WEBVTT 00:00:00.000 --> 00:00:02.120 You know, when you look across the current AI 00:00:02.120 --> 00:00:05.120 landscape, the story has almost always been bigger 00:00:05.120 --> 00:00:07.400 is better. Right. We're kind of constantly waiting 00:00:07.400 --> 00:00:10.720 for the next massive foundational model. But 00:00:10.720 --> 00:00:13.519 what happens when that narrative just flips entirely? 00:00:14.179 --> 00:00:16.800 What if a lightweight model, something like 25 00:00:16.800 --> 00:00:19.739 times cheaper to run than the big guys, actually... 00:00:20.170 --> 00:00:23.410 beats them decisively on key tasks it just turns 00:00:23.410 --> 00:00:25.170 everything upside down doesn't it that whole 00:00:25.170 --> 00:00:28.030 paradox speed versus scale performance versus 00:00:28.030 --> 00:00:30.370 price that's exactly what we're going to crack 00:00:30.370 --> 00:00:33.570 open today welcome back to the deep dive we've 00:00:33.570 --> 00:00:36.369 sifted through a pile the latest most disruptive 00:00:36.369 --> 00:00:38.590 ai intelligence and we're here to give you the 00:00:38.590 --> 00:00:40.810 fastest way to get you know, truly up to speed. 00:00:40.950 --> 00:00:42.850 Yeah, our mission is really to filter out all 00:00:42.850 --> 00:00:44.890 that noise. We'll start with that shocking underdog 00:00:44.890 --> 00:00:48.090 model, Grok4Fast, and how it's kind of rewriting 00:00:48.090 --> 00:00:50.429 the economics of deploying this stuff. And then 00:00:50.429 --> 00:00:52.149 we'll pivot to the strategic side, things like 00:00:52.149 --> 00:00:54.770 major universities adopting ChatGPT and some 00:00:54.770 --> 00:00:57.770 pretty crucial policy shifts that, frankly, demand 00:00:57.770 --> 00:01:00.250 immediate action from you, especially about your 00:01:00.250 --> 00:01:03.159 data. And finally, we hit what feels like the 00:01:03.159 --> 00:01:05.299 ultimate reality check for the whole industry 00:01:05.299 --> 00:01:08.280 right now. We dive into this new gold standard 00:01:08.280 --> 00:01:11.280 for testing AI coding agents, a really tough 00:01:11.280 --> 00:01:14.980 benchmark called SWE Bench Pro. And the results 00:01:14.980 --> 00:01:18.120 are, well, pretty sobering. Reminds us how far 00:01:18.120 --> 00:01:20.959 AI still has to go on its own. So let's start 00:01:20.959 --> 00:01:23.480 the deep dive. Okay, so XAI just dropped Grok 00:01:23.480 --> 00:01:26.599 4 fast. Now, they position it as this lightweight, 00:01:26.859 --> 00:01:29.040 cost -efficient follow -up to their main Grok 00:01:29.040 --> 00:01:32.060 4 model. But honestly. It's way more than just 00:01:32.060 --> 00:01:34.560 a footnote. It immediately blasted into the global 00:01:34.560 --> 00:01:37.019 top 10 on the benchmarks. That's a huge statement 00:01:37.019 --> 00:01:38.819 right out of the gate. The efficiency numbers 00:01:38.819 --> 00:01:40.700 are just staggering. Our sources are confirming 00:01:40.700 --> 00:01:43.340 it runs roughly 25 times cheaper than competitors 00:01:43.340 --> 00:01:46.480 like Gemari 2 .5 Pro. This isn't just tweaking. 00:01:46.500 --> 00:01:48.599 It's fundamental. It's using way fewer computational 00:01:48.599 --> 00:01:50.780 resources. Like, for example, it only needed 00:01:50.780 --> 00:01:53.620 about 61 million tokens for its standard benchmark 00:01:53.620 --> 00:01:56.579 runs. The full Grok 4. That used 120 million. 00:01:56.819 --> 00:02:00.340 Wow. And that huge drop in tokens means less 00:02:00.340 --> 00:02:02.480 processing, right, which translates directly 00:02:02.480 --> 00:02:05.640 into lower costs and much faster answers. This 00:02:05.640 --> 00:02:07.760 is the kind of model that makes AI suddenly seem 00:02:07.760 --> 00:02:10.199 viable for, you know, every small team, every 00:02:10.199 --> 00:02:12.479 startup with an idea. But the real story, I think, 00:02:12.500 --> 00:02:14.539 is the performance it keeps, despite being so 00:02:14.539 --> 00:02:17.419 lightweight. This thing is, well, it's a powerhouse. 00:02:17.520 --> 00:02:19.419 It shot straight to number one in web search 00:02:19.419 --> 00:02:21.719 performance on the El Marina search arena. That's 00:02:21.719 --> 00:02:24.000 a really competitive space. Yeah, that's impressive 00:02:24.000 --> 00:02:25.979 for sure. But here's where the tech side gets 00:02:25.979 --> 00:02:28.770 really interesting. It beat both the older Grok 00:02:28.770 --> 00:02:32.629 4 and GPT -5 on those complex live code bench 00:02:32.629 --> 00:02:35.430 tests. And look at the math scores, too. It's 00:02:35.430 --> 00:02:38.009 competing where AI usually, you know, struggles 00:02:38.009 --> 00:02:40.030 with actual reasoning, not just crunching numbers. 00:02:40.189 --> 00:02:45.580 Getting 92 % on AIM 2025, 93 .3 % on HMMT. For 00:02:45.580 --> 00:02:47.860 people who don't know, AIME and HMMT are tough 00:02:47.860 --> 00:02:50.379 math competitions. They test real problem solving, 00:02:50.520 --> 00:02:53.039 not just calculation, scoring that high. That 00:02:53.039 --> 00:02:55.219 means serious reasoning ability. Right. That 00:02:55.219 --> 00:02:57.699 level of consistency search coding complex math 00:02:57.699 --> 00:02:59.740 is kind of astonishing for a model that's 25 00:02:59.740 --> 00:03:02.009 times cheaper to run. And the way they achieve 00:03:02.009 --> 00:03:04.270 this efficiency is fascinating. They built this 00:03:04.270 --> 00:03:06.849 unified architecture, right? It runs in two modes. 00:03:07.050 --> 00:03:10.590 A thinking mode for the hard stuff, complex reasoning, 00:03:10.830 --> 00:03:14.189 and a fast mode for quick, simple answers. But 00:03:14.189 --> 00:03:16.770 it's smarter than just two gears. It seems like 00:03:16.770 --> 00:03:20.060 it uses tokens dynamically. The system only kicks 00:03:20.060 --> 00:03:22.560 into that expensive thinking mode if its confidence 00:03:22.560 --> 00:03:25.419 dips below some threshold, so it doesn't waste 00:03:25.419 --> 00:03:28.139 processing power on easy questions. That's the 00:03:28.139 --> 00:03:30.800 really clever part, yeah. This is a direct shot 00:03:30.800 --> 00:03:33.460 at the OpenAI pricing wall, basically. It's offering 00:03:33.460 --> 00:03:36.430 Claude or Gemini -level power. But at a price 00:03:36.430 --> 00:03:38.729 that makes advanced AI suddenly accessible for 00:03:38.729 --> 00:03:41.449 like indie developers, it really shifts the economic 00:03:41.449 --> 00:03:43.389 barrier for anyone wanting to build something 00:03:43.389 --> 00:03:45.930 new. So bottom line on the economics. Cheaper, 00:03:45.930 --> 00:03:48.189 faster models make advanced AI accessible for 00:03:48.189 --> 00:03:50.860 small teams and new products. Definitely. OK, 00:03:50.919 --> 00:03:53.099 let's pivot a bit from deployment economics to 00:03:53.099 --> 00:03:56.039 the bigger strategic picture and policy shifts. 00:03:56.259 --> 00:03:58.460 We're seeing mass adoption now in institutions. 00:03:58.900 --> 00:04:01.379 Look at Oxford University first in the UK to 00:04:01.379 --> 00:04:03.900 give all students and staff free access to chat 00:04:03.900 --> 00:04:06.960 GPT edu. That's huge. Yeah, because institutional 00:04:06.960 --> 00:04:09.439 adoption makes it normal. Right. Not just some 00:04:09.439 --> 00:04:11.860 tool for students to cheat, but like an actual 00:04:11.860 --> 00:04:14.800 part of professional work of education. It's 00:04:14.800 --> 00:04:17.240 a big stamp of approval. And on the really practical 00:04:17.240 --> 00:04:20.069 personal level. Good prompting is becoming a 00:04:20.069 --> 00:04:22.709 real skill, almost like a financial skill. We 00:04:22.709 --> 00:04:24.829 saw this amazing story. A traveler used just 00:04:24.829 --> 00:04:28.470 eight specific prompts, basically turned ChatGPT 00:04:28.470 --> 00:04:32.490 into their travel agent, cut an $879 flight down 00:04:32.490 --> 00:04:37.069 to $299. No reward points involved. See? Prompt 00:04:37.069 --> 00:04:39.430 engineering isn't just a buzzword anymore. It's 00:04:39.430 --> 00:04:42.029 a real -world money saver. Ask the right questions, 00:04:42.189 --> 00:04:45.139 save hundreds of dollars. Simple as that. That 00:04:45.139 --> 00:04:46.980 said, with all this AI getting integrated everywhere, 00:04:47.160 --> 00:04:49.259 we absolutely have to talk privacy. Data sharing 00:04:49.259 --> 00:04:52.240 is about to change pretty significantly for millions 00:04:52.240 --> 00:04:54.360 of people. LinkedIn is going to start sharing 00:04:54.360 --> 00:04:56.899 user profiles, activity data directly with Microsoft. 00:04:56.959 --> 00:04:59.120 That data is for AI training and ad targeting. 00:04:59.279 --> 00:05:02.000 And it starts November 3rd. And here's the critical 00:05:02.000 --> 00:05:04.300 part for you listening right now. The default 00:05:04.300 --> 00:05:08.529 setting is on in. It's opt out, not opt in. So 00:05:08.529 --> 00:05:10.550 if you care about your professional data feeding 00:05:10.550 --> 00:05:13.310 into these big AI systems, you have to actively 00:05:13.310 --> 00:05:16.209 go into your settings and turn it off. Like today. 00:05:17.350 --> 00:05:19.410 Meanwhile, the money keeps pouring into the hardware 00:05:19.410 --> 00:05:22.310 side. We noted that $10 billion U .S. government 00:05:22.310 --> 00:05:24.970 allocation, it's specifically to make sure America 00:05:24.970 --> 00:05:27.949 stays dominant in chip manufacturing, a really 00:05:27.949 --> 00:05:30.290 strategic investment for the future AI supply 00:05:30.290 --> 00:05:33.569 chain. And the VC money. It's flowing deep and 00:05:33.569 --> 00:05:35.730 focused. Glilak Capital just raised half a billion 00:05:35.730 --> 00:05:39.389 dollars, $500 million, specifically for AI and 00:05:39.389 --> 00:05:42.490 cybersecurity startups based in Israel. The big 00:05:42.490 --> 00:05:44.449 money is laser focused on those two areas right 00:05:44.449 --> 00:05:46.829 now. So drilling down on these shifts, what's 00:05:46.829 --> 00:05:49.069 the single biggest action a user needs to take 00:05:49.069 --> 00:05:51.470 right now? Users must opt out of LinkedIn's new 00:05:51.470 --> 00:05:53.709 default settings if they care about data sharing. 00:05:54.079 --> 00:05:56.100 No question. Let's talk about a common problem 00:05:56.100 --> 00:05:58.879 in this AI gold rush right now. It's this issue 00:05:58.879 --> 00:06:01.060 of vibe coding. You know, developers rushing 00:06:01.060 --> 00:06:03.019 to build something just because the tech is cool 00:06:03.019 --> 00:06:05.100 without checking if anyone actually needs it 00:06:05.100 --> 00:06:07.500 first. Yeah, we got to move past building AI 00:06:07.500 --> 00:06:10.279 just because the demo looks slick, right? That's 00:06:10.279 --> 00:06:12.620 why our sources are really emphasizing a more 00:06:12.620 --> 00:06:15.199 structured approach. Exactly. They lay out this. 00:06:15.639 --> 00:06:18.500 three -stage framework for successful AI building. 00:06:18.779 --> 00:06:21.079 And it stresses that the absolute first step 00:06:21.079 --> 00:06:23.959 has to be user -first research. Validate the 00:06:23.959 --> 00:06:26.779 product idea. Don't even start the MVP until 00:06:26.779 --> 00:06:29.620 you know the problem is real and, you know, people 00:06:29.620 --> 00:06:31.740 will actually pay for a solution. Well, while 00:06:31.740 --> 00:06:33.360 you're building, remember the whole distribution 00:06:33.360 --> 00:06:36.459 game is changing, mostly thanks to Google's big 00:06:36.459 --> 00:06:40.259 AI push. The old search model is shifting. Businesses 00:06:40.259 --> 00:06:42.180 now have to figure out how to deal with conversational 00:06:42.180 --> 00:06:45.100 ads and these external AI search tools. that 00:06:45.100 --> 00:06:47.139 just bypass Google's traditional results page 00:06:47.139 --> 00:06:49.540 entirely. Right, but that shift also creates 00:06:49.540 --> 00:06:51.560 new openings. There's a great practical guide 00:06:51.560 --> 00:06:53.740 we saw detailing like seven different ways people 00:06:53.740 --> 00:06:56.000 are making money right now using free AI tools. 00:06:56.079 --> 00:06:58.680 It covers everything from creating unique digital 00:06:58.680 --> 00:07:01.060 products to offering super specialized freelance 00:07:01.060 --> 00:07:03.000 services. And for the creative folks out there, 00:07:03.040 --> 00:07:05.899 we're seeing some really cool specific uses for 00:07:05.899 --> 00:07:09.329 AI image generation, especially in video. Think 00:07:09.329 --> 00:07:11.110 about systems that can keep a character looking 00:07:11.110 --> 00:07:14.350 consistent across tons of animated frames. Or 00:07:14.350 --> 00:07:17.069 projects animating famous old paintings for, 00:07:17.189 --> 00:07:19.709 like educational videos. It's getting surprisingly 00:07:19.709 --> 00:07:22.129 detailed. So if that framework is the guide, 00:07:22.410 --> 00:07:25.610 what's the absolute crucial first step for building 00:07:25.610 --> 00:07:28.529 a successful AI product now? The crucial first 00:07:28.529 --> 00:07:31.089 stage is user -first research to validate the 00:07:31.089 --> 00:07:33.939 product idea. Period. All right. Let's hit some 00:07:33.939 --> 00:07:35.699 quick industry shifts. We're seeing this push 00:07:35.699 --> 00:07:39.319 towards decentralization. Maybe Meta's letting 00:07:39.319 --> 00:07:41.939 outside developers build apps for its smart glasses 00:07:41.939 --> 00:07:44.199 now. That opens up a whole new playground for 00:07:44.199 --> 00:07:46.939 wearable AI stuff. And pressure on platforms 00:07:46.939 --> 00:07:49.300 is everywhere. You know, formerly Twitter is 00:07:49.300 --> 00:07:51.839 reportedly moving towards a highly customizable 00:07:51.839 --> 00:07:54.879 AI driven algorithm for the main feed. It feels 00:07:54.879 --> 00:07:56.779 like personalized AI feeds are just becoming 00:07:56.779 --> 00:07:59.379 the expected standard. And all these new AIs, 00:07:59.480 --> 00:08:01.620 the big ones, the fast ones like Grok we talked 00:08:01.620 --> 00:08:03.730 about. They're putting serious competitive heat 00:08:03.730 --> 00:08:05.709 on the big tech companies. Their whole strategy 00:08:05.709 --> 00:08:07.990 has to adapt faster than ever. And honestly, 00:08:08.250 --> 00:08:11.209 just trying to keep up with this constant rapid 00:08:11.209 --> 00:08:13.709 change is exhausting, even for those of us watching 00:08:13.709 --> 00:08:16.589 it closely every day. Yeah, I'll admit it. I 00:08:16.589 --> 00:08:18.750 still wrestle with prompt drift myself. Yeah. 00:08:19.069 --> 00:08:21.269 You know, when these algorithms just shift under 00:08:21.269 --> 00:08:22.930 your feet unexpectedly, you think you've nailed 00:08:22.930 --> 00:08:25.750 a perfect prompt one day, and the next, the output's 00:08:25.750 --> 00:08:27.509 totally different. You're constantly having to 00:08:27.509 --> 00:08:32.059 relearn the tools. And that pace. Well, it brings 00:08:32.059 --> 00:08:34.379 risks we just can't ignore. We've seen some really 00:08:34.379 --> 00:08:37.539 serious misuse fears surface lately. Reports 00:08:37.539 --> 00:08:41.059 of chatbots generating child abuse images, for 00:08:41.059 --> 00:08:43.500 instance, that sparked huge safety concerns, 00:08:43.600 --> 00:08:46.019 rightly so. The industry has got to manage this 00:08:46.019 --> 00:08:48.600 dark side along with the innovation. And finally, 00:08:48.700 --> 00:08:51.200 there's this sense that the sheer speed of the 00:08:51.200 --> 00:08:53.799 gold rush is creating tunnel vision at the very 00:08:53.799 --> 00:08:56.940 top. Sources are suggesting CEOs are just so 00:08:56.940 --> 00:08:59.480 incredibly bullish on AI succeeding and the immediate 00:08:59.480 --> 00:09:01.860 money it brings that they're reportedly ignoring 00:09:01.860 --> 00:09:03.399 pretty much everything else in their business 00:09:03.399 --> 00:09:05.299 plans. It's like an all consuming single focus. 00:09:05.559 --> 00:09:08.159 So what does that CEO bullishness, that intense 00:09:08.159 --> 00:09:10.360 focus, tell us about where the market's priorities 00:09:10.360 --> 00:09:13.539 are right now? The market focus is singularly 00:09:13.539 --> 00:09:16.500 locked onto AI, overshadowing other traditional 00:09:16.500 --> 00:09:20.370 risks. It's AI or nothing, it seems. Okay. This 00:09:20.370 --> 00:09:22.830 brings us to maybe the most important reality 00:09:22.830 --> 00:09:25.509 check we have today. For a while now, we've been 00:09:25.509 --> 00:09:27.870 looking at these AI benchmarks showing models 00:09:27.870 --> 00:09:31.629 getting like 70, 80, 90 % on coding tasks. And 00:09:31.629 --> 00:09:33.370 it kind of creates this false impression that 00:09:33.370 --> 00:09:36.129 AI is almost ready to just replace human engineer. 00:09:36.350 --> 00:09:39.549 That perception is dangerous. And this new benchmark. 00:09:40.529 --> 00:09:43.370 SWB Bench Pro totally changes the math on that. 00:09:43.590 --> 00:09:46.590 So SWB Bench Pro is being called the new gold 00:09:46.590 --> 00:09:48.769 standard. What makes it different is it doesn't 00:09:48.769 --> 00:09:50.830 just check if the model spits out code that looks 00:09:50.830 --> 00:09:53.149 right. It actually verifies if the model solves 00:09:53.149 --> 00:09:55.809 complex real world coding problems. And if it 00:09:55.809 --> 00:09:58.190 applies the fixes correctly within existing code 00:09:58.190 --> 00:10:00.710 bases, it really raises the bar for saying something 00:10:00.710 --> 00:10:02.649 actually works. And when they tested the top 00:10:02.649 --> 00:10:05.309 models against this tougher standard, whoa, the 00:10:05.309 --> 00:10:07.250 gap between what we thought they could do and 00:10:07.250 --> 00:10:09.190 what they actually do in the real world became. 00:10:09.389 --> 00:10:12.149 painfully obvious. It was, frankly, a bloodbath 00:10:12.149 --> 00:10:14.309 compared to the old benchmarks. Get these numbers. 00:10:14.350 --> 00:10:16.590 The top models, the ones we use every day, scored 00:10:16.590 --> 00:10:19.809 way lower than anyone thought. OpenAI's GPT -5, 00:10:19.990 --> 00:10:23.590 only 23 .3 % on the public tasks. Claude Opus, 00:10:23.610 --> 00:10:27.529 4 .1. 23 .1%. And GPT -4, which everyone was 00:10:27.529 --> 00:10:30.909 raving about, it dropped to just 4 .9 % on those 00:10:30.909 --> 00:10:34.440 public tasks. 4 .9%. Oof. And when they tested 00:10:34.440 --> 00:10:36.240 on private commercial code, which is usually 00:10:36.240 --> 00:10:38.500 more complex, more context -dependent, GPT -5 00:10:38.500 --> 00:10:42.580 dropped even further, down to 14 .9%. These scores 00:10:42.580 --> 00:10:44.620 just lay bare the massive difference between 00:10:44.620 --> 00:10:46.480 being a helpful coding assistant, which they 00:10:46.480 --> 00:10:49.360 are, and being a true autonomous coding agent 00:10:49.360 --> 00:10:51.779 that can handle production work. We're not there 00:10:51.779 --> 00:10:54.559 yet. The findings just undeniable, really. Yes, 00:10:54.620 --> 00:10:56.240 the top models still do better than the rest. 00:10:56.320 --> 00:10:58.399 There's consistency there. But that performance 00:10:58.399 --> 00:11:01.090 gap. Between the AI dream and production reality, 00:11:01.470 --> 00:11:04.049 it's now clear as day. SWB Bench Pro basically 00:11:04.049 --> 00:11:06.470 gives us the map. It shows exactly how far these 00:11:06.470 --> 00:11:08.470 agents still need to travel before they can reliably 00:11:08.470 --> 00:11:10.470 handle professional software development. So 00:11:10.470 --> 00:11:13.029 the immediate practical takeaway for any company 00:11:13.029 --> 00:11:14.850 out there thinking about, you know, letting their 00:11:14.850 --> 00:11:18.090 developers go, don't fire your engineers. AI 00:11:18.090 --> 00:11:20.450 still needs significant development before it 00:11:20.450 --> 00:11:22.570 can tackle complex production code on its own. 00:11:22.929 --> 00:11:26.629 Simple as that. Hashtag tag tag outro. So it's 00:11:26.629 --> 00:11:29.370 been a deep dive into these two sort of opposing 00:11:29.370 --> 00:11:31.990 but equally vital realities today, hasn't it? 00:11:32.029 --> 00:11:34.350 On one side, you've got this radical efficiency 00:11:34.350 --> 00:11:37.909 leap with Grok 4 Fast. It shows that small, cheap, 00:11:38.029 --> 00:11:40.769 smart systems can absolutely beat the big, expensive 00:11:40.769 --> 00:11:42.789 ones sometimes. Right. And on the other side, 00:11:42.809 --> 00:11:46.110 you have the necessary sobering reality check 00:11:46.110 --> 00:11:49.470 from SWE Bench Pro, proving that the AI coding 00:11:49.470 --> 00:11:52.149 tools we rely on, they're still failing most 00:11:52.149 --> 00:11:54.639 of the time on the... really complex, verifiable 00:11:54.639 --> 00:11:56.720 stuff needed for professional work. The future 00:11:56.720 --> 00:11:58.700 isn't just going to be about who has the biggest 00:11:58.700 --> 00:12:01.100 model. It's about mastering the economics of 00:12:01.100 --> 00:12:04.080 deployment and using targeted intelligence smartly. 00:12:04.139 --> 00:12:07.919 Whoa. Just imagine scaling that Grok 4 fast efficiency, 00:12:08.200 --> 00:12:10.440 right? Where you're paying maybe 25 times less 00:12:10.440 --> 00:12:13.529 for each query. Imagine scaling that. to handle 00:12:13.529 --> 00:12:16.850 like a billion complex user interactions every 00:12:16.850 --> 00:12:19.750 single day. That kind of shift in profit and 00:12:19.750 --> 00:12:22.450 accessibility, it completely changes the business 00:12:22.450 --> 00:12:25.059 model for AI products. Yeah, I think the real 00:12:25.059 --> 00:12:27.460 competitive edge in the next year or so won't 00:12:27.460 --> 00:12:30.460 be just having the biggest model. It'll be mastering 00:12:30.460 --> 00:12:32.779 the economics of deploying the lightest, most 00:12:32.779 --> 00:12:35.899 targeted models that can still deliver top performance 00:12:35.899 --> 00:12:38.179 where it really matters. So maybe your homework 00:12:38.179 --> 00:12:40.500 from this deep dive is twofold. First, seriously, 00:12:40.700 --> 00:12:43.179 go check your LinkedIn settings right now. Make 00:12:43.179 --> 00:12:44.940 sure your data preferences are where you want 00:12:44.940 --> 00:12:47.200 them. And second, start thinking about those 00:12:47.200 --> 00:12:49.879 efficiency gains, that idea delivering premium 00:12:49.879 --> 00:12:52.759 results at indie level cost. That feels like 00:12:52.759 --> 00:12:55.230 the blueprint. the next wave of successful AI 00:12:55.230 --> 00:12:57.409 products. Until next time keep digging into the 00:12:57.409 --> 00:12:57.610 details.