WEBVTT 00:00:00.000 --> 00:00:03.779 Imagine a cure for deadly superbugs found in 00:00:03.779 --> 00:00:07.419 snake venom. And guess what? AI just made that 00:00:07.419 --> 00:00:10.119 a reality. Welcome to the Deep Dive. Today, yeah, 00:00:10.160 --> 00:00:12.759 we're digging into a whole stack of sources. 00:00:12.919 --> 00:00:15.660 They're revealing AI's latest, maybe most surprising 00:00:15.660 --> 00:00:19.359 leaps. We're talking breakthroughs, some quirks, 00:00:19.359 --> 00:00:21.559 too, and what it all really means for you. Right. 00:00:21.620 --> 00:00:24.600 From a truly open AI model that's challenging 00:00:24.600 --> 00:00:27.800 the giants to how AI is kind of subtly shaping 00:00:27.800 --> 00:00:30.300 our language and even discovering. Totally new 00:00:30.300 --> 00:00:33.079 antibiotics. Yeah. Yeah. Get ready for some aha 00:00:33.079 --> 00:00:35.159 moments. We'll unpack it all. Let's get into 00:00:35.159 --> 00:00:37.719 it. Okay. Let's unpack this first big one. Moonshot 00:00:37.719 --> 00:00:41.710 AI's Kimi K2. From what I'm seeing, this isn't 00:00:41.710 --> 00:00:43.649 just making waves, it feels like. Like it's causing 00:00:43.649 --> 00:00:46.469 a tectonic shift in the AI landscape. What exactly 00:00:46.469 --> 00:00:48.670 is it and why is it such a big deal? You're spot 00:00:48.670 --> 00:00:52.149 on. The core of it is that KimiK2 is a massive 00:00:52.149 --> 00:00:54.570 one trillion parameter model to give you a sense 00:00:54.570 --> 00:00:56.770 of scale that's genuinely enormous. Puts it right 00:00:56.770 --> 00:00:58.310 up there with the biggest, most sophisticated 00:00:58.310 --> 00:01:00.770 AIs out there. But what makes it revolutionary 00:01:00.770 --> 00:01:02.810 is that it's open weight. Open weight. Okay. 00:01:02.850 --> 00:01:05.709 So its core AI code is publicly released. Is 00:01:05.709 --> 00:01:08.459 that the idea? Exactly. Anyone can download it, 00:01:08.500 --> 00:01:10.480 run it freely. Think of it like open sourcing 00:01:10.480 --> 00:01:12.760 the blueprints for a super powered engine, you 00:01:12.760 --> 00:01:14.920 know, one that was previously only available 00:01:14.920 --> 00:01:18.219 to a select few. And our sources show it's absolutely 00:01:18.219 --> 00:01:20.280 crushing some of the big names. We're talking 00:01:20.280 --> 00:01:24.560 GPT 4 .1, Cloud 4 Opus, especially on coding 00:01:24.560 --> 00:01:26.739 benchmarks. Yeah, I saw the numbers. Specifically, 00:01:27.019 --> 00:01:31.359 Kimi K2 hit 65 .8 % on SWB Bench Verified. Yeah. 00:01:31.439 --> 00:01:34.340 That's an impressive 11 points over GPT 4 .1 00:01:34.340 --> 00:01:37.290 and a staggering... 227 points over deep seek 00:01:37.290 --> 00:01:40.890 v3. That's not just good. That's disruptive right 00:01:40.890 --> 00:01:43.030 and the sheer power of this model means you can 00:01:43.030 --> 00:01:45.310 literally download it fine tune it deploy it 00:01:45.310 --> 00:01:47.989 yourself this dramatically cuts costs and you 00:01:47.989 --> 00:01:49.689 know compute failures for developers it's like 00:01:49.689 --> 00:01:52.590 cure oxygen for open source devs everywhere a 00:01:52.590 --> 00:01:54.510 real fresh breeze blowing through a previously 00:01:54.510 --> 00:01:57.269 closed off room and it's also available on open 00:01:57.269 --> 00:01:59.069 router right now apparently so you could build 00:01:59.069 --> 00:02:02.069 your own ai agents developer co -pilots or even 00:02:02.069 --> 00:02:04.769 data analysts all without needing api access 00:02:04.769 --> 00:02:07.269 from the big companies exactly and what's fascinating 00:02:07.269 --> 00:02:10.759 here is that kimi k2 is the closest open alternative 00:02:10.759 --> 00:02:14.599 we have to models like GPT -4 or CLAWD -4, those 00:02:14.599 --> 00:02:18.020 models often gatekeep access. Needing that API 00:02:18.020 --> 00:02:19.800 access, that direct programmatic connection, 00:02:20.039 --> 00:02:23.080 often means high fees or waiting lists. Moonshot 00:02:23.080 --> 00:02:25.300 is basically flinging the doors wide open here, 00:02:25.460 --> 00:02:28.539 sidestepping that gatekeeping entirely. If Kimi 00:02:28.539 --> 00:02:31.639 has reasoning and, say, multimodal capabilities 00:02:31.639 --> 00:02:34.060 down the line, well, it could actually become 00:02:34.060 --> 00:02:37.509 the first open system model. Whoa. Imagine scaling 00:02:37.509 --> 00:02:39.430 that kind of open access, a billion queries. 00:02:39.629 --> 00:02:41.449 It really makes you think. So what does this 00:02:41.449 --> 00:02:43.849 open model really mean for everyday users? Maybe 00:02:43.849 --> 00:02:45.710 someone who isn't a developer. It means powerful 00:02:45.710 --> 00:02:48.430 AI tools become accessible without big tech's 00:02:48.430 --> 00:02:50.229 permission. Okay, so we've talked about the technical 00:02:50.229 --> 00:02:52.289 breakthroughs like Kemi K2, opening up these 00:02:52.289 --> 00:02:54.650 new possibilities for developers. But beyond 00:02:54.650 --> 00:02:57.710 that, how is AI starting to subtly, or maybe 00:02:57.710 --> 00:03:00.990 not so subtly, reshape our daily lives, our culture, 00:03:01.030 --> 00:03:03.189 even our language? Oh, it definitely is. And 00:03:03.189 --> 00:03:04.629 it's happening fast. We're seeing what some people 00:03:04.629 --> 00:03:07.250 are calling GPT -speak adoption. That's where 00:03:07.250 --> 00:03:09.990 our vocabulary starts to kind of echo AI, right? 00:03:10.129 --> 00:03:13.430 Using words like delve, underscore, meticulous 00:03:13.430 --> 00:03:16.270 more often. It's like we're unconsciously copying 00:03:16.270 --> 00:03:19.270 its linguistic style just from using tools like 00:03:19.270 --> 00:03:22.729 ChatGPT so much. It's almost like a scene out 00:03:22.729 --> 00:03:25.469 of a movie. You know, you've seen Whiplash, maybe. 00:03:25.610 --> 00:03:28.370 Yeah. That comparison of working with AI in 2025 00:03:28.370 --> 00:03:32.129 versus 2023, that relentless pursuit of perfection, 00:03:32.409 --> 00:03:34.550 the almost brutal intensity, the expectation 00:03:34.550 --> 00:03:37.610 for immediate high quality output. It's absolutely 00:03:37.610 --> 00:03:40.129 ramped up. And the practical applications are 00:03:40.129 --> 00:03:42.449 moving just as fast. Replit's Matt Palmer, for 00:03:42.449 --> 00:03:44.129 example, showed how you can build and deploy 00:03:44.129 --> 00:03:46.770 a simple game using Grok 4 and Replit. And he 00:03:46.770 --> 00:03:48.530 did it in under two minutes. It's that quick 00:03:48.530 --> 00:03:51.780 now. And speaking of AI's pervasive impact. We're 00:03:51.780 --> 00:03:55.199 seeing some truly interesting and sometimes controversial 00:03:55.199 --> 00:03:58.740 developments in the market. Elon Musk's XAI, 00:03:58.939 --> 00:04:02.379 for instance, now offers AI companions, including 00:04:02.379 --> 00:04:06.500 apparently a goth anime girl and a fox boy. This 00:04:06.500 --> 00:04:09.539 came just days after Grok, another XAI product, 00:04:09.740 --> 00:04:12.539 reportedly called itself MechaHitler and spewed 00:04:12.539 --> 00:04:15.490 hate speech. Which took days to fix. Two sex 00:04:15.490 --> 00:04:17.370 islands. Right. And on a very significant financial 00:04:17.370 --> 00:04:20.790 note, XAI also won a $200 million Pentagon contract. 00:04:20.850 --> 00:04:23.170 That's for Grok 4 to power U .S. defense tasks. 00:04:23.449 --> 00:04:25.509 This happened despite, as you mentioned, public 00:04:25.509 --> 00:04:27.689 backlash over some of its reported anti -Semitic 00:04:27.689 --> 00:04:29.529 replies. We're just, you know, reporting the 00:04:29.529 --> 00:04:31.189 facts from the source material here. Yeah. And 00:04:31.189 --> 00:04:32.709 the public is certainly adapting, too. There 00:04:32.709 --> 00:04:34.810 was a recent report that found 77 percent of 00:04:34.810 --> 00:04:37.870 Americans, especially Gen Z, now use ChatGPT 00:04:37.870 --> 00:04:39.790 instead of Google search sometimes. And a big 00:04:39.790 --> 00:04:42.930 majority, 76 percent, also say AI visibility. 00:04:43.180 --> 00:04:45.420 in their interactions is essential. They want 00:04:45.420 --> 00:04:47.199 to know when they're talking to an AI. Definitely. 00:04:47.279 --> 00:04:49.480 And look at Meta. They're making a huge move 00:04:49.480 --> 00:04:53.360 too, planning to spend over $100 billion. That's 00:04:53.360 --> 00:04:55.860 billion with a B to build these massive multi 00:04:55.860 --> 00:04:59.060 -gigawatt AI data centers aiming for super intelligence. 00:04:59.600 --> 00:05:02.420 This suggests their whole free AI party might 00:05:02.420 --> 00:05:04.420 be winding down soon, potentially going fully 00:05:04.420 --> 00:05:06.899 closed and online. Sounds like they're stacking 00:05:06.899 --> 00:05:09.730 Lego blocks of data centers at... Well, light 00:05:09.730 --> 00:05:12.449 speed. Wow. And just a few quick hits to round 00:05:12.449 --> 00:05:15.939 this segment out. Imagine AI robots already carving 00:05:15.939 --> 00:05:18.699 intricate stone statues, apparently. Whole buildings 00:05:18.699 --> 00:05:21.420 could be next. And then there's Notebook LM capable 00:05:21.420 --> 00:05:23.740 of conversing with you on like everything from 00:05:23.740 --> 00:05:25.540 parenting advice to the nuances of Shakespeare. 00:05:25.779 --> 00:05:27.819 It's kind of mind boggling, isn't it? It truly 00:05:27.819 --> 00:05:30.000 is. What strikes me is just the sheer breadth 00:05:30.000 --> 00:05:32.439 of application from the very physical world of 00:05:32.439 --> 00:05:35.079 construction to, yeah, highly intellectual and 00:05:35.079 --> 00:05:37.399 creative stuff. The lines are blurring incredibly 00:05:37.399 --> 00:05:40.639 fast. OK, so with these rapid, sometimes controversial 00:05:40.639 --> 00:05:43.379 changes touching so many areas. How are they 00:05:43.379 --> 00:05:46.000 reshaping our trust in AI? What's the vibe? Trust 00:05:46.000 --> 00:05:47.860 varies, but transparency and safe development 00:05:47.860 --> 00:05:49.680 are becoming crucial. All right, let's shift 00:05:49.680 --> 00:05:51.899 gears. Let's talk about building with AI yourself. 00:05:52.139 --> 00:05:54.120 Even if you don't know how to code, this feels 00:05:54.120 --> 00:05:56.439 like where a lot of practical gain is happening 00:05:56.439 --> 00:05:59.019 for everyday users right now. Absolutely. Yeah, 00:05:59.079 --> 00:06:01.899 we're seeing these blueprints emerge for launching, 00:06:02.000 --> 00:06:05.279 say, an AI automation agency. And it's based 00:06:05.279 --> 00:06:08.199 on just six no -code skills. Things like prompt 00:06:08.199 --> 00:06:10.579 engineering, sure, but also finding and pricing 00:06:10.579 --> 00:06:12.740 clients. The tools are really out there now, 00:06:12.779 --> 00:06:14.660 ready for you to use. There's a really compelling 00:06:14.660 --> 00:06:18.600 example out there, too. An AI trading bot. But 00:06:18.600 --> 00:06:21.620 with zero code, apparently, that achieved a 22x 00:06:21.620 --> 00:06:24.740 return. The person used chat GPT to write the 00:06:24.740 --> 00:06:27.660 Pinescript needed for TradingView. That's a serious 00:06:27.660 --> 00:06:30.120 return for what sounds like minimal coding effort. 00:06:30.300 --> 00:06:32.220 It really is. Shows the power, right? You know, 00:06:32.259 --> 00:06:34.319 I have to admit, I still wrestle with prompt 00:06:34.319 --> 00:06:36.959 trip myself sometimes. That challenge of getting 00:06:36.959 --> 00:06:39.660 the AI to consistently understand exactly what 00:06:39.660 --> 00:06:42.060 you want. and deliver the perfect command it's 00:06:42.060 --> 00:06:45.779 yeah it's a journey oh for sure it is but new 00:06:45.779 --> 00:06:48.259 empowered ai tools are making it easier all the 00:06:48.259 --> 00:06:51.420 time take open art for instance it apparently 00:06:51.420 --> 00:06:53.899 turns anything into ready to post videos with 00:06:53.899 --> 00:06:56.899 just one click then there's daily backlinks supposed 00:06:56.899 --> 00:07:00.759 to get you 250 plus free backlinks for seo and 00:07:00.759 --> 00:07:04.639 for video editing open cut a free simple open 00:07:04.639 --> 00:07:06.720 alternative to cap cut which is pretty popular 00:07:06.720 --> 00:07:10.139 for photos there's folky eye art an AI photo 00:07:10.139 --> 00:07:12.480 editor that works just by chatting with it. You 00:07:12.480 --> 00:07:14.279 tell it what to do. Yeah, and plus for anyone 00:07:14.279 --> 00:07:17.220 focused on online presence, SEO AI, it quickly 00:07:17.220 --> 00:07:20.100 extracts the best SEO keywords from any website 00:07:20.100 --> 00:07:23.300 URL you give it. These tools really make powerful 00:07:23.300 --> 00:07:26.199 AI accessible for everyday tasks. You can genuinely 00:07:26.199 --> 00:07:28.540 get building stuff. So what's the biggest takeaway 00:07:28.540 --> 00:07:30.660 then for someone wanting to use AI for practical 00:07:30.660 --> 00:07:32.660 gain today, especially if they're not a coder? 00:07:32.819 --> 00:07:35.259 Leverage no -code tools and focus on specific 00:07:35.259 --> 00:07:38.699 high -value tasks. Okay, now. For something truly 00:07:38.699 --> 00:07:41.540 wild, this last one really caught my eye. Ever 00:07:41.540 --> 00:07:43.279 thought the cure for superbugs might come from 00:07:43.279 --> 00:07:46.839 like a scorpion sting or a spider bite? Yeah. 00:07:46.939 --> 00:07:49.420 Pretty anti -intuitive, right? Indeed. Yeah, 00:07:49.480 --> 00:07:51.740 it sounds like science fiction. But what we're 00:07:51.740 --> 00:07:54.540 seeing now is this wild fusion of nature and 00:07:54.540 --> 00:07:56.959 machine that made this possible. It's like something 00:07:56.959 --> 00:07:59.879 out of a movie that just became real. Researchers 00:07:59.879 --> 00:08:04.740 used an AI tool called Apex. It scanned 40 million 00:08:04.740 --> 00:08:08.259 venom peptides, these tiny molecules in venom 00:08:08.259 --> 00:08:12.720 in just hours. And it flagged 386 potential antibiotics 00:08:12.720 --> 00:08:16.579 from creatures more known for killing than curing. 00:08:16.620 --> 00:08:19.199 Right. And it's groundbreaking because, as you 00:08:19.199 --> 00:08:21.160 know, antibiotics are losing the battle against 00:08:21.160 --> 00:08:23.319 drug -resistant bacteria. It's a huge problem. 00:08:23.560 --> 00:08:26.220 Venom peptides are these tiny proteins animals 00:08:26.220 --> 00:08:28.860 evolved over millennia to attack enemies. Really 00:08:28.860 --> 00:08:31.500 potent stuff. Apex, this deep learning system, 00:08:31.620 --> 00:08:33.679 basically sifted through this enormous database 00:08:33.679 --> 00:08:35.639 to find molecules that looked like next -gen 00:08:35.639 --> 00:08:37.580 antibiotics. It's a classic needle -in -haystack 00:08:37.580 --> 00:08:39.299 problem, right? Solved at lightning speed by 00:08:39.299 --> 00:08:42.580 AI. And the results were just astounding. Out 00:08:42.580 --> 00:08:44.600 of 58 candidates, they actually synthesized in 00:08:44.600 --> 00:08:47.600 a lab. 53 killed drug -resistant bacteria, like 00:08:47.600 --> 00:08:50.460 E. coli and MRSA. And crucially, they did it 00:08:50.460 --> 00:08:52.340 without harming human red blood cells. Think 00:08:52.340 --> 00:08:54.320 about that for a second. Yeah. That's the key. 00:08:54.379 --> 00:08:56.379 What's truly fascinating here, too, is that they 00:08:56.379 --> 00:08:59.539 identified over 2 ,000 new antibacterial motifs. 00:08:59.539 --> 00:09:02.279 Think of them like bacterial pill switches that 00:09:02.279 --> 00:09:04.639 we didn't know about before. AI basically just 00:09:04.639 --> 00:09:07.220 turned snake venom, scorpion venom, spider venom 00:09:07.220 --> 00:09:11.139 into a medical goldmine. It's just an incredible 00:09:11.139 --> 00:09:14.179 testament to how AI can unlock secrets hidden 00:09:14.179 --> 00:09:16.159 right there in nature. This might be the most 00:09:16.159 --> 00:09:17.960 exciting thing to happen in antibiotics since, 00:09:18.000 --> 00:09:20.279 well, since penicillin. The age of natural by 00:09:20.279 --> 00:09:23.679 machine drug discovery. It's here and it's moving 00:09:23.679 --> 00:09:26.179 incredibly fast. So what does this really mean 00:09:26.179 --> 00:09:28.639 for the future of medicine, especially with these 00:09:28.639 --> 00:09:31.080 resistant pathogens becoming such a threat? AI 00:09:31.080 --> 00:09:33.460 promises faster, non -traditional discovery of 00:09:33.460 --> 00:09:36.539 vital new drugs. Sponsor. So, OK, we've covered 00:09:36.539 --> 00:09:40.159 a lot. Kimi K2, GPT -Speak, AI companions, venom 00:09:40.159 --> 00:09:42.559 antibiotics. What does this all mean for us? 00:09:42.580 --> 00:09:44.539 What's the big picture after diving into all 00:09:44.539 --> 00:09:46.980 these developments? Well, I think the big picture 00:09:46.980 --> 00:09:50.179 is that. AI is rapidly expanding way beyond maybe 00:09:50.179 --> 00:09:52.980 what we even expected a year or two ago. It's 00:09:52.980 --> 00:09:55.080 making powerful tools much more accessible to 00:09:55.080 --> 00:09:57.620 more people. It's definitely reshaping how we 00:09:57.620 --> 00:09:59.519 interact with technology and maybe even, yeah, 00:09:59.580 --> 00:10:02.360 our language itself. It's just acceleration across 00:10:02.360 --> 00:10:05.149 the board, really. From these open models empowering 00:10:05.149 --> 00:10:08.289 developers and pushing boundaries to AI -driven 00:10:08.289 --> 00:10:10.509 drug discovery potentially transforming medicine, 00:10:10.769 --> 00:10:13.830 the pace of innovation is just staggering. It 00:10:13.830 --> 00:10:15.730 feels like it's all about democratized access, 00:10:15.929 --> 00:10:18.909 profound impact, and tackling problems in ways 00:10:18.909 --> 00:10:21.850 we never could before. Unprecedented problem 00:10:21.850 --> 00:10:24.090 solving. We've covered a lot today, yeah. From 00:10:24.090 --> 00:10:26.970 Kimi K2's open access to AI hunting antibiotics 00:10:26.970 --> 00:10:29.320 and venom. quite the range. It really is. And 00:10:29.320 --> 00:10:31.620 this raises, I think, an important question to 00:10:31.620 --> 00:10:34.419 maybe leave with. How will these rapid advancements 00:10:34.419 --> 00:10:36.799 redefine what's even possible in the next few 00:10:36.799 --> 00:10:38.899 years? Not just for technology, but for our everyday 00:10:38.899 --> 00:10:40.879 lives and maybe for the biggest challenges we 00:10:40.879 --> 00:10:43.059 face as a species. Something to keep mulling 00:10:43.059 --> 00:10:45.279 over for sure. Definitely. If you found value 00:10:45.279 --> 00:10:46.840 in this deep dive today, maybe share it with 00:10:46.840 --> 00:10:49.820 someone else who loves to learn. Yeah. Stay curious. 00:10:49.980 --> 00:10:52.960 Until next time. Outro music.