,990 over the years. Stanford AI Index Report, we get to go 24 00:01:27,990 - > 00:01:29,750 through it. It's always fun. 25 00:01:30,870 - > 00:01:34,230 And kind of kind of level set kinda how things are changing, 26 00:01:34,230 - > 00:01:38,390 and gosh, I mean, things are changing so fast right now. 27 00:01:39,435 - > 00:01:40,235 Whew. 28 00:01:40,795 - > 00:01:46,235 Dan: And for context, so some of you may or may not have have 29 00:01:46,235 - > 00:01:49,195 listened to our previous episodes where Stanford 30 00:01:49,195 - > 00:01:55,630 Stanford's Human Centered Artificial Intelligence center 31 00:01:57,070 - > 00:02:01,310 institute. I forget, the exact of what they call themselves. 32 00:02:01,390 - > 00:02:04,510 But the human centered artificial intelligence effort 33 00:02:04,510 - > 00:02:08,195 there at Stanford, they published this AI Index report, 34 00:02:08,195 - > 00:02:10,595 and they've been doing it for a number of years. We've talked 35 00:02:10,595 - > 00:02:13,155 about it before. If you're interested, we're we're not 36 00:02:13,155 - > 00:02:15,795 gonna go into, like, how it was created. 37 00:02:15,795 - > 00:02:19,395 It's very rigorous. It's very data driven. You can go back and 38 00:02:19,395 - > 00:02:24,610 listen to episode two seventy six. We had some representatives 39 00:02:24,610 - > 00:02:29,010 on from Stanford that actually shared, you know, what it is, 40 00:02:29,010 - > 00:02:32,930 how it's created, and I'm sure that's updated somewhat over 41 00:02:32,930 - > 00:02:36,325 time, but that would be a great context for today. But there's a 42 00:02:36,325 - > 00:02:40,485 lot of takeaways here, Chris, and I think, you know, maybe 43 00:02:40,485 - > 00:02:41,685 we'll get through all of them. 44 00:02:41,685 - > 00:02:45,205 We can try rapid rapid fire here to talk through some of these 45 00:02:45,205 - > 00:02:49,365 and share them with the audience and see see maybe our reaction 46 00:02:49,365 - > 00:02:52,440 to some of these. Some of them were a surprise to me, to be 47 00:02:52,440 - > 00:02:53,160 honest, Chris. 48 00:02:53,160 - > 00:02:57,080 Chris: Yeah, there always are, because I mean, you kind of, it 49 00:02:57,240 - > 00:03:00,200 kind of brings you back after, you know, with the rigorous 50 00:03:00,200 - > 00:03:02,840 approach they have. We all have these perceptions. We're all 51 00:03:02,840 - > 00:03:07,035 watching the news and all the AI hot things that are out there, 52 00:03:07,435 - > 00:03:11,995 and there's times where it kinda level sets you a little bit, and 53 00:03:11,995 - > 00:03:16,875 then other times it kind of goes and I mean, just kicking us off 54 00:03:16,875 - > 00:03:20,780 on number one on their top takeaways list right off the 55 00:03:20,780 - > 00:03:23,580 bat. We kind of This is one of those places where we were going 56 00:03:23,580 - > 00:03:28,300 one way, and then it didn't take the report. We kind of realized 57 00:03:28,300 - > 00:03:31,305 that things were changing back, but for a while we were pretty 58 00:03:31,305 - > 00:03:36,345 convinced open source models were gonna completely catch up 59 00:03:36,345 - > 00:03:38,585 with plateau models because that's the trend that we were 60 00:03:38,585 - > 00:03:39,945 seeing for such a long time. 61 00:03:40,185 - > 00:03:43,225 We realized a little while back that that wasn't happening for a 62 00:03:43,225 - > 00:03:45,305 variety of reasons, which we've actually talked about on 63 00:03:45,305 - > 00:03:48,790 previous episodes, but the very first thing they mention is AI 64 00:03:48,790 - > 00:03:52,790 capability is not plateauing. It is accelerating and reaching 65 00:03:52,790 - > 00:03:59,665 more people than ever, and yeah, I think we're seeing that in 66 00:03:59,665 - > 00:04:00,865 2026. 67 00:04:00,865 - > 00:04:04,545 Dan: Yeah. So one one way one of the ways they express this is 68 00:04:04,545 - > 00:04:10,305 that over 90% of notable frontier models were produced in 69 00:04:10,320 - > 00:04:14,720 in 2025, and several of those now meet or exceed human 70 00:04:14,720 - > 00:04:18,160 baselines on a number of things, and they and they go into those 71 00:04:18,160 - > 00:04:22,695 things. Obviously, one of the things one of the hot takes that 72 00:04:22,695 - > 00:04:27,015 I'm always sharing, Chris, are these baseline or or these 73 00:04:27,015 - > 00:04:32,055 benchmarks, let's say, on PhD level science questions. They're 74 00:04:32,055 - > 00:04:36,050 very they're interesting, and I think they're some somehow 75 00:04:36,050 - > 00:04:39,730 representative of how we're advancing, but benchmarks in 76 00:04:39,730 - > 00:04:45,090 general are quite flawed. So even with that, you know, caveat 77 00:04:45,090 - > 00:04:48,210 in there, it does seem like there is advancement that's 78 00:04:48,210 - > 00:04:54,005 that's happening and, you know, a lot of that reaching or 79 00:04:54,005 - > 00:04:58,565 exceeding human level performance is is impressive and 80 00:04:58,565 - > 00:05:00,405 maybe scary for some people. 81 00:05:00,405 - > 00:05:01,525 I'm I'm not sure. But 82 00:05:01,765 - > 00:05:04,405 Chris: Yeah. I mean, we're we've gotten these frontier models, 83 00:05:04,405 - > 00:05:09,560 you know, in recent months that are are just so capable, 84 00:05:09,560 - > 00:05:12,840 especially when combined with these new agentic systems that 85 00:05:12,840 - > 00:05:18,440 everyone's been obviously huge topic this year, and able to 86 00:05:18,440 - > 00:05:22,685 productively do a lot of stuff, which creating a lot of upheaval 87 00:05:22,685 - > 00:05:25,485 in the job markets and how different companies are 88 00:05:25,485 - > 00:05:29,405 perceiving that. Yeah. But, yeah, I mean, it's it's a new 89 00:05:29,645 - > 00:05:32,845 it's very different from a year ago today, I would say. Yeah. 90 00:05:32,845 - > 00:05:34,125 If you if you look back. 91 00:05:34,525 - > 00:05:38,950 Dan: I I think, like, they talk about the majority, four out of 92 00:05:38,950 - > 00:05:43,030 five university students using Gen AI. And I was actually 93 00:05:43,030 - > 00:05:47,190 thinking Chris, as like a gauge on this. I was thinking back to 94 00:05:47,190 - > 00:05:53,795 my own PhD, which is about five years long. Like how long would 95 00:05:53,795 - > 00:05:57,875 that amount of work taken me with the tools that are 96 00:05:57,875 - > 00:06:02,595 available now? Both research wise, there was a coding element 97 00:06:02,595 - > 00:06:03,155 to it. 98 00:06:03,155 - > 00:06:08,520 There was a writing element to it, obviously. And I think like 99 00:06:08,520 - > 00:06:12,440 the amount of work I did was a lot, but I think with the tools 100 00:06:12,440 - > 00:06:16,840 now, it's got to at least cut that down by half. I would I 101 00:06:16,840 - > 00:06:20,125 would assume. And I don't know, like, obviously, universities 102 00:06:20,365 - > 00:06:24,285 are wrestling with this and how to deal with it. And, you know, 103 00:06:24,285 - > 00:06:27,565 maybe people are just getting more done in their PhDs now, 104 00:06:27,565 - > 00:06:29,565 which would be which would probably be good. 105 00:06:29,565 - > 00:06:32,125 Chris: But And and and I know we're already blowing through. 106 00:06:32,125 - > 00:06:34,610 We had we only have we have We had a bunch of items to get 107 00:06:34,610 - > 00:06:36,290 through, probably not gonna get through them all, but I will 108 00:06:36,290 - > 00:06:39,650 note that I think that also translates into the workplace. I 109 00:06:39,650 - > 00:06:44,770 know in my own job, I am much more productive with the tooling 110 00:06:45,455 - > 00:06:50,095 cape you know, that we are all using here, and so the notion of 111 00:06:50,095 - > 00:06:53,215 what what not too far back, it would have been like a research 112 00:06:53,215 - > 00:06:55,215 project, and you would have been trying to think about all the 113 00:06:55,215 - > 00:06:58,840 things you gotta do for that, and now that's like, it's like, 114 00:06:58,840 - > 00:07:01,000 I'm gonna dive into it, and by the end of the week, I'm gonna 115 00:07:01,000 - > 00:07:04,440 have this thing done. That would have been a large body of work 116 00:07:04,440 - > 00:07:08,680 prior to that, and so, yeah. I mean, it's definitely changing 117 00:07:08,680 - > 00:07:11,240 jobs, which is another thing that I think we're gonna get to 118 00:07:11,240 - > 00:07:11,800 down the road. 119 00:07:12,885 - > 00:07:16,965 Number two is AI model performance between The United 120 00:07:16,965 - > 00:07:21,845 States and China has closed. It's effectively closed, and 121 00:07:21,845 - > 00:07:25,685 we're looking at two it no longer kind of a leader follower 122 00:07:26,180 - > 00:07:29,860 effect, but two co leaders in the world market. 123 00:07:30,340 - > 00:07:34,740 Dan: Yeah. Yeah. And actually, this was one of the ones well, I 124 00:07:34,740 - > 00:07:37,300 don't know. Maybe there there's different elements of this, 125 00:07:37,300 - > 00:07:41,300 Chris. I would say just in my own practical experience on the, 126 00:07:42,075 - > 00:07:48,235 you know, people use and govern, you know, both closed and open 127 00:07:48,235 - > 00:07:51,915 models in in our platform now, so I get exposed to kind of both 128 00:07:51,915 - > 00:07:52,875 of those. 129 00:07:53,035 - > 00:07:56,990 I would say on the open model side, just practically, China 130 00:07:56,990 - > 00:08:01,150 seems to clearly have the lead. Now maybe that's that's 131 00:08:01,150 - > 00:08:06,430 different on the, I guess, on the closed model side. If so 132 00:08:06,430 - > 00:08:09,630 maybe there is a little bit of nuance there where, 133 00:08:11,015 - > 00:08:11,495 Chris: Fair. 134 00:08:11,495 - > 00:08:16,375 Dan: Maybe in my mind, had this perception of of Chinese models 135 00:08:16,375 - > 00:08:21,015 being almost superior, but that doesn't really factor in the 136 00:08:21,015 - > 00:08:24,230 frontier model, closed model provider side of that. 137 00:08:24,230 - > 00:08:28,070 Chris: I mean, that's true. We talked about that recently on an 138 00:08:28,070 - > 00:08:33,590 episode just a few weeks ago, and the notion of, in a lot of 139 00:08:33,590 - > 00:08:37,865 ways, The US has kind of walked away from open models to some 140 00:08:37,865 - > 00:08:42,585 degree, you know, Meta you know, with Meta walking away, they're 141 00:08:42,585 - > 00:08:45,065 they're now going entirely closed, and that was kinda 142 00:08:45,065 - > 00:08:48,585 leading The US contingent. It's not to say that there aren't 143 00:08:48,585 - > 00:08:52,340 smaller, but in that top tier, we've kinda walked away. Whereas 144 00:08:52,340 - > 00:08:58,180 China has largely embraced the open model approach, which I 145 00:08:58,180 - > 00:09:01,380 think the fallout from that in the West will be interesting in 146 00:09:01,380 - > 00:09:03,780 terms of how much, you know, there's a certain amount of 147 00:09:03,780 - > 00:09:10,005 geopolitical division across, you know, between the West and 148 00:09:10,005 - > 00:09:15,125 the East in terms of of how they adopt models and what models are 149 00:09:15,125 - > 00:09:19,365 okay to use in different contexts. And so as we see open 150 00:09:19,365 - > 00:09:23,770 models predominantly, in the large scale happening in the 151 00:09:23,770 - > 00:09:28,330 East, closed models only in the West, how that ends up shuffling 152 00:09:28,330 - > 00:09:31,370 things will be, an interesting thing to watch in the months and 153 00:09:31,370 - > 00:09:32,010 years ahead. 154 00:09:32,010 - > 00:09:35,130 Dan: Yeah. I'm connecting a little bit more of that dynamic 155 00:09:35,555 - > 00:09:40,915 kind of US and abroad dynamic. Takeaway number three on the AI 156 00:09:40,915 - > 00:09:46,595 Index report was that The United States hosts the most AI data 157 00:09:46,595 - > 00:09:51,170 centers, but the majority of their chips are fab fabricated 158 00:09:51,170 - > 00:09:58,370 by a single Taiwanese foundry. So, it this one, actually, Chris 159 00:09:58,610 - > 00:10:03,730 so not the chip fabrication piece. I I knew that piece, but, 160 00:10:03,810 - > 00:10:07,305 you know, you always hear about China just spinning up data 161 00:10:07,305 - > 00:10:08,745 centers everywhere. 162 00:10:08,745 - > 00:10:14,105 So it was interesting to me to hear that the The US still hosts 163 00:10:14,105 - > 00:10:19,290 the most AI data centers because obviously, you know, if you're 164 00:10:19,290 - > 00:10:22,730 putting in a data center in some town in The United States, you 165 00:10:22,730 - > 00:10:27,930 could get, the local city council against you and people 166 00:10:27,930 - > 00:10:32,405 are up in arms and there's more hoops to jump through. Whereas 167 00:10:32,405 - > 00:10:38,165 in China, you know, a lot of a lot of that can just happen. I 168 00:10:38,165 - > 00:10:41,765 And there's a huge amount of investment. So, yeah, this one 169 00:10:41,765 - > 00:10:45,130 was actually pretty interesting to me to hear that current 170 00:10:45,130 - > 00:10:45,530 state. 171 00:10:45,530 - > 00:10:48,730 Chris: I'm it'll be interesting, like, when this same report 172 00:10:48,730 - > 00:10:52,730 comes out next year, you know, year by year to track that, and 173 00:10:52,730 - > 00:10:56,330 see if our if the if the The United States, kind of that 10 174 00:10:56,330 - > 00:11:00,815 time number that's in here shrinks, maybe even shrinks very 175 00:11:00,815 - > 00:11:04,895 rapidly. And so we'll see. We'll see what happens on there. 176 00:11:04,895 - > 00:11:10,335 Dan: Yeah. Yeah. For sure. So takeaway number four, we're 177 00:11:10,335 - > 00:11:14,980 gradually working our way through here, Chris. AI models 178 00:11:14,980 - > 00:11:18,900 can win a gold medal at the International Mathematical 179 00:11:18,980 - > 00:11:23,300 Olympiad but cannot reliably tell time, an example of what 180 00:11:23,300 - > 00:11:27,860 researchers call the, quote, jagged frontier of AI. 181 00:11:28,315 - > 00:11:32,395 Interesting. It it almost seems like I remember we had a a 182 00:11:32,395 - > 00:11:35,755 guest. I'm pretty sure I I don't know. I remember talking to 183 00:11:35,755 - > 00:11:37,595 them. I think they were on the show. 184 00:11:37,595 - > 00:11:40,875 Sometimes I forget what was on the show and what conversations 185 00:11:40,875 - > 00:11:43,450 I had in real life, Chris. I don't know how do that same. 186 00:11:43,450 - > 00:11:45,050 People. Yeah. We do. 187 00:11:45,050 - > 00:11:48,090 But at some point, there was a conversation that happened with, 188 00:11:48,890 - > 00:11:53,050 someone from the Allen Institute, for AI, and a lot of 189 00:11:53,050 - > 00:11:56,215 what they were looking at for some time was around, like, 190 00:11:56,215 - > 00:12:00,135 common sense. So AI models can do a lot of really impressive 191 00:12:00,135 - > 00:12:03,015 things, but when it comes to common sense, it's like they 192 00:12:03,015 - > 00:12:09,030 fall fall over because there's no actual connection to the real 193 00:12:09,030 - > 00:12:12,150 world. Right? They're producing tokens or they're, you know, 194 00:12:12,150 - > 00:12:17,430 producing, know, tokens based on probabilities of what they've 195 00:12:17,430 - > 00:12:21,595 seen before. And so there's these many seemingly seeming 196 00:12:21,595 - > 00:12:24,475 coherence and impressive things that happen, and then all of a 197 00:12:24,475 - > 00:12:29,755 sudden, they can't do the most simple thing that involves some 198 00:12:29,755 - > 00:12:33,195 connection to the real world, like here, you know, telling 199 00:12:33,195 - > 00:12:34,235 time, for example. 200 00:12:34,235 - > 00:12:37,740 Chris: Yeah. The example they talk about is Gemini DeepThink 201 00:12:37,740 - > 00:12:41,500 getting the gold medal at the IMO, but only being able to read 202 00:12:41,500 - > 00:12:47,500 an analog clock 50.1 of the time in terms of accurately. And and 203 00:12:47,500 - > 00:12:50,915 and they offer some other stats along the way. And I think, you 204 00:12:50,915 - > 00:12:53,395 know, this goes back to another topic that we've talked about a 205 00:12:53,395 - > 00:12:56,275 number of times, and that is we're still talking about 206 00:12:56,275 - > 00:13:00,595 language, you know, being what these models are are trained on, 207 00:13:00,915 - > 00:13:03,875 and we've we've talked you know, it's becoming increasingly 208 00:13:03,875 - > 00:13:08,620 popular to talk about the notion of what a genuine what you know, 209 00:13:08,620 - > 00:13:11,740 it's there's several names it goes by, but a world model, you 210 00:13:11,740 - > 00:13:14,860 know, something that actually has context for all the things 211 00:13:14,860 - > 00:13:19,495 in life because, you know, you don't have that with with our 212 00:13:19,495 - > 00:13:22,695 existing frontier models that are, you know, based on on 213 00:13:22,695 - > 00:13:26,935 training on language. So there's a lot of research. 214 00:13:27,015 - > 00:13:30,775 Know, you know, that's famously Jan Lecun, one of the the 215 00:13:30,775 - > 00:13:34,040 godfathers of AI, has talked about that many times over the 216 00:13:34,040 - > 00:13:38,440 last few years, the need to move past LLMs and have world models 217 00:13:38,440 - > 00:13:43,400 that actually have context. So and increasingly, especially, I 218 00:13:43,400 - > 00:13:46,360 know as I as my world over the last few years has gotten more 219 00:13:46,360 - > 00:13:51,025 and more focused on on edge cases and autonomy, the notion 220 00:13:51,025 - > 00:13:55,025 of world models and how they would impact our field has 221 00:13:55,025 - > 00:13:57,585 become increasingly important. So it'll it'll be interesting as 222 00:13:57,585 - > 00:14:01,665 well to see how how this measures up, you know, next year 223 00:14:01,665 - > 00:14:05,180 in the same report as we get to that point on, on those kinds of 224 00:14:05,180 - > 00:14:05,740 upgrades. 225 00:14:05,740 - > 00:14:10,700 Dan: Maybe a bit of an opinion here, Chris, is I, I almost 226 00:14:10,700 - > 00:14:16,915 think we're not being fair in some sense to the, 227 00:14:17,235 - > 00:14:17,635 Chris: I don't 228 00:14:17,635 - > 00:14:21,475 Dan: know if it's to the models or to the, the way people do AI 229 00:14:21,475 - > 00:14:25,635 because you, you know, like Claude, for example, like an 230 00:14:25,635 - > 00:14:29,670 anthropic model knows nothing about the tickets in my ClickUp 231 00:14:29,670 - > 00:14:33,110 platform. Right? And so you could call that model dumbaw. It 232 00:14:33,110 - > 00:14:37,030 doesn't know, like, blah blah blah. But I can perfectly well 233 00:14:37,030 - > 00:14:41,990 just connect Claude via Claude code skill to my ClickUp, and 234 00:14:41,990 - > 00:14:44,885 all of a sudden, now I have all that context about what I should 235 00:14:44,885 - > 00:14:49,285 be working on this sprint and all that context is there and it 236 00:14:49,285 - > 00:14:52,085 knows about my PRs and all of this stuff. 237 00:14:52,165 - > 00:14:56,740 Right? So it's kind of like what we talked about with, when we 238 00:14:56,740 - > 00:15:03,220 were talking about Hermes agent where the no one would expect a 239 00:15:03,220 - > 00:15:08,900 a brain absent a body to be sort of take useful action in the 240 00:15:08,900 - > 00:15:14,025 world. And so that agent harness around which we surround the 241 00:15:14,025 - > 00:15:17,945 model is really part of that connection. So it could be that 242 00:15:17,945 - > 00:15:21,145 like there are these world models and such that are 243 00:15:21,145 - > 00:15:25,465 relevant and that seems like good research, but also like 244 00:15:24,750 - > 00:15:28,110 part of this is that these models need a body. They need a 245 00:15:28,110 - > 00:15:29,310 harness around them. 246 00:15:29,310 - > 00:15:29,710 Right? 247 00:15:29,710 - > 00:15:32,030 Chris: We've and we've we've addressed that a whole bunch of 248 00:15:32,030 - > 00:15:36,270 times on the show and and the the you you can't a model a 249 00:15:36,270 - > 00:15:40,355 model in isolation doesn't do you a whole lot of good. You've 250 00:15:40,355 - > 00:15:44,035 gotta have that connection with the world, and I think as things 251 00:15:44,035 - > 00:15:48,915 evolve, we will I personally, my own personal belief is that 252 00:15:48,915 - > 00:15:52,675 world model development requires the same. You have to have 253 00:15:52,675 - > 00:15:56,250 feedback from the real world to be able to incorporate that into 254 00:15:56,250 - > 00:15:59,930 training to actually get you what you're looking for. And as 255 00:15:59,930 - > 00:16:02,570 inspiration for that, if we look at our own human brains, we're 256 00:16:02,570 - > 00:16:06,490 born, we're babies, we don't you know, we have these amazing baby 257 00:16:06,490 - > 00:16:09,135 brains, but they they haven't gotten a lot of experience 258 00:16:09,135 - > 00:16:12,335 against the real world. And it is all those feedback loops in 259 00:16:12,335 - > 00:16:15,535 those first two decades of life that kinda get us to, to 260 00:16:15,535 - > 00:16:16,175 functioning. 261 00:16:16,175 - > 00:16:19,535 So, there'll be some sort of analog presumably to that 262 00:16:19,535 - > 00:16:23,295 notion, in the world model development world as we go 263 00:16:23,295 - > 00:16:23,775 forward. So 264 00:16:24,030 - > 00:16:27,710 Dan: speaking of AI in the physical world, takeaway number 265 00:16:27,710 - > 00:16:33,470 five, robots still fail at most household tasks even as they 266 00:16:33,470 - > 00:16:36,750 excel in controlled environments. Do you have any 267 00:16:36,750 - > 00:16:38,270 any robots in your house, Chris? 268 00:16:38,295 - > 00:16:41,895 Chris: Well, only the ones that most people have. We have the 269 00:16:41,895 - > 00:16:46,455 vacuum going around and such, but, you know, I keep seeing you 270 00:16:46,455 - > 00:16:51,175 know, especially this is a much bigger thing in China than it is 271 00:16:51,175 - > 00:16:55,250 here with with robots in production in a lot of 272 00:16:55,250 - > 00:16:59,410 households. I saw something just a few days ago about a humanoid 273 00:16:59,410 - > 00:17:03,410 robot that, you know, is doing elderly care and washes the 274 00:17:03,410 - > 00:17:06,975 dishes and things like that, and that was exactly what I was 275 00:17:06,975 - > 00:17:11,455 wondering is like if you took that out of of their version of 276 00:17:11,455 - > 00:17:18,270 CES and and kind of explore it, I I wonder when you have an kind 277 00:17:18,270 - > 00:17:20,830 of a somewhat unique environment, you know, as 278 00:17:20,830 - > 00:17:23,950 different configurations of households and what duties, how 279 00:17:23,950 - > 00:17:27,070 well that really performs. And I honestly, I I don't know. I 280 00:17:27,070 - > 00:17:30,350 haven't had direct exposure to robots of that that nature. 281 00:17:30,350 - > 00:17:33,725 Dan: Yeah. Yeah. I Petal Petal Woman. 282 00:17:33,885 - > 00:17:34,685 Chris: I'm ready. 283 00:17:34,685 - > 00:17:38,285 Dan: You're ready? I'm ready. I guess controlled environments 284 00:17:38,285 - > 00:17:43,965 here, they're referring to maybe manufacturing facilities or, 285 00:17:44,525 - > 00:17:48,760 they also mention kind of software based simulations. 286 00:17:49,160 - > 00:17:53,320 Right? And, and yeah, that that that will be an interesting one 287 00:17:53,320 - > 00:17:55,000 to follow for for sure. 288 00:17:55,000 - > 00:17:57,560 Chris: I'm I'm just curious that when we when I get my first 289 00:17:57,560 - > 00:18:02,335 humanoid robot and we assign it the tasks around the house and I 290 00:18:02,335 - > 00:18:05,135 have five dogs running around the house, and the dog starts 291 00:18:05,135 - > 00:18:08,575 jumping up on it to play. And it may be the puppy at first, but 292 00:18:08,575 - > 00:18:11,375 then the the big dog that weighs, you know, 80 pounds 293 00:18:11,375 - > 00:18:15,110 jumps up on it. It will be interesting to see if it can 294 00:18:15,110 - > 00:18:16,950 survive the chaos of the Benson household. 295 00:18:16,950 - > 00:18:21,510 Dan: Listen, one of the I forget the so we had a friend in the 296 00:18:21,510 - > 00:18:26,710 restaurant industry, and he gave my wife and I tickets to I 297 00:18:26,710 - > 00:18:29,990 forget the name of the show. It's like the Chicago restaurant 298 00:18:30,125 - > 00:18:33,325 convention or something. It's basically, you could go there. 299 00:18:33,325 - > 00:18:36,845 There's a bunch of vendors that sell products into restaurants. 300 00:18:36,845 - > 00:18:40,205 So everything from, like, utensils to appliances to 301 00:18:40,205 - > 00:18:42,045 software to whatever, you know. 302 00:18:42,690 - > 00:18:46,290 And, so big networks of restaurants go there and look at 303 00:18:46,290 - > 00:18:49,810 things and they had all the robots there in, you know, 304 00:18:49,810 - > 00:18:52,930 different sections that did different things. And I have to 305 00:18:52,930 - > 00:18:57,730 say, you know, not having exposure to that world a ton, 306 00:18:58,175 - > 00:19:04,175 but kind of expecting way more than I saw. I I guess I I was 307 00:19:04,175 - > 00:19:08,255 fairly disappointed. Like, most of the robots that did, like, 308 00:19:08,255 - > 00:19:13,760 cooking, for example, it seemed to just be, you know, like, you 309 00:19:14,000 - > 00:19:18,800 the the washer dryer that spends and, you know, it spends around 310 00:19:18,880 - > 00:19:22,320 it was basically like a drum like that that heated up and 311 00:19:22,320 - > 00:19:25,520 you'd it just sort of dropped ingredients in the drum and it 312 00:19:25,520 - > 00:19:29,135 spun around and kinda stir fried them or something like that. 313 00:19:29,135 - > 00:19:32,735 It's like, I I was I was disappointed that I didn't see 314 00:19:32,735 - > 00:19:36,895 any, like, humanoid robots, like, chopping up, you know, 315 00:19:36,895 - > 00:19:38,975 making sushi or or something. 316 00:19:39,295 - > 00:19:44,630 Very far from, yeah. The the heated washing machine was very 317 00:19:44,630 - > 00:19:46,550 far from what I expected to see. 318 00:19:46,550 - > 00:19:50,150 Chris: What you envisioned. I I think and and while we're 319 00:19:50,150 - > 00:19:53,670 obviously questioning, you know, how how real the performance 320 00:19:53,670 - > 00:19:58,235 capabilities are, I do, at least my personal belief is that I 321 00:19:58,235 - > 00:20:02,715 think you're gonna find, a a a certain leveling up in China 322 00:20:02,715 - > 00:20:05,835 above what we've done in The US and probably, a fairly 323 00:20:05,835 - > 00:20:08,460 substantial one at that. So Is that 324 00:20:08,460 - > 00:20:11,980 Dan: because of safety restrictions in The US, or is 325 00:20:12,220 - > 00:20:15,820 Chris: I'm I'm not actually sure. Okay. I I think I think 326 00:20:15,820 - > 00:20:21,500 that my sense is that robots have just been a higher priority 327 00:20:21,500 - > 00:20:26,005 for quite a long time, drones and robots, obviously, and, you 328 00:20:26,005 - > 00:20:29,605 know, we go back to the turn of the millennium, like looking way 329 00:20:29,605 - > 00:20:32,725 back, or we're looking at old school drones going up and 330 00:20:32,725 - > 00:20:37,820 selling you know, performances way back before we were really 331 00:20:37,820 - > 00:20:40,860 even thinking about such things at all here. I think that if 332 00:20:40,860 - > 00:20:43,980 you've been doing something for a long time and it's more of an 333 00:20:43,980 - > 00:20:48,540 evolutionary step each way, whereas for us, we are we are 334 00:20:48,540 - > 00:20:52,285 surging here in The US, but I think we're coming from behind 335 00:20:52,285 - > 00:20:56,045 on the experience side of that. So, it'll be interesting to see 336 00:20:56,045 - > 00:20:57,485 how that plays out over time. 337 00:20:57,485 - > 00:21:00,765 Dan: Yeah. Make make sense. Well, something that is playing 338 00:21:00,765 - > 00:21:05,390 out over time, take away number six, responsible AI is not 339 00:21:05,390 - > 00:21:09,470 keeping pace with AI cap capability with safety 340 00:21:09,470 - > 00:21:14,110 benchmarks lagging and incidents rising sharply. Chris, 341 00:21:14,430 - > 00:21:19,055 obviously, this one, hits close to home for me. This is part of 342 00:21:19,055 - > 00:21:22,335 what we're hopefully helping people deal with, but it also 343 00:21:22,335 - > 00:21:27,935 reminded me of, you know, of course, incidents and other 344 00:21:27,935 - > 00:21:30,975 things that we've seen from my work with Prediction Guard, but 345 00:21:30,975 - > 00:21:35,260 also a show that we had actually somewhat recently. 346 00:21:35,260 - > 00:21:40,300 So back in February, episode three forty six, which was AI 347 00:21:40,300 - > 00:21:43,740 incidence audits and the limits of benchmarks, which basically 348 00:21:43,740 - > 00:21:48,515 hits directly at this. We had Sean MacGregor on that show, and 349 00:21:48,515 - > 00:21:53,395 and he was talking about the AI incident database that he was 350 00:21:53,395 - > 00:21:57,475 helping helping create and and manage. And, yeah, that it was 351 00:21:57,475 - > 00:22:03,390 very interesting just the the diversity of AI incidents that 352 00:22:03,390 - > 00:22:09,230 we're seeing now and the sharp rise in those. And this is also, 353 00:22:09,310 - > 00:22:13,445 you know, only documented AI incident, you know, cases. 354 00:22:13,445 - > 00:22:16,725 There's many things that are happening just anecdotally that 355 00:22:16,725 - > 00:22:21,045 I see that I'm sure aren't being documented in that AI incident 356 00:22:21,045 - > 00:22:22,005 database. 357 00:22:22,165 - > 00:22:24,885 Chris: Yeah. You know, I'm gonna I'm gonna make a comment that 358 00:22:24,885 - > 00:22:30,020 probably will surprise most people tuning in, and that is, 359 00:22:31,220 - > 00:22:34,100 first of all, making it very clear I only speak for myself 360 00:22:34,100 - > 00:22:38,100 and not for my employer or any other organization. I actually 361 00:22:38,100 - > 00:22:40,980 think people would be surprised in the defense industry that 362 00:22:40,980 - > 00:22:45,175 there is probably more guardrails and responsible AI 363 00:22:45,175 - > 00:22:49,335 efforts around our industry than most commercial industries. 364 00:22:49,575 - > 00:22:53,175 There are federal regulations here in The US that prohibit 365 00:22:54,055 - > 00:22:56,920 certain things that in the commercial space people might 366 00:22:56,920 - > 00:23:02,280 just surge and go do in terms of safety issues. And so as I was 367 00:23:02,280 - > 00:23:05,160 reading that earlier before the show, I was thinking, you know, 368 00:23:05,160 - > 00:23:08,680 I'm actually in an industry that it may hold us back at times 369 00:23:08,680 - > 00:23:12,685 because we're not surging the way the commercial industries 370 00:23:12,685 - > 00:23:17,725 have the freedom to, but I think it also there's a there's quite 371 00:23:17,725 - > 00:23:24,210 an intense focus on keeping things that need to be safe, and 372 00:23:24,210 - > 00:23:25,250 I have noticed that. 373 00:23:25,250 - > 00:23:28,050 And sometimes, as someone who is always enthusiastic on new 374 00:23:28,050 - > 00:23:31,490 technologies, I get a little bit frustrated, but then I stop and 375 00:23:31,490 - > 00:23:35,650 go, no, I'm glad we're that way. So I just thought I'd mention 376 00:23:35,650 - > 00:23:36,130 that. 377 00:23:36,850 - > 00:23:39,795 Dan: Yeah. I think it's encouraging and a good 378 00:23:40,195 - > 00:23:43,955 inspiration for the rest of us to consider those, those 379 00:23:44,115 - > 00:23:48,755 responsible AI governance enforcement, you know, policy, 380 00:23:48,755 - > 00:23:53,170 whatever, however, what, what whatever form that takes in your 381 00:23:53,170 - > 00:23:56,530 level of maturity as an organization, I think there are 382 00:23:56,690 - > 00:24:02,530 there are those out there that are pushing, that direction. And 383 00:24:01,905 - > 00:24:05,825 we've seen development even over this last year where people I 384 00:24:05,825 - > 00:24:11,265 think are moving beyond the sort of trust me phase of of AI 385 00:24:11,265 - > 00:24:18,240 governance towards, you know, exportable proof and, even 386 00:24:18,240 - > 00:24:22,320 certification types of things. And I think we'll see, you know, 387 00:24:22,320 - > 00:24:24,720 one of my predictions this coming year, I think is we're 388 00:24:24,720 - > 00:24:30,015 gonna start to see some of that needed exportable proof being 389 00:24:30,015 - > 00:24:34,495 part of even audits and certifications, whether that's 390 00:24:34,495 - > 00:24:38,575 SOC two or AI specific types of certifications for companies. 391 00:24:38,575 - > 00:24:41,215 Chris: I I not only agree with that, but I think that the 392 00:24:41,215 - > 00:24:45,170 marketplace will demand it as we go forward. I think I think we 393 00:24:45,170 - > 00:24:48,770 will continue to have some big news events where things are 394 00:24:48,770 - > 00:24:52,930 going off the wheels from various organizations and 395 00:24:52,930 - > 00:24:56,175 industries at different places, probably a variety of them, and 396 00:24:56,175 - > 00:24:59,375 there's gonna be a point where where people say, we need we 397 00:24:59,375 - > 00:25:02,415 need to know that there are safety measures in place before 398 00:25:02,415 - > 00:25:05,695 we're willing to deploy it within our organization. And so 399 00:25:05,695 - > 00:25:08,540 I I think the market will will demand that going forward. 400 00:25:11,100 - > 00:25:13,260 Sponsor: If you've been listening to the show over the 401 00:25:13,260 - > 00:25:17,580 past few months, you realize just how transformative AgenTic 402 00:25:17,580 - > 00:25:22,220 AI is, whether that's Claude Code or Hermes Agent or custom 403 00:25:22,220 - > 00:25:25,345 built software that you're deploying for operational 404 00:25:25,345 - > 00:25:29,505 efficiencies or as new products to your customers. Regardless of 405 00:25:29,505 - > 00:25:33,345 your maturity now, this is the world that we're headed towards, 406 00:25:33,345 - > 00:25:37,400 this agentic AI world. And there's a lot of security and 407 00:25:37,400 - > 00:25:40,200 governance teams that aren't letting these agents go into 408 00:25:40,200 - > 00:25:45,640 production because of risks related to agency and autonomy 409 00:25:45,640 - > 00:25:50,440 and how do you take care of things like prompt injections or 410 00:25:49,815 - > 00:25:53,415 insecure tool usage. There's a lot to take care of, and that's 411 00:25:53,415 - > 00:25:57,895 why I'm personally spending my time outside of the show working 412 00:25:57,895 - > 00:26:01,655 with an amazing team of AI engineers to build 413 00:26:01,590 - > 00:26:04,630 PredictionGuard. PredictionGuard is an AI control plane that you 414 00:26:04,630 - > 00:26:08,390 run-in your own infrastructure behind your firewall. 415 00:26:08,550 - > 00:26:11,670 Developers can build on top of this control plane using 416 00:26:11,670 - > 00:26:14,790 everything that they wanna use, OpenAI and Anthropic compatible 417 00:26:14,790 - > 00:26:19,685 APIs, MCP servers, frameworks like LangChain, but all of this 418 00:26:19,685 - > 00:26:24,245 is plugged into a built in governance harness that enforces 419 00:26:24,405 - > 00:26:28,965 your organization's AI policies and all of that telemetry goes 420 00:26:28,965 - > 00:26:32,040 back to your monitoring and alerting systems. I'd encourage 421 00:26:32,040 - > 00:26:35,160 you to check out what we're doing at prediction guard dot 422 00:26:35,160 - > 00:26:39,480 com slash practical AI. You can schedule a demo with me and the 423 00:26:39,480 - > 00:26:42,760 team, and I'd love to get your feedback on what we're doing. So 424 00:26:42,760 - > 00:26:47,185 visit us at predictionguard.com/practicalai. 425 00:26:47,185 - > 00:26:51,505 That's predictionguard.com/practicalai. 426 00:26:54,145 - > 00:27:00,320 Dan: Well, Chris, we are almost half ish way through, the the 427 00:27:00,320 - > 00:27:04,560 the takeaways. We might not get to all of them in detail, but 428 00:27:04,560 - > 00:27:08,960 number seven, United States leads in AI investment, but its 429 00:27:08,960 - > 00:27:13,745 ability to attract global talent is declining. Interesting. 430 00:27:13,745 - > 00:27:19,105 Chris: Yes. It is interesting. It's, I think we are and I I'm 431 00:27:19,105 - > 00:27:23,105 not I'm not terribly surprised by that one either, as we have 432 00:27:23,105 - > 00:27:30,830 seen diversification, and also, frankly as political priorities 433 00:27:30,830 - > 00:27:34,430 in The United States have changed. I'll leave that to 434 00:27:34,830 - > 00:27:38,510 folks tuning in to decide kind of how they're looking at it, 435 00:27:38,510 - > 00:27:42,755 but I think that that has also impacted that. If you're it's 436 00:27:42,755 - > 00:27:45,235 certainly hard to attract if you're talking about global 437 00:27:45,235 - > 00:27:49,715 talent, it's hard to attract global talent if things that are 438 00:27:49,715 - > 00:27:52,995 ancillary to that, things like immigration, are challenge. 439 00:27:53,715 - > 00:27:57,430 So I don't, for me at least, given the current circumstances, 440 00:27:57,430 - > 00:28:02,870 that wasn't a surprise to run across that. I'm hoping that we 441 00:28:02,870 - > 00:28:04,790 don't lose our edge in that capacity. 442 00:28:04,870 - > 00:28:08,545 Dan: Yeah. Number that shocked me in this report was, I guess, 443 00:28:08,545 - > 00:28:13,025 the scale of that. So apparently, there's been an 80% 444 00:28:13,025 - > 00:28:17,985 decline just in the last year in terms of the number of AI 445 00:28:17,985 - > 00:28:23,430 researchers and developers moving to The US, which, yeah. 446 00:28:23,430 - > 00:28:25,270 Yeah. This is made right there. 447 00:28:25,510 - > 00:28:32,390 Is pretty, astounding. Yeah. Yeah. So it's, you you reap what 448 00:28:32,390 - > 00:28:35,590 Chris: you sow. I said we'll see how things change in the in the 449 00:28:35,590 - > 00:28:40,145 years ahead. But, yeah, interesting interesting point. 450 00:28:40,145 - > 00:28:45,265 I'm I'm curious with any thoughts on kind of future of AI 451 00:28:45,265 - > 00:28:47,505 investment? You know, we're we're leading right now. 452 00:28:47,505 - > 00:28:50,705 Do you have any any positions on your side on where you think 453 00:28:50,705 - > 00:28:51,505 things will go? 454 00:28:51,505 - > 00:28:58,360 Dan: Yeah. I mean, I I think the The US still kind of corners the 455 00:28:58,360 - > 00:29:07,005 market on VC driven startup startups that are funded by that 456 00:29:07,005 - > 00:29:11,325 world and, you know, Silicon Valley still holds a special 457 00:29:11,645 - > 00:29:16,125 place there in terms of the VC funds that are there, you know, 458 00:29:16,125 - > 00:29:19,725 other places to New York and kind of growing markets in the 459 00:29:19,725 - > 00:29:24,060 Midwest and that sort of thing. I think that's gonna be that's 460 00:29:24,060 - > 00:29:29,100 gonna hold true. The report also shows still the greatest number 461 00:29:29,100 - > 00:29:34,335 of companies that are being started are starting in AI 462 00:29:34,335 - > 00:29:39,455 companies or starting in The US and maybe funded by US VCs. But 463 00:29:39,455 - > 00:29:42,095 now in this world, I think a couple of things are true. 464 00:29:42,095 - > 00:29:46,815 It it's becoming more and more possible to maintain smaller 465 00:29:46,815 - > 00:29:51,590 teams and get a lot done. And so there's less people needed to 466 00:29:51,590 - > 00:29:55,350 build a company that's even doing tens or hundreds of 467 00:29:55,350 - > 00:29:59,830 millions of dollars in in revenue. But then also teams are 468 00:29:59,830 - > 00:30:03,910 necessarily distributed these days and maybe it's not 469 00:30:03,910 - > 00:30:08,255 necessary for those AI researchers or developers to 470 00:30:08,255 - > 00:30:11,695 move from other places around the world to those countries to 471 00:30:11,695 - > 00:30:15,055 do that work. So even if the companies are funded there and 472 00:30:15,055 - > 00:30:19,055 the VC is here, that that might not reflect where the 473 00:30:19,055 - > 00:30:22,660 individuals in the company live or operate. Right. 474 00:30:22,740 - > 00:30:24,180 That makes perfect sense. 475 00:30:24,660 - > 00:30:28,100 Chris: And, you know, talking about, you know, VC and the the 476 00:30:28,100 - > 00:30:34,420 adoption rate and stuff, we we the last few months, the amount 477 00:30:34,420 - > 00:30:39,655 of AI adoption has just skyrocketed. People were until 478 00:30:39,655 - > 00:30:43,015 until 2026, maybe, you know, late second half of twenty 479 00:30:43,015 - > 00:30:45,975 twenty five, I I knew a lot of people who were not in 480 00:30:45,975 - > 00:30:49,095 technology, and they were touching on different model 481 00:30:49,095 - > 00:30:53,140 programs here and there. Most of the, you know, the the people 482 00:30:53,140 - > 00:30:56,340 that I would be talking to would be touching the closed frontier 483 00:30:56,340 - > 00:31:00,260 models, whichever one they particularly went for. But we've 484 00:31:00,260 - > 00:31:03,300 seen a marked improve improvement in terms of 485 00:31:03,300 - > 00:31:08,905 adoption. I really don't think I've run across anyone recently, 486 00:31:10,105 - > 00:31:12,745 including I'm all all age brackets. 487 00:31:13,625 - > 00:31:17,780 I have friends who are in their nineties who are using who are 488 00:31:17,780 - > 00:31:21,620 using the tools that are out there now. Mostly free mostly 489 00:31:21,620 - > 00:31:25,220 free, and that's what Stanford had noted as the different free 490 00:31:25,220 - > 00:31:29,460 tiers. But but that has that's been interesting to see, but I 491 00:31:29,460 - > 00:31:32,180 noticed that I think the one thing that may have surprised me 492 00:31:32,475 - > 00:31:36,555 was that Stanford had noticed that we are still, in The US, 493 00:31:36,715 - > 00:31:41,595 ranked twenty fourth at only a 28.3% adoption. So obviously, 494 00:31:41,595 - > 00:31:45,915 I'm not I'm not the, you know, representative in terms of my 495 00:31:45,915 - > 00:31:49,000 own experience of that. What are your thoughts there? 496 00:31:49,560 - > 00:31:55,640 Dan: Yeah. It's it's super interesting. I I don't know, the 497 00:31:55,640 - > 00:32:02,985 full economics of some of these free access systems and how 498 00:32:03,145 - > 00:32:09,065 much, you know, usage for example, like Gemini's Google 499 00:32:09,065 - > 00:32:15,490 Gemini's getting off of usage on Android phones from free users 500 00:32:15,490 - > 00:32:20,690 versus like paid workspace accounts. I I I don't know how 501 00:32:20,690 - > 00:32:25,730 all the economics play out there, but certainly I know just 502 00:32:25,730 - > 00:32:30,975 anecdotally the last couple flights I've been on, for 503 00:32:30,975 - > 00:32:33,855 example, I look around and of course people are on their 504 00:32:33,855 - > 00:32:37,615 phones and a significant number of those people on their phones 505 00:32:37,615 - > 00:32:44,340 are chatting with an AI app of some type and some chatting with 506 00:32:44,340 - > 00:32:48,420 that AI app throughout the entire flight on, you know, WiFi 507 00:32:48,420 - > 00:32:53,220 and, you know, not even watching a movie or whatever. Like, it 508 00:32:53,140 - > 00:32:55,675 Chris: so I've been guilty of that myself. 509 00:32:55,675 - > 00:33:00,795 Dan: Yeah. Yeah. And so, yeah, it it will be it will be 510 00:33:00,795 - > 00:33:05,275 interesting to see how that that level of usage continues to 511 00:33:05,275 - > 00:33:09,800 spread. I don't under like I say, understand some of those 512 00:33:09,800 - > 00:33:15,560 direct to consumer mechanisms as well as, some of the b to b type 513 00:33:15,560 - > 00:33:18,680 of things that that I that I interact with day to day. 514 00:33:18,680 - > 00:33:22,115 Chris: I will well, before we leave the topic of kind of, you 515 00:33:22,115 - > 00:33:25,795 know, general population AI adoption, there's something that 516 00:33:25,795 - > 00:33:28,515 I've noticed that I was meaning to bring into the show at the 517 00:33:28,515 - > 00:33:32,915 right moment anyway, and that was my mother is a is a 518 00:33:32,915 - > 00:33:37,860 technologist. She's retired mid eighties. Did old school AI way 519 00:33:37,860 - > 00:33:44,020 back. But she's been kind of reengaging in recent times, and 520 00:33:44,260 - > 00:33:48,580 I noticed that she was she was talking to me the other night 521 00:33:48,580 - > 00:33:51,805 about she likes to paint, and then she'll capture thing an 522 00:33:51,805 - > 00:33:55,165 image in Photoshop historically and work on it in Photoshop and 523 00:33:55,165 - > 00:33:57,725 get it just the way she wants it. And I was like, well, mom, 524 00:33:57,725 - > 00:34:00,125 you could do that in any of these tools and just have it do 525 00:34:00,125 - > 00:34:00,365 it. 526 00:34:00,365 - > 00:34:03,090 Like, just tell it what you want and it will do it. And I 527 00:34:03,090 - > 00:34:07,090 realized that she was taking pride in working in Photoshop 528 00:34:07,090 - > 00:34:10,370 with her skills there, and I got about halfway through trying to 529 00:34:10,370 - > 00:34:12,930 convince her and I backed away because I suddenly realized, 530 00:34:12,930 - > 00:34:15,775 like, this is her hobby and there's fun, and even if she 531 00:34:15,775 - > 00:34:19,695 could, in a productive way, move right to the end goal, maybe 532 00:34:19,695 - > 00:34:23,055 this is a moment where AI is not the right thing to bring into 533 00:34:23,055 - > 00:34:26,815 play just because she enjoys doing it. Yeah. So I just 534 00:34:26,815 - > 00:34:29,295 thought I wanted to bring that human element back into it, that 535 00:34:29,295 - > 00:34:32,060 not AI for all things is always the answer. 536 00:34:32,060 - > 00:34:35,580 Dan: Yeah, I think that there is a distinction there, Chris, 537 00:34:35,580 - > 00:34:40,380 because you could look at other, other examples of this, right? 538 00:34:40,380 - > 00:34:45,275 It's, it, I know people that brew their own beer. Right? 539 00:34:45,275 - > 00:34:50,235 There's no reason conceivable that people should brew their 540 00:34:50,235 - > 00:34:54,155 own beer because they can just go down the street for less 541 00:34:54,155 - > 00:34:58,760 money with probably better results and and and get 542 00:34:58,760 - > 00:35:02,120 something, you know, that that tastes great's already cold. 543 00:35:02,120 - > 00:35:02,520 Right? 544 00:35:02,520 - > 00:35:06,520 But that's not that's not the point, right, to to what you 545 00:35:06,520 - > 00:35:10,680 were saying. That's not why they're doing things in that 546 00:35:10,680 - > 00:35:15,205 way. And that it'll be interesting to see what kind of 547 00:35:16,165 - > 00:35:23,045 what elements also see we we see a resurgence of, I don't know, 548 00:35:23,045 - > 00:35:29,190 people that just want to use use Excel and do analysis because 549 00:35:29,190 - > 00:35:33,590 they enjoy doing it. Yeah. I I think what is true is if you're 550 00:35:33,590 - > 00:35:37,430 in a job where there's productivity expectations and 551 00:35:37,430 - > 00:35:41,005 you're doing those things, then that's no longer gonna be 552 00:35:41,005 - > 00:35:41,885 acceptable. 553 00:35:41,885 - > 00:35:43,165 Right? So in the same way 554 00:35:43,245 - > 00:35:44,845 Chris: hobbyist and professional. 555 00:35:44,925 - > 00:35:48,525 Dan: Yeah. In the same way that, like, some people might like 556 00:35:48,525 - > 00:35:52,580 writing physical snail mail letters. But if you were to 557 00:35:52,580 - > 00:35:56,500 insist that you're only going to write physical mail letters, 558 00:35:56,580 - > 00:35:59,220 send them through the mail as part of your company 559 00:35:59,220 - > 00:36:02,180 communication and you're not gonna use email, that's not 560 00:36:02,180 - > 00:36:05,220 gonna work in the world that we live in. Right? 561 00:36:05,220 - > 00:36:08,505 Chris: Yeah. I I will tie a bow on this by saying to your 562 00:36:08,505 - > 00:36:12,985 example, once upon a time, I did I took a a hand at at trying to 563 00:36:12,985 - > 00:36:17,865 brew some beer and and also wine, and both my beer and my 564 00:36:17,865 - > 00:36:23,120 wine were terrible, but I took great pride and I drank them 565 00:36:23,120 - > 00:36:26,960 because I took a lot of pride in it, whereas I certainly wouldn't 566 00:36:26,960 - > 00:36:31,120 have done that in a professional context, but yes, it's 567 00:36:31,120 - > 00:36:35,145 interesting to see the human element inching back in and 568 00:36:35,145 - > 00:36:36,345 recognizing So, that there's a 569 00:36:37,465 - > 00:36:39,865 Dan: you were trying to try a bow on this, but I think one 570 00:36:39,865 - > 00:36:46,825 more anecdote is because really one of my friends that brews 571 00:36:46,825 - > 00:36:50,190 their own beer, one of the things that they told me that 572 00:36:50,190 - > 00:36:54,510 they did recently was they said they they actually used their 573 00:36:54,510 - > 00:37:01,710 creativity in a in in an AI tool and said, I'm I kind of wanna 574 00:37:01,710 - > 00:37:06,295 brew a beer that's, you know, like this, and it has this ABV 575 00:37:06,295 - > 00:37:11,095 and, like, these notes, and, it's kinda this style. I kinda 576 00:37:11,095 - > 00:37:14,615 want it to turn out like this. And he gave all that description 577 00:37:14,615 - > 00:37:19,400 and actually had the AI system create the full build materials 578 00:37:19,400 - > 00:37:23,080 and and recipe for that. And then he went to the brewing 579 00:37:23,080 - > 00:37:27,160 store and, you know, actually had the person there like, hey, 580 00:37:27,160 - > 00:37:29,555 could you look at this and see if this is legit? And they're 581 00:37:29,555 - > 00:37:32,035 like, yeah, this seems this seems great to to us. 582 00:37:32,035 - > 00:37:36,755 And they got that and he he brewed the beer and it was it 583 00:37:36,755 - > 00:37:40,515 was it worked out great. So I think actually this could be a 584 00:37:40,515 - > 00:37:47,490 spark to some of those hobbies on that side of things. I know, 585 00:37:47,730 - > 00:37:51,570 you know, even Well, I just, of course, on, like, travel and 586 00:37:51,570 - > 00:37:55,250 trips and that sort of thing, I heavily use AI systems to help 587 00:37:55,250 - > 00:37:59,250 me plan out things and do research and such, and that's, 588 00:37:59,250 - > 00:38:02,705 you know, part of what I enjoy about doing the planning of 589 00:38:02,705 - > 00:38:05,265 trips or something like that. Absolutely. 590 00:38:05,425 - > 00:38:06,305 Chris: Moving on. 591 00:38:06,785 - > 00:38:10,305 Dan: Yeah. What are we on? We're on Takeaway 9. 592 00:38:11,025 - > 00:38:11,505 Chris: That's right. 593 00:38:11,920 - > 00:38:16,720 Dan: Which is productivity gains from AI are appearing in many of 594 00:38:16,720 - > 00:38:21,360 the same fields where entry level employment is starting to 595 00:38:21,360 - > 00:38:23,040 decline. That's 596 00:38:23,360 - > 00:38:25,760 Chris: I think we've talked a lot about this on the show over 597 00:38:25,760 - > 00:38:26,480 the months. Yes. 598 00:38:27,255 - > 00:38:31,575 Dan: Yeah. There's no way you can get a junior software dev 599 00:38:31,575 - > 00:38:34,295 position writing SQL queries anymore. 600 00:38:34,295 - > 00:38:37,735 Chris: That's that's right. Yeah. That's gone. It's it's 601 00:38:37,735 - > 00:38:41,540 interesting, and and and the senior level people have have 602 00:38:41,540 - > 00:38:46,420 been learning very rapidly that it is time to embrace. I I 603 00:38:46,420 - > 00:38:47,620 really it's funny. 604 00:38:47,620 - > 00:38:51,300 As we went into New Year's, I knew a lot of people who were 605 00:38:51,300 - > 00:38:54,945 still pushing back on that. At this point. As we're recording 606 00:38:54,945 - > 00:38:59,585 in late May, I can't think of anybody I know that's pushing 607 00:38:59,585 - > 00:39:02,625 back on that at this point, at least not not that I know well 608 00:39:02,625 - > 00:39:05,185 enough to have these conversations with. So that's 609 00:39:05,185 - > 00:39:08,000 changed. The world has changed in a in a very short amount of 610 00:39:08,000 - > 00:39:10,960 time, certainly in coding, but I think in a lot of other areas. 611 00:39:10,960 - > 00:39:14,400 You know, you talked about Excel. If you're not a technical 612 00:39:14,400 - > 00:39:17,120 person, but you're using Microsoft Office, and to your 613 00:39:17,120 - > 00:39:21,745 point a few minutes ago, a lot of the a lot of the basic stuff, 614 00:39:22,065 - > 00:39:24,705 you're gonna be using an AI assistant in your tool, and I 615 00:39:24,705 - > 00:39:28,545 think that's going that's going across many, many industries. So 616 00:39:28,545 - > 00:39:29,745 not not surprising. 617 00:39:29,745 - > 00:39:33,500 Dan: Yeah. Yeah. Yeah. And I I think, I've been thinking about 618 00:39:33,500 - > 00:39:39,420 this over the previous weeks, Chris, also because, I mean, we 619 00:39:39,420 - > 00:39:43,580 have hired folks into our company and I'm sure I'll be 620 00:39:43,580 - > 00:39:48,575 part of hiring processes in the future, And it's interesting 621 00:39:48,575 - > 00:39:54,495 because there you could say there's no longer the chance for 622 00:39:54,495 - > 00:39:58,815 the entry level jobs, but there's still a chance to hire 623 00:39:58,815 - > 00:40:04,980 people in and give them, even if they're a more junior engineer, 624 00:40:05,540 - > 00:40:09,780 like we have, for example, in our company, maybe like we have 625 00:40:09,780 - > 00:40:14,020 our own repo with all the skills, quad code skills that 626 00:40:14,020 - > 00:40:17,565 are relevant to our stack and connect to this and that and 627 00:40:17,565 - > 00:40:22,845 help you get up and going. And really like that, that's that's 628 00:40:22,845 - > 00:40:25,485 a lot of power that you can give someone. 629 00:40:25,485 - > 00:40:29,245 Now, obviously, there are debugging things and 630 00:40:29,245 - > 00:40:32,660 architecture things that are very take take a very skilled 631 00:40:32,660 - > 00:40:36,740 person, a more senior person to get to the bottom of. But I 632 00:40:36,740 - > 00:40:41,940 think, I guess what I'm saying is I think these tools can also 633 00:40:41,940 - > 00:40:46,945 help junior folks coming into a position to level up more 634 00:40:46,945 - > 00:40:51,025 rapidly than they were doing before. And maybe even if 635 00:40:51,025 - > 00:40:55,665 universities or educational systems embrace those tools and 636 00:40:55,665 - > 00:40:58,625 help them level up even before they were on the job market, 637 00:40:59,080 - > 00:41:02,600 then they might, you know, have have more of a, more of a 638 00:41:02,600 - > 00:41:08,280 chance. Not to not to not, you know, I I definitely acknowledge 639 00:41:08,280 - > 00:41:11,960 that jobs will be lost here. There there certainly will be, 640 00:41:11,960 - > 00:41:12,200 right? 641 00:41:12,425 - > 00:41:15,305 And that is a hard thing for many people. 642 00:41:15,385 - > 00:41:17,945 Chris: Yeah. Going back, before we leave this, because I think 643 00:41:17,945 - > 00:41:20,105 this is important, there there are other things we can skip 644 00:41:20,105 - > 00:41:24,105 over in the interest of time, but I think the way I think your 645 00:41:24,185 - > 00:41:27,225 a really big point there is the way we're learning things has to 646 00:41:27,225 - > 00:41:32,230 change too in that and I'll share a two second experience, 647 00:41:32,310 - > 00:41:35,190 is that both of us have been through many programming 648 00:41:35,190 - > 00:41:40,630 languages over the years. And and I had gotten an interest in 649 00:41:40,630 - > 00:41:43,965 Rust, and I was dipping in and out of it and not really taking 650 00:41:43,965 - > 00:41:47,645 it on board, and I'd get caught up into something else, and and 651 00:41:47,645 - > 00:41:51,325 it it's famous for its steep learning curve. It's taking a 652 00:41:51,325 - > 00:41:54,845 little better this time because one of the things that I've done 653 00:41:54,845 - > 00:42:00,780 to try to get go from very entry level to beyond that is to to 654 00:42:00,780 - > 00:42:05,020 use the tools that are out there, using Claude code, and 655 00:42:05,020 - > 00:42:09,740 and and creating it, but also having it explain and having 656 00:42:09,740 - > 00:42:13,865 discussions about what's going on and why. So it's not just 657 00:42:13,865 - > 00:42:18,425 make the thing, but it's let's have a conversation about what 658 00:42:18,425 - > 00:42:20,825 the thing, how you make the thing and why the choices are 659 00:42:20,825 - > 00:42:24,185 being made, and you can use the model to say, well, why are you 660 00:42:24,185 - > 00:42:24,745 doing that? 661 00:42:24,745 - > 00:42:27,880 What's the what's the rationale? Would it make sense to do this 662 00:42:27,880 - > 00:42:33,320 and have that? And and I that has been transforming for me, is 663 00:42:33,320 - > 00:42:38,680 not fresh out of college, to to continue to learn at a rapid 664 00:42:38,680 - > 00:42:42,565 pace. And it's been a great experience. It's been different, 665 00:42:42,565 - > 00:42:45,125 but I would encourage people to be open to that. 666 00:42:45,125 - > 00:42:49,205 And so don't just have the model do whatever your thing is. Don't 667 00:42:49,205 - > 00:42:53,605 just have it do it. Have it explain and share in the load as 668 00:42:53,605 - > 00:42:57,050 you go so that it is it is a creative, but also a learning 669 00:42:57,050 - > 00:43:00,410 process as you do it. And then you'll come out of it better 670 00:43:00,410 - > 00:43:04,410 than you started as well. So I I I really wanted to to just kinda 671 00:43:04,410 - > 00:43:06,570 get out get out there and urge people to give it a shot. 672 00:43:06,570 - > 00:43:12,425 Dan: Yeah. It does tie into one of the other takeaways, which 673 00:43:12,665 - > 00:43:15,625 yeah. Chris, we can just mention a few of these as we get closer 674 00:43:15,625 - > 00:43:19,385 to closing out here, but formal education is lagging behind AI, 675 00:43:19,385 - > 00:43:22,265 but people are learning AI skills at every stage of life. 676 00:43:22,265 - > 00:43:25,980 That was one of the other takeaways. I think, you know, 677 00:43:25,980 - > 00:43:29,980 related to related to what you're what you're talking 678 00:43:29,980 - > 00:43:30,540 about. 679 00:43:31,660 - > 00:43:35,980 And, yeah, it's it's like, I I think the stat they gave is 80% 680 00:43:35,980 - > 00:43:39,505 of high school and college students now use AI for school 681 00:43:39,505 - > 00:43:45,665 related things, but a very small percentage of teachers, you 682 00:43:45,825 - > 00:43:50,305 know, have any sort of policy in place around around that usage 683 00:43:50,305 - > 00:43:54,750 or either positive or negative. Right? Which I think both of us 684 00:43:54,750 - > 00:43:58,110 would hope that some of that is on the positive side and there's 685 00:43:58,110 - > 00:44:01,550 teachers and universities that are helping students learn how 686 00:44:01,550 - > 00:44:04,990 to use these tools and encouraging their use versus 687 00:44:05,165 - > 00:44:08,365 trying to always police it and shut it down, which isn't gonna 688 00:44:08,365 - > 00:44:08,685 work. 689 00:44:08,685 - > 00:44:10,685 Chris: Which is the right yeah. It's not gonna work. And I have 690 00:44:10,685 - > 00:44:14,205 an incoming high school student of my own, and I have I have 691 00:44:14,205 - > 00:44:17,325 been telling her for years now to use the tool. I want her to 692 00:44:17,325 - > 00:44:21,380 use the tools, but I'm but going back to my previous comment, we 693 00:44:21,460 - > 00:44:25,300 I'll sit down with her, and we will use the tools to learn so 694 00:44:25,300 - > 00:44:28,820 that she actually comes out. And by way of example, even though 695 00:44:28,820 - > 00:44:33,620 she's a heavy user of AI tools for school, without those tools, 696 00:44:33,620 - > 00:44:37,065 she'll go in she went in for her finals in middle school just 697 00:44:37,065 - > 00:44:37,465 now. 698 00:44:37,465 - > 00:44:40,345 Straight a's. You know? And that's without any AI tools 699 00:44:40,345 - > 00:44:43,225 being available to her to do that. She learned the material, 700 00:44:43,225 - > 00:44:45,785 and that's trying to use it the right way. Use it to where 701 00:44:45,785 - > 00:44:49,545 you're learning in addition to just getting the job done like a 702 00:44:49,545 - > 00:44:50,745 lot of students might do. 703 00:44:50,745 - > 00:44:54,700 So it's it's the right way to do it. And I I as a parent, I would 704 00:44:54,700 - > 00:44:58,940 I would urge teachers, and I know you're restricted by your 705 00:44:58,940 - > 00:45:03,660 school policies and such, but try to be open to that, open to 706 00:45:03,660 - > 00:45:07,205 to to taking on the new tools over with your students. 707 00:45:07,365 - > 00:45:10,005 Dan: Yeah. Well, just to mention a few of these that we didn't 708 00:45:10,005 - > 00:45:14,005 get to, Chris, I encourage people to look up the report. 709 00:45:14,005 - > 00:45:17,365 We'll link it in our show notes. You can read all the details. 710 00:45:17,365 - > 00:45:21,930 The full report is 425 pages long, so you probably wanna 711 00:45:21,930 - > 00:45:26,170 stick that into some type of AI thing and and ask some questions 712 00:45:26,170 - > 00:45:27,450 and and work through it. 713 00:45:27,450 - > 00:45:30,330 Some of the other ones that, just so people know them, are 714 00:45:30,330 - > 00:45:36,355 AI's environmental footprint is expanding. AI models for science 715 00:45:36,355 - > 00:45:39,315 outperform human scientists, though bigger models do not 716 00:45:39,315 - > 00:45:44,195 always perform better. AI is transforming clinical care, but 717 00:45:44,195 - > 00:45:48,390 rigorous evidence remains limited. AI sovereignty is 718 00:45:48,390 - > 00:45:53,510 becoming a defining feature of national policy, and AI experts 719 00:45:53,510 - > 00:45:56,710 and the public have very different perspectives on the 720 00:45:56,710 - > 00:46:00,550 technology's future. So all of those super interesting. 721 00:46:00,630 - > 00:46:03,725 Of course, go and look at the article, article, dig into the 722 00:46:03,725 - > 00:46:07,405 details, have some fun exploring it, and and thanks again to 723 00:46:07,405 - > 00:46:10,685 Stanford for continuing to to do this great work. 724 00:46:10,685 - > 00:46:13,565 Chris: They do great great reports every year. We loved I 725 00:46:13,965 - > 00:46:16,940 we wait for this every year to dig in to and have fun with it. 726 00:46:17,020 - > 00:46:20,700 Dan: So awesome, Chris. Well, have have fun yourself with, 727 00:46:21,100 - > 00:46:29,180 with your non nonhumanoid robots at home and the AI tools for for 728 00:46:29,180 - > 00:46:30,540 school and all the things. 729 00:46:31,595 - > 00:46:32,635 Chris: See you next time. 730 00:46:37,035 - > 00:46:40,235 Narrator: Alright. That's our show for this week. If you 731 00:46:40,235 - > 00:46:44,395 haven't checked out our website, head to practicalai.fm, and be 732 00:46:44,395 - > 00:46:47,830 sure connect with us on LinkedIn, X, or Blue Sky. You'll 733 00:46:47,830 - > 00:46:51,190 see us posting insights related to the latest AI developments, 734 00:46:51,190 - > 00:46:54,150 and we would love for you to join the conversation. Thanks to 735 00:46:54,150 - > 00:46:57,030 our partner Prediction Guard for providing operational support 736 00:46:57,030 - > 00:46:57,510 for the show. 737 00:46:58,035 - > 00:47:01,235 Check them out at predictionguard.com. Also, 738 00:47:01,315 - > 00:47:03,875 thanks to Breakmaster Cylinder for the beats and to you for 739 00:47:03,875 - > 00:47:06,835 listening. That's all for now, but you'll hear from us again 740 00:47:06,835 - > 00:47:07,395 next week.