1 00:00:00,000 --> 00:00:00,960  Welcome to Crash News.  2 00:00:00,960 --> 00:00:03,760  Today is May 15th, 2025.  3 00:00:03,760 --> 00:00:05,720  Ever feel like you're just scrolling endlessly  4 00:00:05,720 --> 00:00:08,480  through news headlines, kind of drowning in it all,  5 00:00:08,480 --> 00:00:09,760  especially with AI?  6 00:00:09,760 --> 00:00:10,580  - Oh, absolutely.  7 00:00:10,580 --> 00:00:11,600  It's moving so fast.  8 00:00:11,600 --> 00:00:12,440  - Exactly.  9 00:00:12,440 --> 00:00:14,900  So today, we're trying to cut through that noise.  10 00:00:14,900 --> 00:00:18,380  Think of this as your essential AI briefing.  11 00:00:18,380 --> 00:00:19,960  We're gonna unpack the big stuff,  12 00:00:19,960 --> 00:00:22,120  explain what it means, why you should care.  13 00:00:22,120 --> 00:00:23,600  - Yeah, not just skimming the surface.  14 00:00:23,600 --> 00:00:26,280  We'll dive into the sources, connect the dots,  15 00:00:26,280 --> 00:00:29,020  make it clear and hopefully engaging.  16 00:00:29,020 --> 00:00:31,280  - We'll break down any tech terms along the way too.  17 00:00:31,280 --> 00:00:33,260  By the end, you should have a really solid grasp  18 00:00:33,260 --> 00:00:34,760  of what's important right now.  19 00:00:34,760 --> 00:00:37,160  - And it's fascinating the sheer pace of what's happening.  20 00:00:37,160 --> 00:00:39,000  Some really foundational shifts.  21 00:00:39,000 --> 00:00:40,200  - Okay, where should we start then?  22 00:00:40,200 --> 00:00:43,720  - Let's kick off with Google DeepMind's AlphaVolve.  23 00:00:43,720 --> 00:00:46,160  - AlphaVolve, okay, sounds significant.  24 00:00:46,160 --> 00:00:46,980  What is it?  25 00:00:46,980 --> 00:00:48,280  - Well, basically it's a coding agent.  26 00:00:48,280 --> 00:00:50,760  It's powered by Google's Gemini AI.  27 00:00:50,760 --> 00:00:52,440  But here's the key thing.  28 00:00:52,440 --> 00:00:54,920  It's not just suggesting, you know, little code fixes.  29 00:00:54,920 --> 00:00:56,280  - Ah, okay.  30 00:00:56,280 --> 00:00:57,140  So it's doing more.  31 00:00:57,140 --> 00:00:57,980  - Much more.  32 00:00:57,980 --> 00:01:00,200  - It's main job is to iteratively generate  33 00:01:00,200 --> 00:01:03,160  and actually improve entire algorithm solutions.  34 00:01:03,160 --> 00:01:05,020  - Wow, so it's not just tweaking,  35 00:01:05,020 --> 00:01:08,280  it's building from the ground up almost,  36 00:01:08,280 --> 00:01:10,160  like designing a whole new program.  37 00:01:10,160 --> 00:01:13,120  - Exactly, think of it less like a code spell checker  38 00:01:13,120 --> 00:01:15,800  and more like an AI system architect.  39 00:01:15,800 --> 00:01:17,480  - Okay, that's a big claim.  40 00:01:17,480 --> 00:01:18,720  How does it actually do that?  41 00:01:18,720 --> 00:01:21,960  - Well, the process is pretty clever.  42 00:01:21,960 --> 00:01:25,700  AlphaVolve first proposes a full algorithm solution.  43 00:01:25,700 --> 00:01:28,420  Then it uses these automated evaluators,  44 00:01:28,420 --> 00:01:30,500  you can think of them as AI judges,  45 00:01:30,500 --> 00:01:34,360  basically to test how well the code performs objectively.  46 00:01:34,360 --> 00:01:35,800  - Right, so it gets feedback.  47 00:01:35,800 --> 00:01:36,860  - Precisely.  48 00:01:36,860 --> 00:01:38,580  And based on those test results,  49 00:01:38,580 --> 00:01:40,900  AlphaVolve learns which approaches work best  50 00:01:40,900 --> 00:01:42,480  and then builds on them in the next round.  51 00:01:42,480 --> 00:01:44,560  It keeps refining, iterating, improving.  52 00:01:44,560 --> 00:01:46,080  - And this isn't just theoretical, is it?  53 00:01:46,080 --> 00:01:47,320  Like, is it being used?  54 00:01:47,320 --> 00:01:48,240  - Oh yeah, absolutely.  55 00:01:48,240 --> 00:01:50,880  It's already having a real impact inside Google.  56 00:01:50,880 --> 00:01:54,320  people, if you look at their core infrastructure, they're seeing tangible benefits.  57 00:01:54,320 --> 00:01:55,320  Like what?  58 00:01:55,320 --> 00:01:58,400  We're talking about improvements in data center efficiency, you know, those huge places  59 00:01:58,400 --> 00:02:04,520  that use tons of power, also advances in designing their computer chips and even making the process  60 00:02:04,520 --> 00:02:06,960  of training other AI models more efficient.  61 00:02:06,960 --> 00:02:07,960  Okay.  62 00:02:07,960 --> 00:02:12,240  Those sound like fundamental improvements, things that could save energy, boost performance  63 00:02:12,240 --> 00:02:13,240  across the board.  64 00:02:13,240 --> 00:02:14,240  Exactly.  65 00:02:14,240 --> 00:02:15,240  That's the kind of foundational stuff we're talking about.  66 00:02:15,240 --> 00:02:21,520  Now, I did see a number mentioned, something about a 75% success rate on math problems.  67 00:02:21,520 --> 00:02:22,940  Why is that a big deal?  68 00:02:22,940 --> 00:02:24,240  Ah, yes.  69 00:02:24,240 --> 00:02:28,940  That 75% figure on quite complex math problems is really interesting.  70 00:02:28,940 --> 00:02:33,560  It shows AlphaVolve's ability to reason and problem solve at a pretty high level.  71 00:02:33,560 --> 00:02:36,260  It's not just recalling solutions it's seen before.  72 00:02:36,260 --> 00:02:37,800  It's actually figuring things out.  73 00:02:37,800 --> 00:02:38,800  It seems so.  74 00:02:38,800 --> 00:02:42,940  It shows it can independently discover effective strategies for new challenges.  75 00:02:42,940 --> 00:02:48,600  And what's maybe even more impressive is that in 20% of those successful cases, AlphaVolve  76 00:02:48,600 --> 00:02:52,100  actually found solutions that were better than the known best ones.  77 00:02:52,100 --> 00:02:53,760  Better than human design solutions.  78 00:02:53,760 --> 00:02:56,400  In some cases, yes, improved upon them.  79 00:02:56,400 --> 00:02:57,600  But I guess it has its limits, right?  80 00:02:57,600 --> 00:03:00,560  It's not about to write the next great novel or compose music.  81 00:03:00,560 --> 00:03:01,880  Hey, no, you're right.  82 00:03:01,880 --> 00:03:05,460  That hits on a key limitation currently.  83 00:03:05,460 --> 00:03:09,480  AlphaVolve really shines on problems where the solution's success can be automatically  84 00:03:09,480 --> 00:03:11,180  and objectively measured.  85 00:03:11,180 --> 00:03:15,320  Think numerical problems, tasks with clear performance goals.  86 00:03:15,320 --> 00:03:20,460  It's less effective right now when dealing with things that are more subjective, you  87 00:03:20,460 --> 00:03:24,400  know, non-numerical issues where you need human judgment to say if it's good.  88 00:03:24,400 --> 00:03:25,400  Right.  89 00:03:25,400 --> 00:03:26,400  Makes sense.  90 00:03:26,400 --> 00:03:27,400  Powerful, but in specific areas.  91 00:03:27,400 --> 00:03:28,400  Yeah.  92 00:03:28,400 --> 00:03:29,400  Okay.  93 00:03:29,400 --> 00:03:30,400  Let's shift gears.  94 00:03:30,400 --> 00:03:31,400  Another big name.  95 00:03:31,400 --> 00:03:32,400  OpenAI.  96 00:03:32,400 --> 00:03:34,880  They've just put new GPT models into ChatGPT itself.  97 00:03:34,880 --> 00:03:35,880  That's right.  98 00:03:35,880 --> 00:03:41,080  GPT 4.1 and GPT 4.1 mini are now available directly within ChatGPT.  99 00:03:41,080 --> 00:03:42,400  - What are the main upgrades there?  100 00:03:42,400 --> 00:03:43,240  What's new?  101 00:03:43,240 --> 00:03:44,600  - Well, GPT 4.1 is faster,  102 00:03:44,600 --> 00:03:46,400  which makes using it feel smoother.  103 00:03:46,400 --> 00:03:49,320  It's also significantly better at programming tasks,  104 00:03:49,320 --> 00:03:51,320  which is a big plus for developers.  105 00:03:51,320 --> 00:03:52,720  And maybe the headline feature,  106 00:03:52,720 --> 00:03:56,320  it supports a 1 million token context window.  107 00:03:56,320 --> 00:03:58,440  - A million tokens, that sounds huge.  108 00:03:58,440 --> 00:03:59,280  Can you break that down?  109 00:03:59,280 --> 00:04:00,320  What does that mean practically?  110 00:04:00,320 --> 00:04:01,760  - Yeah, it's a massive amount.  111 00:04:01,760 --> 00:04:04,040  Think of a token as roughly a word  112 00:04:04,040 --> 00:04:05,860  or maybe a piece of code.  113 00:04:05,860 --> 00:04:09,200  So a million tokens means the AI can process  114 00:04:09,200 --> 00:04:12,480  and remember context from an enormous chunk of information  115 00:04:12,480 --> 00:04:14,680  like the equivalent of a pretty hefty book  116 00:04:14,680 --> 00:04:16,440  or a very large code base all at once.  117 00:04:16,440 --> 00:04:18,760  - So it can handle much more complex tasks,  118 00:04:18,760 --> 00:04:21,200  keep track of longer conversations or documents.  119 00:04:21,200 --> 00:04:23,040  - Exactly, deeper understanding,  120 00:04:23,040 --> 00:04:26,320  more coherent responses over long interactions.  121 00:04:26,320 --> 00:04:30,340  And this GPT 4.1, it's available for ChatGPT Plus,  122 00:04:30,340 --> 00:04:32,160  Pro and Team subscribers.  123 00:04:32,160 --> 00:04:35,280  - Okay, and what about the mini version, GPT 4.1 mini?  124 00:04:35,280 --> 00:04:36,620  - Right, the mini is important  125 00:04:36,620 --> 00:04:38,620  because it's actually the new default model  126 00:04:38,620 --> 00:04:40,660  for all ChatGPT users now.  127 00:04:40,660 --> 00:04:42,300  That includes people on the free plan.  128 00:04:42,300 --> 00:04:43,980  - Oh, interesting, so everyone gets an upgrade.  129 00:04:43,980 --> 00:04:47,020  - Pretty much, it replaces the older GPT 4.0 mini  130 00:04:47,020 --> 00:04:49,100  and OpenAI says both these new models  131 00:04:49,100 --> 00:04:50,520  perform better across the board  132 00:04:50,520 --> 00:04:52,000  compared to the previous versions.  133 00:04:52,000 --> 00:04:54,320  - It's interesting they were API only first.  134 00:04:54,320 --> 00:04:56,240  Why add them to ChatGPT now?  135 00:04:56,240 --> 00:04:57,680  Was it user demand?  136 00:04:57,680 --> 00:05:00,560  - It really seems like user feedback played a big role.  137 00:05:00,560 --> 00:05:02,780  Putting these models directly into ChatGPT  138 00:05:02,780 --> 00:05:05,780  makes them accessible to way more people.  139 00:05:05,780 --> 00:05:07,680  While APIs are great for developers  140 00:05:07,680 --> 00:05:09,320  building specific things,  141 00:05:09,320 --> 00:05:11,280  this move caters to everyday users  142 00:05:11,280 --> 00:05:12,360  wanting better performance  143 00:05:12,360 --> 00:05:14,600  right in the chat interface they already use.  144 00:05:14,600 --> 00:05:17,000  It sort of democratizes access to the newer tech.  145 00:05:17,000 --> 00:05:18,160  - Makes total sense,  146 00:05:18,160 --> 00:05:20,840  putting the power right there for everyone.  147 00:05:20,840 --> 00:05:22,700  Okay, let's switch domains again.  148 00:05:22,700 --> 00:05:24,720  Audio, Stability AI,  149 00:05:24,720 --> 00:05:27,580  and something called Stable Audio Open Small.  150 00:05:27,580 --> 00:05:30,680  - Yes, Stability AI has open sourced this model,  151 00:05:30,680 --> 00:05:31,520  Stable Audio.  152 00:05:31,560 --> 00:05:35,360  open-small. Open-sourced, meaning the code is out there for anyone. Exactly.  153 00:05:35,360 --> 00:05:40,120  Free to use, modify, build upon. It's a text-to-audio model, so you give it text,  154 00:05:40,120 --> 00:05:44,000  it generates audio. Like sound effects music. It can generate short audio clips,  155 00:05:44,000 --> 00:05:48,360  sound effects, maybe simple musical elements based on text descriptions. It's  156 00:05:48,360 --> 00:05:53,800  a relatively small model, architecturally speaking, 341 million parameters. Which  157 00:05:53,800 --> 00:05:57,240  sounds like a lot, but I guess it's small in the AI world. Compared to the huge  158 00:05:57,240 --> 00:06:00,960  language models, yes, it's quite compact, but it's surprisingly efficient.  159 00:06:00,960 --> 00:06:06,400  Efficient? How? Like fast. Yeah, very fast. It can apparently generate an 11-second  160 00:06:06,400 --> 00:06:11,360  audio clip on a regular smartphone in under 8 seconds. On a phone, wow. Right.  161 00:06:11,360 --> 00:06:16,400  That kind of speed and ability to run locally on mobile devices really  162 00:06:16,400 --> 00:06:20,440  highlights its efficiency. It lowers the barrier for creating audio content quite  163 00:06:20,440 --> 00:06:24,780  a bit. Very cool. Okay, moving from creation to safety. That's always a huge  164 00:06:24,780 --> 00:06:30,000  topic. OpenAI launched a public safety metrics hub. Yes, and this is quite  165 00:06:30,000 --> 00:06:35,600  significant I think. It's a step towards more transparency. The hub's purpose is to  166 00:06:35,600 --> 00:06:41,000  publicly track how OpenAI's models are performing on key safety benchmarks. What  167 00:06:41,000 --> 00:06:44,880  kind of benchmarks are we talking about? Things like measuring hallucinations, you  168 00:06:44,880 --> 00:06:47,960  know, and the AI just confidently makes things up. Right, that's a big one.  169 00:06:47,960 --> 00:06:52,120  Definitely. Also how well the model refuses to generate harmful or  170 00:06:52,120 --> 00:06:56,360  inappropriate content. Its resistance to jailbreaking, which is people trying to  171 00:06:56,360 --> 00:07:01,040  trick it into bypassing safety rules. Okay. And also just its general ability  172 00:07:01,040 --> 00:07:04,240  to follow the instructions it's given accurately and safely. So it's like a  173 00:07:04,240 --> 00:07:08,160  public dashboard for AI safety performance. Allows researchers, the  174 00:07:08,160 --> 00:07:12,000  public, everyone to see the progress. Exactly. It provides a public record,  175 00:07:12,000 --> 00:07:16,440  allows for more scrutiny, and helps people understand the ongoing safety  176 00:07:16,440 --> 00:07:21,680  work. It's a positive move I'd say. Good to hear. Now beyond those big headlines, it  177 00:07:21,680 --> 00:07:24,040  feels like  178 00:07:24,040 --> 00:07:26,900  there's just a constant flood of other AI news. Our source has mentioned a few interesting  179 00:07:26,900 --> 00:07:27,900  bits and pieces.  180 00:07:27,900 --> 00:07:35,460  Yeah, the landscape is just incredibly busy. One area getting more focus is the broader  181 00:07:35,460 --> 00:07:38,880  societal impacts of AI, what people call second order effects.  182 00:07:38,880 --> 00:07:40,120  The indirect consequences.  183 00:07:40,120 --> 00:07:45,200  Right. How AI is changing things in ways we didn't immediately expect. Also, seeing companies  184 00:07:45,200 --> 00:07:50,700  like Klarna, the finance tech firm, adjusting their AI strategy that might hint at broader  185 00:07:50,700 --> 00:07:56,280  industry trends or challenges as companies figure out how to actually use AI effectively  186 00:07:56,280 --> 00:07:57,280  and profitably.  187 00:07:57,280 --> 00:07:59,320  Hmm, interesting. What else?  188 00:07:59,320 --> 00:08:02,960  Well, there was a meta-analysis mentioned looking at lots of studies suggesting chat  189 00:08:02,960 --> 00:08:07,800  GPT can be quite effective in education, especially for problem-based learning, getting students  190 00:08:07,800 --> 00:08:08,800  engaged.  191 00:08:08,800 --> 00:08:10,460  Okay, so potential in the classroom.  192 00:08:10,460 --> 00:08:12,800  Seems like it. And then there's the practical engineering side too.  193 00:08:12,800 --> 00:08:14,800  Right, like building actual tools.  194 00:08:14,800 --> 00:08:19,520  Exactly. Deploying smaller models like Gemini Nano directly onto phones to help fight scams  195 00:08:19,520 --> 00:08:21,640  in real time, that's a direct benefit for users.  196 00:08:21,640 --> 00:08:23,240  Yeah, definitely useful.  197 00:08:23,240 --> 00:08:28,260  Also work on AI news tools to help us filter information and open source code editors like  198 00:08:28,260 --> 00:08:31,320  Void which help the developer community itself innovate.  199 00:08:31,320 --> 00:08:35,420  It really is everywhere. And there were some super quickfire announcements too, just highlighting  200 00:08:35,420 --> 00:08:36,420  the speed.  201 00:08:36,420 --> 00:08:42,080  Oh yeah. Things like that huge five plus billion dollar AI partnership announced in Saudi Arabia  202 00:08:42,080 --> 00:08:44,540  shows massive global investment.  203 00:08:44,540 --> 00:08:45,540  Wow.  204 00:08:45,540 --> 00:08:49,440  Then discussions about risks with things like the model context protocol ongoing work on  205 00:08:49,440 --> 00:08:55,080  security and that stat for Microsoft. AI writing up to 30% of their code now.  206 00:08:55,080 --> 00:08:59,200  30%. That's a lot. Shows how integrated it's becoming in actual development.  207 00:08:59,200 --> 00:09:04,600  Absolutely. And then just the rapid pace stuff, discounts on AI events, PayPal popping up  208 00:09:04,600 --> 00:09:09,640  in AI search results, chatbots having glitches sometimes reminding us it's still evolving.  209 00:09:09,640 --> 00:09:12,640  And of course, new models like GPT 4.1.  210 00:09:12,640 --> 00:09:14,500  instantly appearing, it never stops.  211 00:09:14,500 --> 00:09:17,000  - Okay, that's a lot on the development side.  212 00:09:17,000 --> 00:09:18,700  For listeners looking for something practical,  213 00:09:18,700 --> 00:09:20,300  maybe something they can use today,  214 00:09:20,300 --> 00:09:22,980  we found an interesting AI tutorial tip.  215 00:09:22,980 --> 00:09:24,840  It's about deciding what to watch.  216 00:09:24,840 --> 00:09:26,540  Tell us about this PIX tool.  217 00:09:26,540 --> 00:09:28,060  - Ah, PIX from Likewise.  218 00:09:28,060 --> 00:09:29,100  Yeah, this is a neat one.  219 00:09:29,100 --> 00:09:31,620  It's basically an AI tool designed purely  220 00:09:31,620 --> 00:09:33,580  for entertainment recommendations.  221 00:09:33,580 --> 00:09:36,540  Books, podcasts, TV shows, movies.  222 00:09:36,540 --> 00:09:39,020  It learns your tastes and suggests things you might like.  223 00:09:39,020 --> 00:09:40,660  - Okay, sounds useful.  224 00:09:40,660 --> 00:09:41,760  How do people access it?  225 00:09:41,760 --> 00:09:42,600  Is it an app?  226 00:09:42,600 --> 00:09:43,960  - You can use it on their website  227 00:09:43,960 --> 00:09:45,880  or sign up for email suggestions,  228 00:09:45,880 --> 00:09:49,120  but maybe the easiest way is just texting the word PIX,  229 00:09:49,120 --> 00:09:51,960  P-I-X to the number five hour zero five 50.  230 00:09:51,960 --> 00:09:53,480  - Texting, that's pretty simple.  231 00:09:53,480 --> 00:09:54,680  - Yeah, very accessible.  232 00:09:54,680 --> 00:09:55,520  - Could you give an example?  233 00:09:55,520 --> 00:09:56,880  Like how would you ask it for something?  234 00:09:56,880 --> 00:09:58,680  - Sure, let's say you want a scary movie,  235 00:09:58,680 --> 00:10:00,500  but you hate slasher films.  236 00:10:00,500 --> 00:10:02,040  You could text something like,  237 00:10:02,040 --> 00:10:04,740  send me some scary movies on Prime or Hulu,  238 00:10:04,740 --> 00:10:06,340  no slashers please.  239 00:10:06,340 --> 00:10:08,080  - And it understands that.  240 00:10:08,080 --> 00:10:10,420  The platforms, the no slashers part.  241 00:10:10,420 --> 00:10:11,520  - That's the idea.  242 00:10:11,520 --> 00:10:14,360  - The AI is meant to parse those details,  243 00:10:14,360 --> 00:10:17,360  genre, platforms, specific exclusions,  244 00:10:17,360 --> 00:10:19,040  and give you tailored recommendations  245 00:10:19,040 --> 00:10:20,400  based on what's available.  246 00:10:20,400 --> 00:10:22,600  And the best part, it's free.  247 00:10:22,600 --> 00:10:24,400  Just designed to help with that,  248 00:10:24,400 --> 00:10:26,560  what should I watch tonight problem.  249 00:10:26,560 --> 00:10:28,060  - Okay, I might actually try that one.  250 00:10:28,060 --> 00:10:30,000  Solves a common frustration.  251 00:10:30,000 --> 00:10:32,760  - We also, a quick list of some other new AI tools  252 00:10:32,760 --> 00:10:33,600  just popping up.  253 00:10:33,600 --> 00:10:36,200  - Yeah, just to give a flavor of the variety out there.  254 00:10:36,200 --> 00:10:37,860  There's one called Intellisay,  255 00:10:37,860 --> 00:10:40,380  that's focused on voice commands for productivity.  256 00:10:40,380 --> 00:10:42,080  Another called WhisperFlow,  257 00:10:42,080 --> 00:10:43,780  designed to help you write faster,  258 00:10:43,780 --> 00:10:46,000  kind of integrating into your workflow.  259 00:10:46,000 --> 00:10:48,580  Then Greta, which aims to let people deploy  260 00:10:48,580 --> 00:10:50,260  full web applications quickly,  261 00:10:50,260 --> 00:10:52,200  even without deep coding skills.  262 00:10:52,200 --> 00:10:54,880  - Interesting, democratizing development.  263 00:10:54,880 --> 00:10:56,960  - Potentially, there's also VidMay,  264 00:10:56,960 --> 00:10:59,900  which helps creators scale content on TikTok and Instagram,  265 00:10:59,900 --> 00:11:01,600  using AI to generate stuff  266 00:11:01,600 --> 00:11:03,600  that looks like user generated content.  267 00:11:03,600 --> 00:11:04,120  Okay.  268 00:11:04,120 --> 00:11:09,320  And lastly, Whisperlarge V3, a new version of OpenAI's transcription model, offering  269 00:11:09,320 --> 00:11:13,280  supposedly simple one-click transcription for long audio files.  270 00:11:13,280 --> 00:11:17,920  Could be great for researchers, journalists, anyone dealing with lots of recordings.  271 00:11:17,920 --> 00:11:20,000  Lots of tools emerging constantly.  272 00:11:20,000 --> 00:11:25,360  Okay, wrapping up our dive into the sources, there are also some broader findings and resources  273 00:11:25,360 --> 00:11:26,360  mentioned.  274 00:11:26,360 --> 00:11:27,360  What stood out?  275 00:11:27,360 --> 00:11:28,360  A few things.  276 00:11:28,360 --> 00:11:33,100  For listeners wanting to go deeper technically, Stanford's CS229 course on building large  277 00:11:33,100 --> 00:11:35,840  language models was highlighted as a great resource.  278 00:11:35,840 --> 00:11:36,840  Good tip.  279 00:11:36,840 --> 00:11:41,540  Also, the observation that chat GPT written posts are becoming really common on places  280 00:11:41,540 --> 00:11:45,840  like Reddit that raises, you know, interesting questions about online content and authenticity.  281 00:11:45,840 --> 00:11:48,120  Yeah, how do we know it's human anymore?  282 00:11:48,120 --> 00:11:49,120  Exactly.  283 00:11:49,120 --> 00:11:52,720  Then there was a note about job growth in big tech being kind of stagnant for the last  284 00:11:52,720 --> 00:11:53,800  three years.  285 00:11:53,800 --> 00:11:56,460  That could signal shifts in the industry.  286 00:11:56,460 --> 00:11:59,960  And AI tools being tested to gauge public opinion on things.  287 00:11:59,960 --> 00:12:04,940  The example was Botox, but it shows AI moving into market research and sentiment analysis.  288 00:12:04,940 --> 00:12:07,300  And any overarching trends or maybe concerns.  289 00:12:07,300 --> 00:12:12,020  Well, the trend of AI getting baked into existing platforms continues, like YouTube planning  290 00:12:12,020 --> 00:12:14,880  to use Gemini for contextual ads and videos.  291 00:12:14,880 --> 00:12:16,980  That could really change online advertising.  292 00:12:16,980 --> 00:12:20,260  Ads tailored by AI understanding the video content.  293 00:12:20,260 --> 00:12:21,260  Precisely.  294 00:12:21,260 --> 00:12:27,360  Also, Microsoft testing a "Hey" co-pilot, WakeWord for Windows 11, making AI assistance  295 00:12:27,360 --> 00:12:30,720  even more integrated into the basic computer experience.  296 00:12:30,720 --> 00:12:31,720  Always listening.  297 00:12:31,720 --> 00:12:33,240  Potentially, yes.  298 00:12:33,240 --> 00:12:37,000  And Perplexity partnering with PayPal for shopping within the chat interface, blurring  299 00:12:37,000 --> 00:12:39,720  the lines between search, AI, and e-commerce.  300 00:12:39,720 --> 00:12:40,720  Right.  301 00:12:40,720 --> 00:12:44,560  We already mentioned OpenAI's transparency hub for safety, which is positive.  302 00:12:44,560 --> 00:12:51,780  And new models keep coming, like Alibaba releasing a video generation model called 1.2.1VACE.  303 00:12:51,780 --> 00:12:53,740  the progress in multimedia generation.  304 00:12:53,740 --> 00:12:57,960  is just relentless. Okay, so pulling this all together, what does this actually mean for you  305 00:12:57,960 --> 00:13:03,000  lifting right now? We've covered a huge range today. AI designing algorithms, making audio.  306 00:13:03,000 --> 00:13:07,400  Powering better chatbots, helping you pick a movie. Right. The pace is just incredible,  307 00:13:07,400 --> 00:13:11,720  isn't it? And understanding these key pieces seems more and more crucial just to navigate,  308 00:13:11,720 --> 00:13:16,680  well, everything. What's really striking is how interconnected it all is. Improvements in core  309 00:13:16,680 --> 00:13:21,560  AI research feed into practical tools, which then change how we work, how we get information,  310 00:13:21,560 --> 00:13:27,400  even how we entertain ourselves. It's weaving itself deeper into daily life. Exactly. And that  311 00:13:27,400 --> 00:13:32,520  leads to the big question, doesn't it? As AI gets more integrated, everywhere from the hidden  312 00:13:32,520 --> 00:13:39,240  infrastructure to the apps on our phones, how do we make sure its development, its deployment,  313 00:13:39,240 --> 00:13:45,080  actually aligns with what we value as a society with our goals? It's a huge question. No easy  314 00:13:45,080 --> 00:13:50,440  answers. Definitely something worth pondering as you see these new AI features pop up everywhere.  315 00:13:50,440 --> 00:13:55,960  It's not just about the cool new tech, but the deeper impact. Yeah, maybe something for you to  316 00:13:55,960 --> 00:14:01,720  think about next time you use an AI tool. What's it doing for you? But also, how might it be subtly  317 00:14:01,720 --> 00:14:07,480  shaping things, your choices, the information you see? Good point. Well, thanks for taking this deep  318 00:14:07,480 --> 00:14:11,960  dive with us today. Stay curious, keep learning. We really encourage that. And we'll catch you next  319 00:14:11,960 --> 00:14:16,040  time with more insights from The Cutting Edge. And hey, if you found this valuable, if you learned  320 00:14:16,040 --> 00:14:20,600  something, the best free way to support us is to rate, like, and subscribe on whatever platform you  321 00:14:20,600 --> 00:14:24,600  use. It really helps others find the show. Yeah, and join the conversation. We have a growing  322 00:14:24,600 --> 00:14:29,080  Discord community. The link's usually in the show notes. Absolutely. And for those who want to go  323 00:14:29,080 --> 00:14:33,640  even deeper and help us keep doing these deep dives, you might consider becoming a Patreon  324 00:14:33,640 --> 00:14:39,080  supporter. We really see this as more than just news. It's about learning together in this rapidly  325 00:14:39,080 --> 00:14:43,800  changing field. Couldn't agree more. Your support makes it possible. Thanks again for listening.