WEBVTT 00:00:00.000 --> 00:00:02.040 You know, sometimes the really big shifts, they 00:00:02.040 --> 00:00:04.419 happen kind of quietly, away from the main headlines. 00:00:04.519 --> 00:00:06.860 Imagine a giant in the tech world, someone everyone 00:00:06.860 --> 00:00:11.529 knows, getting sort of overtaken. their most 00:00:11.529 --> 00:00:13.570 important fight. That's what we're unpacking 00:00:13.570 --> 00:00:15.449 today. Right. And it's not just about, you know, 00:00:15.449 --> 00:00:17.210 who's winning right now. It's really about how 00:00:17.210 --> 00:00:19.730 AI itself is evolving. We're going to deep dive 00:00:19.730 --> 00:00:22.129 into a whole stack of fresh insights. We'll look 00:00:22.129 --> 00:00:24.489 at new market dynamics and breakthrough tools 00:00:24.489 --> 00:00:27.789 and even these autonomous AI agents that might 00:00:27.789 --> 00:00:30.449 just change, well, everything. So what does all 00:00:30.449 --> 00:00:32.609 this mean for you? For anyone trying to navigate 00:00:32.609 --> 00:00:35.689 this fast -paced world, we've got the nuggets, 00:00:35.810 --> 00:00:38.189 the insights, and maybe a few surprising facts 00:00:38.189 --> 00:00:41.780 along the way. Let's explore it. Okay, so our 00:00:41.780 --> 00:00:43.520 deep dive kicks off with what a lot of people 00:00:43.520 --> 00:00:45.039 are calling the most surprising shift in the 00:00:45.039 --> 00:00:48.240 AI market, looking at mid -2025. For the very 00:00:48.240 --> 00:00:50.520 first time, it seems Anthropix cloud models have 00:00:50.520 --> 00:00:53.960 jumped to number one. For enterprise AI, and 00:00:53.960 --> 00:00:55.840 this isn't just a small nudge, it feels like 00:00:55.840 --> 00:00:58.679 a pretty monumental change. It really is. And 00:00:58.679 --> 00:01:01.000 what's fascinating is the data that backs this 00:01:01.000 --> 00:01:03.579 up. This comes from Menlo Ventures. Claude now 00:01:03.579 --> 00:01:06.280 apparently holds about 32 % of enterprise usage. 00:01:06.519 --> 00:01:08.959 That actually puts it ahead of OpenAI, which 00:01:08.959 --> 00:01:12.260 is sitting at 25%. And if you zoom in specifically 00:01:12.260 --> 00:01:14.560 on coding workloads, you know, where AI is really 00:01:14.560 --> 00:01:16.959 being put through its paces daily, Claude's lead 00:01:16.959 --> 00:01:21.469 is even bigger. 42 % compared to OpenAI's 21%. 00:01:21.469 --> 00:01:24.030 And this is a complete flip from just last year, 00:01:24.150 --> 00:01:27.670 2023. Back then, OpenAI had, what, 50 %? Dominant. 00:01:27.689 --> 00:01:31.250 Totally. And Claude was way back at 12%. So, 00:01:31.250 --> 00:01:34.689 yeah. A seismic shift in barely a year. It is 00:01:34.689 --> 00:01:36.890 a really striking contrast, isn't it? Yeah. I 00:01:36.890 --> 00:01:38.650 mean, OpenAI has definitely captured the public 00:01:38.650 --> 00:01:40.689 imagination, dominated the consumer headlines, 00:01:40.810 --> 00:01:43.909 processing, what was it, an incredible 2 .5 billion 00:01:43.909 --> 00:01:47.010 prompts a day. Wow. Huge numbers. But while all 00:01:47.010 --> 00:01:48.790 that consumer spotlight was burning so bright, 00:01:48.890 --> 00:01:51.849 Cloud was quietly, almost like stealthily, rising 00:01:51.849 --> 00:01:53.450 up, becoming what a lot of developers are now 00:01:53.450 --> 00:01:56.170 calling their best friend. Especially for companies 00:01:56.170 --> 00:01:58.010 actually building and shipping real products. 00:01:58.109 --> 00:01:59.829 It highlights a very different battleground for 00:01:59.829 --> 00:02:02.219 AI. Exactly. And the dev chatter, you know, the 00:02:02.219 --> 00:02:05.439 buzz among developers really backs this up. We're 00:02:05.439 --> 00:02:07.819 hearing companies consistently choose Claude 00:02:07.819 --> 00:02:10.460 because it's better at long form tasks. It also 00:02:10.460 --> 00:02:12.860 seems to be easier to plug into their existing 00:02:12.860 --> 00:02:15.840 workflows, experiences fewer failures, apparently, 00:02:16.020 --> 00:02:19.370 and just generally feels well. More enterprise 00:02:19.370 --> 00:02:21.210 ready. That's the phrase people use. These aren't 00:02:21.210 --> 00:02:23.250 just minor points. They speak to that core liability, 00:02:23.530 --> 00:02:26.150 that predictability businesses really need. And 00:02:26.150 --> 00:02:28.629 building that kind of deep reliability, that 00:02:28.629 --> 00:02:31.050 trust, especially in the enterprise world, that's 00:02:31.050 --> 00:02:33.569 incredibly hard work. Once you've got it, it's 00:02:33.569 --> 00:02:36.270 even harder for someone else to unseat you. So 00:02:36.270 --> 00:02:38.669 this signals that, OK, open AI might still be 00:02:38.669 --> 00:02:42.030 the king of consumer AI for now. But they absolutely 00:02:42.030 --> 00:02:45.090 need a stronger B2B strategy. Yeah. And probably 00:02:45.090 --> 00:02:47.969 fast. Yeah. Because the enterprise AI space is 00:02:47.969 --> 00:02:50.409 evolving super quickly and others are clearly 00:02:50.409 --> 00:02:52.669 leapfrogging them right here. Yeah. And if you 00:02:52.669 --> 00:02:54.289 connect this to the bigger picture, like why 00:02:54.289 --> 00:02:57.409 it matters, it basically means cloud is increasingly 00:02:57.409 --> 00:03:00.289 seen as the safe bet for both startups and big 00:03:00.289 --> 00:03:02.370 enterprises when they're picking their first 00:03:02.370 --> 00:03:05.150 or their main large language model, their LLM. 00:03:05.250 --> 00:03:07.349 Right. That foundational AI model trained on 00:03:07.349 --> 00:03:10.069 huge amounts of text to understand and talk like 00:03:10.069 --> 00:03:12.430 a. human exactly so Claude's got this image now 00:03:12.430 --> 00:03:16.629 quiet reliable well supported constantly improving 00:03:16.629 --> 00:03:19.770 makes it a very appealing kind of low -risk choice 00:03:19.770 --> 00:03:21.389 when you're putting something into production 00:03:21.389 --> 00:03:24.729 so this quiet reliability of Claude what's the 00:03:24.729 --> 00:03:27.289 fundamental impact for businesses then It lets 00:03:27.289 --> 00:03:30.889 companies ship products faster, more confidently. 00:03:30.909 --> 00:03:33.229 It's just a dependable choice. Okay, let's switch 00:03:33.229 --> 00:03:35.770 gears. Let's talk about some other quickfire 00:03:35.770 --> 00:03:38.009 AI highlights that caught our eye, starting with 00:03:38.009 --> 00:03:40.710 a really important one about privacy. So it came 00:03:40.710 --> 00:03:44.050 out that Google had kind of quietly indexed shared 00:03:44.050 --> 00:03:47.689 links from chat GPT, which potentially made some 00:03:47.689 --> 00:03:51.129 private chats public. OpenAI stepped in pretty 00:03:51.129 --> 00:03:53.879 quickly, shut that down. Thank you. Yeah. But 00:03:53.879 --> 00:03:55.819 it's really potent reminder for you listening. 00:03:55.939 --> 00:03:58.419 If you've ever shared a GPT link, maybe double 00:03:58.419 --> 00:04:00.659 check your settings. Be mindful of what goes 00:04:00.659 --> 00:04:02.879 public. Yeah, definitely a good wake up call 00:04:02.879 --> 00:04:05.400 for, you know, digital hygiene. Right. And speaking 00:04:05.400 --> 00:04:07.620 of shifts, Microsoft put out these fascinating 00:04:07.620 --> 00:04:10.699 lists recently looking at AI's impact on jobs. 00:04:10.819 --> 00:04:14.300 They flagged like. 40 jobs potentially impacted, 00:04:14.539 --> 00:04:16.680 maybe doomed, as some headlines put it. A little 00:04:16.680 --> 00:04:19.300 dramatic, maybe. Ah, yeah. And then 40 jobs considered 00:04:19.300 --> 00:04:21.839 safe. But what's really insightful isn't just 00:04:21.839 --> 00:04:24.779 the specific jobs. It's the types of tasks AI 00:04:24.779 --> 00:04:27.240 is automating. You know, repetitive stuff, data 00:04:27.240 --> 00:04:29.920 -heavy, predictable tasks. Versus the jobs proving 00:04:29.920 --> 00:04:32.379 resilient, which often need complex problem solving, 00:04:32.540 --> 00:04:34.579 creativity, that kind of unique human interaction. 00:04:34.939 --> 00:04:37.319 It's a snapshot of how work itself is changing. 00:04:37.540 --> 00:04:39.899 That's a great way to look at it. And talking 00:04:39.899 --> 00:04:42.300 about breakthroughs that show AI. AI is evolving 00:04:42.300 --> 00:04:44.860 power. This is where it gets really interesting. 00:04:45.040 --> 00:04:47.480 There's this viral super agent called Manus. 00:04:47.600 --> 00:04:50.100 It just launched a feature called wide research. 00:04:50.300 --> 00:04:53.540 Now, Manus is this AI system focused on automated 00:04:53.540 --> 00:04:56.579 research, putting knowledge together and wide 00:04:56.579 --> 00:04:59.519 research. It lets multiple AI agents work together 00:04:59.519 --> 00:05:02.660 simultaneously, processing truly massive amounts 00:05:02.660 --> 00:05:06.139 of data. Whoa. I mean, imagine scaling that. 00:05:06.750 --> 00:05:09.029 billion queries you let multiple specialized 00:05:09.029 --> 00:05:12.250 agents just attack that data together each from 00:05:12.250 --> 00:05:14.610 a different angle that's a whole new level of 00:05:14.610 --> 00:05:16.889 scalable intelligent insight it's pretty incredible 00:05:16.889 --> 00:05:19.589 to think about and uh on the app development 00:05:19.589 --> 00:05:23.019 side there was this cool real world test A founder 00:05:23.019 --> 00:05:25.800 took Lovable, that's an AI app builder, and ReapBlit, 00:05:25.879 --> 00:05:27.920 the online coding platform, and tried to clone 00:05:27.920 --> 00:05:30.920 five viral apps using both head to head. Apparently, 00:05:30.920 --> 00:05:33.019 a clear winner emerged pretty quickly for speed 00:05:33.019 --> 00:05:35.259 and ease of replication. Just highlights how 00:05:35.259 --> 00:05:36.899 accessible app development is getting, even if 00:05:36.899 --> 00:05:39.060 you're not a hardcore coder. Right. And Apple. 00:05:39.120 --> 00:05:41.240 They've been getting some heat for maybe lagging 00:05:41.240 --> 00:05:43.939 a bit in the AI race. But it sounds like they're 00:05:43.939 --> 00:05:47.310 making significant moves now. planning to significantly 00:05:47.310 --> 00:05:50.029 grow their AI investments, being very open to 00:05:50.029 --> 00:05:52.850 acquisitions. Their goal isn't necessarily like 00:05:52.850 --> 00:05:55.990 launching standalone AI gadgets. It seems more 00:05:55.990 --> 00:05:58.410 about deeply integrating AI into their existing 00:05:58.410 --> 00:06:01.170 stuff. Yeah, the ecosystem. Exactly. Enhancing 00:06:01.170 --> 00:06:04.529 Siri, maybe better photo processing, just making 00:06:04.529 --> 00:06:06.949 user experiences smoother across their products. 00:06:07.009 --> 00:06:09.170 It's more about that pervasive, maybe subtle 00:06:09.170 --> 00:06:12.050 AI integration. And underpinning all these advances, 00:06:12.149 --> 00:06:14.430 you need the infrastructure, right? So find an 00:06:14.430 --> 00:06:16.139 AI infrastructure. company, pretty significant 00:06:16.139 --> 00:06:18.680 one, they just raised a big round, $125 million, 00:06:18.959 --> 00:06:22.259 led by Meritech, but Salesforce Ventures and 00:06:22.259 --> 00:06:25.600 Google's AI Futures Fund chipped in too. Shows 00:06:25.600 --> 00:06:27.420 there's still huge investment going into not 00:06:27.420 --> 00:06:29.660 just the AI models themselves, but the kind of 00:06:29.660 --> 00:06:32.199 plumbing, the tech that lets them scale and run. 00:06:32.439 --> 00:06:34.480 So thinking about all these different pieces, 00:06:34.680 --> 00:06:37.779 privacy fixes, job impacts, scalable research, 00:06:38.000 --> 00:06:40.720 app building, how quickly are these advancements 00:06:40.720 --> 00:06:43.129 really shaping our daily tech lives? Oh, they're 00:06:43.129 --> 00:06:45.949 integrating super fast, making everyday tasks 00:06:45.949 --> 00:06:48.430 and even complex development feel more intelligent. 00:06:48.910 --> 00:06:51.569 Okay, let's pivot now. Let's talk about how AI 00:06:51.569 --> 00:06:55.509 is moving beyond just chatting. Moving into real 00:06:55.509 --> 00:06:58.069 automation, we're starting to see this new standard 00:06:58.069 --> 00:07:00.550 emerge. It's called the Model Context Protocol, 00:07:00.829 --> 00:07:03.750 or MCP. Yeah, MCP. What's cool about it is how 00:07:03.750 --> 00:07:06.129 much it simplifies things, potentially. Think 00:07:06.129 --> 00:07:08.810 of it as giving an AI hands to interact with 00:07:08.810 --> 00:07:10.870 the digital world, not just a voice. That's the 00:07:10.870 --> 00:07:14.089 analogy people use. It lets AI use your existing 00:07:14.089 --> 00:07:16.329 tools, connect to different applications, and 00:07:16.329 --> 00:07:19.360 really automate complex workflows. So before 00:07:19.360 --> 00:07:21.519 MCP, maybe an AI could tell you how to do something 00:07:21.519 --> 00:07:24.259 complex online. Now it could actually do it for 00:07:24.259 --> 00:07:26.720 you. That's a massive leap in practical automation 00:07:26.720 --> 00:07:29.420 from just advising to actually executing. It 00:07:29.420 --> 00:07:31.660 really is. And that leap brings us to things 00:07:31.660 --> 00:07:33.980 like building AI web apps. You can actually build 00:07:33.980 --> 00:07:36.620 custom -led magnet web apps now. You know, those 00:07:36.620 --> 00:07:38.540 little tools websites use to capture customer 00:07:38.540 --> 00:07:41.639 info. You can build those in minutes using tools 00:07:41.639 --> 00:07:45.139 like ChatGPT or Claude, often with no coding 00:07:45.139 --> 00:07:47.680 required, which means even if you're not a developer, 00:07:47.899 --> 00:07:51.019 you can create these powerful customer -attracting 00:07:51.019 --> 00:07:54.019 tools. lowers the barrier to entry for digital 00:07:54.019 --> 00:07:57.100 business so much. And Google Gemini is also pushing 00:07:57.100 --> 00:07:59.319 workflows forward. They've got this suite of 00:07:59.319 --> 00:08:01.339 like 28 free features. And these aren't just 00:08:01.339 --> 00:08:03.360 minor tricks. They let you automate tasks, analyze 00:08:03.360 --> 00:08:05.959 data, even build working apps pretty quickly. 00:08:06.439 --> 00:08:10.360 Imagine an AI quickly going through a huge spreadsheet 00:08:10.360 --> 00:08:13.579 of customer feedback, identifying the key sentiment 00:08:13.579 --> 00:08:16.120 patterns, and then just automatically drafting 00:08:16.120 --> 00:08:18.220 a summary report for you. It's all about working 00:08:18.220 --> 00:08:21.339 smarter, you know, not necessarily harder. Exactly. 00:08:21.600 --> 00:08:26.040 And here are just a few specific new AI tools 00:08:26.040 --> 00:08:28.360 that caught our eye, really illustrating this 00:08:28.360 --> 00:08:31.560 shift to practical automation. First, there's 00:08:31.560 --> 00:08:35.740 Vibe N8n. This helps you build and tweak N8n 00:08:35.740 --> 00:08:37.539 workflows. Which is that open source automation 00:08:37.539 --> 00:08:40.139 software. Right. But you do it just by prompting. 00:08:40.350 --> 00:08:42.730 Using natural language, like telling your automation 00:08:42.730 --> 00:08:45.009 software what to build instead of clicking through 00:08:45.009 --> 00:08:47.750 endless menus. Then there's Project OS. This 00:08:47.750 --> 00:08:49.690 one helps you get a personalized resume that 00:08:49.690 --> 00:08:51.710 really grabs attention. It doesn't just format. 00:08:51.769 --> 00:08:54.750 It kind of crafts your story. Next up, Buki. 00:08:55.450 --> 00:08:58.110 Plans, writes, and even distributes content that's 00:08:58.110 --> 00:09:00.149 designed to be authentic and convert customers. 00:09:01.049 --> 00:09:04.029 Huge for marketing. Content creation is big. 00:09:04.210 --> 00:09:07.639 And finally, Launch. This one creates fully functional 00:09:07.639 --> 00:09:10.179 apps, but with AI assistance and human support. 00:09:10.740 --> 00:09:13.059 Kind of bridging the gap between just an idea 00:09:13.059 --> 00:09:16.320 and a real deployable product makes app creation 00:09:16.320 --> 00:09:18.720 way more accessible. So looking at all these 00:09:18.720 --> 00:09:20.899 new tools, what's the main thing someone should 00:09:20.899 --> 00:09:22.500 take away if they're thinking about exploring 00:09:22.500 --> 00:09:25.700 them? They really simplify complex jobs. From 00:09:25.700 --> 00:09:28.259 automating workflows to creating content, powerful 00:09:28.259 --> 00:09:31.100 tools are becoming much more accessible. Sponsor. 00:09:31.480 --> 00:09:33.659 Okay, we've got a couple more AI quick hits for 00:09:33.659 --> 00:09:35.620 you before we move on. There's this clever new 00:09:35.620 --> 00:09:37.580 tutorial floating around, shows you how to turn 00:09:37.580 --> 00:09:40.080 your resume into a Netflix -style web page using 00:09:40.080 --> 00:09:43.460 AI. Huh, like browser skills? Kind of. It treats 00:09:43.460 --> 00:09:45.700 your profile like a personalized streaming experience. 00:09:45.820 --> 00:09:47.980 It's creative, right? Speaks to that broader 00:09:47.980 --> 00:09:50.639 trend of AI enabling really customized digital 00:09:50.639 --> 00:09:54.179 experiences. Interesting. And back to privacy 00:09:54.179 --> 00:09:56.860 again for a second. OpenAI recently canceled 00:09:56.860 --> 00:09:58.559 a feature they were working on. Yeah. It would 00:09:58.559 --> 00:10:00.919 let users search through their private GPC chats. 00:10:01.470 --> 00:10:04.590 Oh, wow. Yeah. So canceling it reinforces that 00:10:04.590 --> 00:10:06.490 theme we talked about earlier. Companies are 00:10:06.490 --> 00:10:09.129 really having to react quickly to public and 00:10:09.129 --> 00:10:11.990 regulatory concerns about data access security. 00:10:12.309 --> 00:10:14.549 Shows that constant tension, doesn't it? Between 00:10:14.549 --> 00:10:18.090 convenience and privacy and AI. Definitely. And, 00:10:18.210 --> 00:10:20.169 you know, there's also this growing chatter about 00:10:20.169 --> 00:10:23.470 the real cost of chasing AGI. Artificial General 00:10:23.470 --> 00:10:26.370 Intelligence. Machines with like human level 00:10:26.370 --> 00:10:29.039 thinking. Exactly. And the consensus seems to 00:10:29.039 --> 00:10:31.379 be that power consolidation is just the beginning. 00:10:31.600 --> 00:10:34.659 The sheer resources needed, computation, money 00:10:34.659 --> 00:10:37.980 to even attempt, AGI means maybe only a handful 00:10:37.980 --> 00:10:40.580 of big players can truly compete, raises some 00:10:40.580 --> 00:10:42.419 big questions about who controls that kind of 00:10:42.419 --> 00:10:44.899 power, who gets access. Yeah, deep questions. 00:10:45.139 --> 00:10:47.679 And speaking of resources and expansion, OpenAI 00:10:47.679 --> 00:10:50.620 also announced Stargate Norway. That's going 00:10:50.620 --> 00:10:52.700 to be its first European AI data center, which 00:10:52.700 --> 00:10:55.080 is significant, not just for raw computing power, 00:10:55.299 --> 00:10:58.049 but for things like data sovereignty. Regional 00:10:58.049 --> 00:11:01.049 AI development. Could kickstart more localized 00:11:01.049 --> 00:11:04.490 AI hubs in Europe. And one more quick one. Poe, 00:11:04.490 --> 00:11:06.590 that's the platform letting you use a bunch of 00:11:06.590 --> 00:11:08.470 different AI models. Right, like a buffet of 00:11:08.470 --> 00:11:11.090 AIs. Ha, yeah. They just released a developer 00:11:11.090 --> 00:11:13.950 API. So now developers can programmatically access 00:11:13.950 --> 00:11:15.789 all those different cutting -edge models through 00:11:15.789 --> 00:11:18.710 one single point. Makes it way easier to experiment, 00:11:18.889 --> 00:11:20.909 mix and mash, integrate different AI skills into 00:11:20.909 --> 00:11:23.950 their own apps. Okay, but now here's where things 00:11:23.950 --> 00:11:28.259 get maybe really interesting. Thank you. And 00:11:28.259 --> 00:11:30.700 perhaps a little mind -bending. Are you tired 00:11:30.700 --> 00:11:34.179 of cleaning data sets, painstakingly tuning models, 00:11:34.419 --> 00:11:37.399 debugging those awful CUDA errors yourself? Yeah, 00:11:37.399 --> 00:11:39.399 been there. Imagine an AI that just does all 00:11:39.399 --> 00:11:41.440 that for you. Meet NEO. They're calling it the 00:11:41.440 --> 00:11:44.620 Kaggle Killer AI agent. Yeah, this is genuinely 00:11:44.620 --> 00:11:46.940 remarkable stuff from a company called NeoAI. 00:11:47.159 --> 00:11:49.799 So NEO is an AI agent, but it's actually made 00:11:49.799 --> 00:11:51.919 of 11 specialized sub -agents working together. 00:11:52.000 --> 00:11:54.460 And it's designed meticulously to do basically 00:11:54.460 --> 00:11:56.960 everything a full -stack machine learning engineer 00:11:56.960 --> 00:12:01.460 does. And here's the kicker. Autonomously. Autonomously? 00:12:01.559 --> 00:12:03.580 Yeah. It's like having a whole AI engineering 00:12:03.580 --> 00:12:05.860 team just ready to go. Okay, let's break that 00:12:05.860 --> 00:12:08.399 down. So you give it a problem like, build me 00:12:08.399 --> 00:12:12.200 a predictive model using this data, right? Then 00:12:12.200 --> 00:12:16.100 NEO plans the whole project, writes the code. 00:12:16.610 --> 00:12:19.490 Debugs the code, deploys the solution, all in 00:12:19.490 --> 00:12:22.309 this continuous loop, iterating. And it runs 00:12:22.309 --> 00:12:24.570 in a safe sandbox environment so it can try stuff 00:12:24.570 --> 00:12:27.269 without messing things up. It communicates with 00:12:27.269 --> 00:12:29.309 you through chat. You can jump in, guide it if 00:12:29.309 --> 00:12:31.830 you want, or you can just watch it work, solving 00:12:31.830 --> 00:12:34.090 complex problems by itself. It's pretty incredible 00:12:34.090 --> 00:12:35.730 to think about. Yeah, it's way more than just 00:12:35.730 --> 00:12:38.870 one tool. It's like you took ChatGPT plus AutoML 00:12:38.870 --> 00:12:41.970 that's automated machine learning, simplifies 00:12:41.970 --> 00:12:44.450 building models. Right. Plus DevOps, the whole 00:12:44.450 --> 00:12:46.289 software development and operations pipeline 00:12:46.289 --> 00:12:48.909 and put them all together. But like on steroids, 00:12:49.129 --> 00:12:51.490 it's this integrated intelligence system handling 00:12:51.490 --> 00:12:54.009 the whole workflow. And the performance. This 00:12:54.009 --> 00:12:56.730 is what's truly astonishing. They benchmarked 00:12:56.730 --> 00:13:00.350 NEO across 75 Kaggle competitions. Yeah. Used 00:13:00.350 --> 00:13:03.929 a standard framework, MLE bench. And it didn't 00:13:03.929 --> 00:13:08.169 just like participate. It meddled in 34 .2 % 00:13:08.169 --> 00:13:11.809 of them. Wow. Yeah. Actual competition grade 00:13:11.809 --> 00:13:15.549 performance. often beating human teams. It even 00:13:15.549 --> 00:13:18.330 outperformed other advanced AI agents like RD 00:13:18.330 --> 00:13:22.210 Agent and OpenAI's own aid stack. You know, I 00:13:22.210 --> 00:13:24.289 still wrestle with prompt drift myself sometimes 00:13:24.289 --> 00:13:26.789 where the AI output just kind of changes unexpectedly. 00:13:27.029 --> 00:13:29.149 Oh yeah, frustrating. So an agent that can actually 00:13:29.149 --> 00:13:32.190 course correct autonomously, learn and adapt. 00:13:32.700 --> 00:13:35.139 as it hits problems. That feels like a real game 00:13:35.139 --> 00:13:37.620 changer for the field. And crucially, they built 00:13:37.620 --> 00:13:41.299 in this human in the loop mode, which means you 00:13:41.299 --> 00:13:43.919 can intervene any time. You can tweak its logic, 00:13:44.000 --> 00:13:46.120 give it feedback, add constraints midstream, 00:13:46.200 --> 00:13:48.679 or just chat with it like it's another member 00:13:48.679 --> 00:13:50.940 of the team. It's designed to be fast, iterative, 00:13:51.080 --> 00:13:53.120 and really importantly, it's not a black box. 00:13:53.360 --> 00:13:56.539 Ah, transparency. Good. Yeah, it's... Processes 00:13:56.539 --> 00:13:58.559 are transparent, so you can oversee it, collaborate 00:13:58.559 --> 00:14:00.620 with it. This feels like more than just a neat 00:14:00.620 --> 00:14:03.039 tool, though. It feels like a clear signal. OpenAI, 00:14:03.299 --> 00:14:07.000 Google, Adept. You can bet someone's already 00:14:07.000 --> 00:14:09.120 working hard on their own version of Neo. Oh, 00:14:09.120 --> 00:14:11.519 for sure. The agent wars, where these autonomous 00:14:11.519 --> 00:14:14.179 AI systems compete to solve really complex problems, 00:14:14.419 --> 00:14:17.600 they feel like they're truly just getting started 00:14:17.600 --> 00:14:21.929 now. So thinking about NEO. How does this level 00:14:21.929 --> 00:14:24.830 of autonomy really change the game for human 00:14:24.830 --> 00:14:28.409 ML engineers? It handles the complex, often tedious 00:14:28.409 --> 00:14:31.370 tasks freeing up human engineers for higher level 00:14:31.370 --> 00:14:34.570 innovation and strategy. Okay, let's try and 00:14:34.570 --> 00:14:36.490 bring this all together. What's really clear 00:14:36.490 --> 00:14:38.850 from all this is, wow, the AI landscape is shifting 00:14:38.850 --> 00:14:42.139 fast, almost daily. And in ways that are both, 00:14:42.220 --> 00:14:44.840 you know, subtle and really profound. Yeah, we're 00:14:44.840 --> 00:14:46.360 definitely seeing that big pivot, aren't we? 00:14:46.419 --> 00:14:49.460 From AI as maybe a fun consumer novelty towards 00:14:49.460 --> 00:14:52.399 its role in serious enterprise reliability, with 00:14:52.399 --> 00:14:54.279 players like Claude just quietly gaining massive 00:14:54.279 --> 00:14:56.639 ground in the business world, really challenging 00:14:56.639 --> 00:14:58.600 the established leaders there. Two -sec silence. 00:14:59.460 --> 00:15:02.059 And then beyond just those market battles, these 00:15:02.059 --> 00:15:04.779 AI tools themselves are becoming incredibly practical, 00:15:05.000 --> 00:15:06.639 really accessible. They're empowering pretty 00:15:06.639 --> 00:15:08.659 much anyone, really, regardless of their tech 00:15:08.659 --> 00:15:11.299 background. To automate complex tasks, analyze 00:15:11.299 --> 00:15:13.679 data, even build functional apps without needing 00:15:13.679 --> 00:15:16.539 to write a single line of code. But maybe the 00:15:16.539 --> 00:15:18.419 most profound development, the one that really 00:15:18.419 --> 00:15:22.860 sparks the most curiosity, is the rise of these 00:15:22.860 --> 00:15:26.559 truly autonomous agents. like NEO, systems that 00:15:26.559 --> 00:15:28.779 are now capable of handling complex engineering 00:15:28.779 --> 00:15:32.279 tasks, often with minimal human oversight. This 00:15:32.279 --> 00:15:34.740 feels like where AI really starts to genuinely 00:15:34.740 --> 00:15:37.899 collaborate with us or maybe even lead in the 00:15:37.899 --> 00:15:40.179 act of creation itself. It's an incredibly exciting 00:15:40.179 --> 00:15:42.120 time, isn't it? Just watching these technologies 00:15:42.120 --> 00:15:44.799 grow up, mature from like fascinating concepts 00:15:44.799 --> 00:15:49.059 and to tools that feel indispensable, redefining 00:15:49.059 --> 00:15:51.559 what's even possible. And leaves us with an important 00:15:51.559 --> 00:15:53.600 question, I think, for you, our listener, to 00:15:53.600 --> 00:15:56.039 consider. As AI does become more autonomous, 00:15:56.220 --> 00:15:59.320 more capable, what new forms of human -AI collaboration 00:15:59.320 --> 00:16:01.779 are really going to emerge? And maybe more importantly, 00:16:01.940 --> 00:16:04.360 what skills become the most valuable for us humans 00:16:04.360 --> 00:16:06.740 in this rapidly evolving landscape? Definitely 00:16:06.740 --> 00:16:08.360 something to chew on as you go about your day. 00:16:08.519 --> 00:16:10.379 Thanks so much for joining us for this deep dive 00:16:10.379 --> 00:16:12.360 into the latest in AI. Yeah, we hope you found 00:16:12.360 --> 00:16:14.019 some valuable nuggets in there, something to 00:16:14.019 --> 00:16:15.879 help you stay well -informed and hopefully curious. 00:16:16.279 --> 00:16:17.960 Until next time, out to your own music.