WEBVTT 00:00:00.000 --> 00:00:02.680 You ask a simple question, and then you just 00:00:02.680 --> 00:00:05.660 sit there. You stare blankly at the screen. Yeah, 00:00:05.900 --> 00:00:08.119 it's an incredibly frustrating experience. One 00:00:08.119 --> 00:00:12.320 second passes, then two, then five. By the time 00:00:12.320 --> 00:00:14.660 it finally answers, your thought is completely 00:00:14.660 --> 00:00:16.960 gone. Right, your creative momentum just completely 00:00:16.960 --> 00:00:19.710 dies in that silent gap. It really is a momentum 00:00:19.710 --> 00:00:21.670 killer for anyone trying to work quickly. Well, 00:00:21.670 --> 00:00:24.410 welcome to today's deep dive. We're exploring 00:00:24.410 --> 00:00:27.410 a major shift in artificial intelligence today. 00:00:27.570 --> 00:00:30.289 We are looking really closely at Google Gemini 00:00:30.289 --> 00:00:33.890 3 .1 Flash. Our exact mission today is to explore 00:00:33.890 --> 00:00:37.189 a specific technical breakthrough. We want to 00:00:37.189 --> 00:00:39.689 understand how this model completely eliminates 00:00:39.689 --> 00:00:42.070 that awkward pause. Yeah, and we have a really 00:00:42.070 --> 00:00:44.009 exciting roadmap planned out for you. We're going 00:00:44.009 --> 00:00:46.429 to unpack how it processes complex information 00:00:46.429 --> 00:00:49.450 instantly. It handles text, voice, images, and 00:00:49.450 --> 00:00:52.250 video in real time. We will also explore how 00:00:52.250 --> 00:00:55.149 its vision solves physical world problems. Plus, 00:00:55.369 --> 00:00:57.270 we'll show you something truly wild later on. 00:00:57.399 --> 00:01:00.240 You can build a custom voice assistant without 00:01:00.240 --> 00:01:02.679 a single line of code. It represents a completely 00:01:02.679 --> 00:01:05.219 different way to interact with machines. It feels 00:01:05.219 --> 00:01:07.719 so much more like a collaboration than a simple 00:01:07.719 --> 00:01:10.739 query. So let's start by looking deeply at the 00:01:10.739 --> 00:01:14.040 underlying problem of latency. Latency is basically 00:01:14.040 --> 00:01:16.780 that frustrating gap between speaking and getting 00:01:16.780 --> 00:01:19.799 an answer. I always felt like older AI models 00:01:19.799 --> 00:01:23.260 were basically clunky walkie talkies. You push 00:01:23.260 --> 00:01:26.129 a button, you speak your piece. and you wait. 00:01:26.250 --> 00:01:28.349 Yeah, you're just waiting for the digital static 00:01:28.349 --> 00:01:30.769 to finally clear. Right, and it completely ruins 00:01:30.769 --> 00:01:33.269 the natural flow of human conversation. The rhythm 00:01:33.269 --> 00:01:37.030 is entirely broken. But Gemini 3 .1 Flash feels 00:01:37.030 --> 00:01:39.370 like a seamless phone call. The back and forth 00:01:39.370 --> 00:01:41.810 flows much more smoothly and naturally. The core 00:01:41.810 --> 00:01:44.810 design goal was dramatically reducing that exact 00:01:44.810 --> 00:01:47.629 latency gap. It strikes a beautiful balance between 00:01:47.629 --> 00:01:50.189 speed and raw reasoning capability. It's very 00:01:50.189 --> 00:01:53.530 light, very quick, and still remarkably capable. 00:01:53.769 --> 00:01:56.930 It stops feeling like you are querying a sterile 00:01:56.930 --> 00:01:58.790 distant database. It actually starts feeling 00:01:58.790 --> 00:02:00.930 like you are talking to someone in the room. 00:02:01.230 --> 00:02:03.310 Let's look at the impartial data from the recent 00:02:03.310 --> 00:02:06.150 benchmark testing. Right, we have the big bench 00:02:06.150 --> 00:02:09.680 audio benchmark from artificial analysis. Gemini 00:02:09.680 --> 00:02:14.379 3 .1 Flash Live scored 95 .9 % on speech reasoning. 00:02:14.639 --> 00:02:17.360 That is a very impressive number for this specific 00:02:17.360 --> 00:02:20.240 logic test. It places it just behind Step Audio 00:02:20.240 --> 00:02:23.800 R1 .1 in the overall rankings. But it leaves 00:02:23.800 --> 00:02:26.199 other major models well behind it in the dust. 00:02:26.500 --> 00:02:30.400 Yeah, GBT Realtime scored a much lower 83 .3 00:02:30.400 --> 00:02:33.520 % on this benchmark. And the older Gemini 2 .5 00:02:33.520 --> 00:02:37.139 Flash Native Audio hit 90 .7%. To put that in 00:02:37.139 --> 00:02:39.780 perspective, the leap in accuracy is highly significant. 00:02:39.939 --> 00:02:42.460 It means the model makes far fewer logical errors 00:02:42.460 --> 00:02:45.360 when listening. Speed is great, but it also handles 00:02:45.360 --> 00:02:48.199 complex external tasks beautifully. It fully 00:02:48.199 --> 00:02:50.259 supports something called function calling mid 00:02:50.259 --> 00:02:52.300 -conversation. And for those newer to the space, 00:02:52.439 --> 00:02:54.080 we should clarify that term. Function calling 00:02:54.080 --> 00:02:57.370 means... The AI safely uses outside tools to 00:02:57.370 --> 00:02:59.530 do tasks for you. It might check your calendar, 00:03:00.229 --> 00:03:02.189 or it might search the web directly to find live 00:03:02.189 --> 00:03:04.590 information. Doing that quickly usually breaks 00:03:04.590 --> 00:03:06.889 the memory of an AI system. But its memory is 00:03:06.889 --> 00:03:08.909 significantly better than previous software versions. 00:03:09.129 --> 00:03:12.370 The complex Funkbench audio test absolutely proves 00:03:12.370 --> 00:03:14.930 this massive architectural improvement. It scored 00:03:14.930 --> 00:03:18.490 a very impressive 90 .8 % accuracy on function 00:03:18.490 --> 00:03:21.539 calling. The older Gemini 2 .5 versions were 00:03:21.539 --> 00:03:23.819 way behind that specific mark. Right, they only 00:03:23.819 --> 00:03:28.460 scored 71 .5 % and 66 .0 % on the exact same 00:03:28.460 --> 00:03:30.780 test. That is the difference between a novelty 00:03:30.780 --> 00:03:33.360 and a reliable tool. You actually need that high 00:03:33.360 --> 00:03:36.139 accuracy to trust the system entirely. Exactly. 00:03:36.379 --> 00:03:38.780 It handles complex tools without losing the thread 00:03:38.780 --> 00:03:40.819 of the conversation. So we have to look at how 00:03:40.819 --> 00:03:43.599 that impacts long chats. But does moving that 00:03:43.599 --> 00:03:47.280 fast mean it forgets what we just said? No. Upgraded 00:03:47.280 --> 00:03:50.300 memory keeps your entire long conversation perfectly 00:03:50.300 --> 00:03:52.639 on track. That brings us to a really fascinating 00:03:52.639 --> 00:03:55.259 philosophical and technical hurdle. We really 00:03:55.259 --> 00:03:57.500 need to talk about the messy reality of human 00:03:57.500 --> 00:04:00.120 speech. Human communication is incredibly chaotic 00:04:00.120 --> 00:04:02.400 when you actually analyze the audio. We simply 00:04:02.400 --> 00:04:05.080 do not speak in perfectly formed, complete sentences. 00:04:05.259 --> 00:04:07.919 We hesitate, we pause, we leave half -finished 00:04:07.919 --> 00:04:10.539 thoughts dangling in the air constantly. Older 00:04:10.539 --> 00:04:12.479 models dealt with this by using a very rigid 00:04:12.479 --> 00:04:15.159 pipeline. They usually just read a sterile text 00:04:15.159 --> 00:04:17.699 transcript of your voice. The system translated 00:04:17.699 --> 00:04:20.579 your audio into text before the brain saw it. 00:04:20.660 --> 00:04:23.079 And in that translation process, you lose so 00:04:23.079 --> 00:04:26.160 much vital context. They completely missed the 00:04:26.160 --> 00:04:28.740 subtle nuance of human communication. A sigh 00:04:28.740 --> 00:04:31.980 becomes nothing. A long pause is just deleted 00:04:31.980 --> 00:04:36.899 entirely. But Gemini 3 .1 Flash actually processes 00:04:36.899 --> 00:04:39.319 the raw audio natively. It doesn't need a text 00:04:39.319 --> 00:04:40.860 transcript to understand what you're saying. 00:04:41.079 --> 00:04:43.740 It analyzes the actual sound waves in real time. 00:04:43.980 --> 00:04:46.680 It hears the specific tone and feeling behind 00:04:46.680 --> 00:04:48.879 your spoken words. It knows when you are actually 00:04:48.879 --> 00:04:51.040 done talking to it. It also knows when you are 00:04:51.040 --> 00:04:55.250 just pausing to think. Beat. That makes the exchange 00:04:55.250 --> 00:04:58.189 feel far less robotic and much more intuitive. 00:04:58.430 --> 00:05:00.550 It essentially reads the room and responds to 00:05:00.550 --> 00:05:02.589 your emotional state. If you sound frustrated 00:05:02.589 --> 00:05:05.389 or confused, it picks up on that instantly. That 00:05:05.389 --> 00:05:08.230 is a massive leap forward for user experience. 00:05:08.370 --> 00:05:10.850 So how does the model actually alter its response 00:05:10.850 --> 00:05:13.189 based on emotion? The response becomes much more 00:05:13.189 --> 00:05:15.829 patient, encouraging, and grounded in reality. 00:05:15.949 --> 00:05:17.810 It slows down slightly to make sure you're following 00:05:17.810 --> 00:05:19.610 along. And if you sound very confident and want 00:05:19.610 --> 00:05:22.519 to move fast, It simply speeds right up to match 00:05:22.519 --> 00:05:25.699 your exact energy level. This tonal awareness 00:05:25.699 --> 00:05:29.560 unlocks some truly amazing real -world use cases. 00:05:30.000 --> 00:05:32.220 Think about practicing a completely new language 00:05:32.220 --> 00:05:34.819 in real time. You need that immediate feedback 00:05:34.819 --> 00:05:37.639 without awkward pauses breaking your mental flow. 00:05:37.959 --> 00:05:40.579 Or getting step -by -step cooking guidance while 00:05:40.579 --> 00:05:42.639 your hands are deeply busy. You can just talk 00:05:42.639 --> 00:05:45.139 to it while covered in flour and oil. You can 00:05:45.139 --> 00:05:48.000 ask quick complex questions while driving without 00:05:48.000 --> 00:05:51.269 losing visual focus. Brainstorming ideas out 00:05:51.269 --> 00:05:54.209 loud without waiting is incredibly powerful for 00:05:54.209 --> 00:05:56.410 creatives. It feels like having a really smart 00:05:56.410 --> 00:05:58.689 passenger sitting right next to you. It allows 00:05:58.689 --> 00:06:00.810 you to process your thoughts verbally, which 00:06:00.810 --> 00:06:03.730 is how many humans think best. I know this shift 00:06:03.730 --> 00:06:05.589 can feel a bit weird at first. Well, I still 00:06:05.589 --> 00:06:07.589 feel a bit silly talking out loud to my computer. 00:06:07.769 --> 00:06:09.649 Yeah, it feels slightly unnatural to perform 00:06:09.649 --> 00:06:11.930 my thoughts for a machine. That is completely 00:06:11.930 --> 00:06:14.569 normal when trying this entirely new interface 00:06:14.569 --> 00:06:17.449 paradigm. It might feel a bit strange during 00:06:17.449 --> 00:06:20.199 the very first attempt. But that initial fiction 00:06:20.199 --> 00:06:23.100 vanishes very quickly with regular daily use. 00:06:23.379 --> 00:06:25.300 It definitely becomes second nature after a while. 00:06:25.500 --> 00:06:27.740 Can this actually help me design or code just 00:06:27.740 --> 00:06:31.100 by talking? Yes. Vibe coding lets you brainstorm 00:06:31.100 --> 00:06:34.379 and shape ideas purely out loud. Hearing perfectly 00:06:34.379 --> 00:06:37.220 is definitely a huge step forward for artificial 00:06:37.220 --> 00:06:39.980 intelligence. But true, seamless collaboration 00:06:39.980 --> 00:06:42.480 requires seeing what the human user is seeing. 00:06:42.819 --> 00:06:45.500 This is where things get really futuristic and 00:06:45.610 --> 00:06:48.269 deeply impressive technically. Let's talk about 00:06:48.269 --> 00:06:51.410 screen sharing inside the Google AI Studio environment. 00:06:51.769 --> 00:06:54.069 The source material provides a brilliant practical 00:06:54.069 --> 00:06:56.750 example of this feature. A user shared a live 00:06:56.750 --> 00:06:59.589 Google Search Console SEO report. directly. It 00:06:59.589 --> 00:07:03.250 was just a massive screen filled with raw, complex 00:07:03.250 --> 00:07:06.189 keyword data. For a human, scanning that wall 00:07:06.189 --> 00:07:08.970 of text takes significant time. The AI intelligently 00:07:08.970 --> 00:07:11.930 analyzed that complex data in real time instantly. 00:07:12.089 --> 00:07:14.610 It didn't just blindly read the raw numbers out 00:07:14.610 --> 00:07:17.050 loud like a screen reader. It actually synthesized 00:07:17.050 --> 00:07:19.389 the information and looked for broader strategic 00:07:19.389 --> 00:07:21.730 patterns. Right. And it noticed a large surplus 00:07:21.730 --> 00:07:24.649 of branded search keywords. But it also saw a 00:07:24.649 --> 00:07:27.449 glaring lack of how -to keywords in the data. 00:07:27.660 --> 00:07:30.959 That is a very high -level strategic observation 00:07:30.959 --> 00:07:34.019 for a machine to make. Two -sec silence. It interpreted 00:07:34.019 --> 00:07:36.180 the underlying marketing strategy and offered 00:07:36.180 --> 00:07:38.720 a smart recommendation. It acted more like a 00:07:38.720 --> 00:07:40.980 senior consultant than a simple data processor. 00:07:41.220 --> 00:07:43.540 And it gets even more fascinating when you activate 00:07:43.540 --> 00:07:46.339 vision mode. You can point your webcam at the 00:07:46.339 --> 00:07:49.100 messy physical world directly. The user turned 00:07:49.100 --> 00:07:52.100 on the camera and simply waved a hand. The AI 00:07:52.100 --> 00:07:54.579 identified the waving hand perfectly without 00:07:54.579 --> 00:07:57.180 any hesitation or lag. They held up a pen and 00:07:57.180 --> 00:07:59.899 asked for the exact color. The AI got the color 00:07:59.899 --> 00:08:03.620 right on every single rapid attempt. It is actively 00:08:03.620 --> 00:08:06.759 watching the live video feed like a dedicated 00:08:06.759 --> 00:08:11.360 collaborator. Whoa. Imagine an AI instantly guiding 00:08:11.360 --> 00:08:13.980 your hands. through a complex motherboard assembly. 00:08:14.100 --> 00:08:16.899 That is a wildly powerful image to consider for 00:08:16.899 --> 00:08:19.139 a moment. It completely changes how we might 00:08:19.139 --> 00:08:21.480 troubleshoot difficult hardware issues forever. 00:08:21.660 --> 00:08:23.899 You just show the broken physical component directly 00:08:23.899 --> 00:08:25.939 to the camera lens. You can get instant assembly 00:08:25.939 --> 00:08:28.180 instructions by pointing at a confusing product. 00:08:28.360 --> 00:08:30.939 You can identify an unknown plant on a hike or 00:08:30.939 --> 00:08:33.200 a weird ingredient. You can walk through an empty 00:08:33.200 --> 00:08:36.379 room. and get layout feedback instantly. It bridges 00:08:36.379 --> 00:08:39.360 the gap between the digital realm and physical 00:08:39.360 --> 00:08:42.659 reality. But we must honestly report the limitations 00:08:42.659 --> 00:08:45.539 of this specific technology today. It is not 00:08:45.539 --> 00:08:48.279 entirely perfect in every single complex scenario 00:08:48.279 --> 00:08:51.320 yet. That's true. Screen sharing while using 00:08:51.320 --> 00:08:54.450 voice simultaneously can cause some multitasking 00:08:54.450 --> 00:08:57.429 lag. The system is processing an enormous amount 00:08:57.429 --> 00:09:00.049 of data all at once. Sometimes the voice might 00:09:00.049 --> 00:09:02.269 cut out or it takes slightly longer to reply. 00:09:02.429 --> 00:09:04.889 It definitely works best when you do one specific 00:09:04.889 --> 00:09:07.690 thing at a time. Vision mode also gets confused 00:09:07.690 --> 00:09:10.129 if you move objects too quickly. The camera needs 00:09:10.129 --> 00:09:12.169 a moment to focus clearly on the subject. You 00:09:12.169 --> 00:09:14.690 have to move things slowly and deliberately for 00:09:14.690 --> 00:09:17.549 the highest accuracy. Is it safe to let the AI 00:09:17.549 --> 00:09:20.289 watch my live scream? Hide your passwords and 00:09:20.289 --> 00:09:22.490 bank details to actively protect your privacy. 00:09:22.710 --> 00:09:25.090 Let's take a very brief pause right here. Sponsor. 00:09:25.230 --> 00:09:28.149 Okay. We know how powerful its digital and physical 00:09:28.149 --> 00:09:30.950 senses are now. You can hear tone and it can 00:09:30.950 --> 00:09:33.590 see the real world accurately. How do we actually 00:09:33.590 --> 00:09:37.129 mold it to our specific daily needs? And what 00:09:37.129 --> 00:09:39.509 does it actually cost to use this technology 00:09:39.509 --> 00:09:43.210 regularly? Building a custom voice app is surprisingly 00:09:43.210 --> 00:09:46.960 easy to do today. Inside Google AI Studio, there 00:09:46.960 --> 00:09:49.799 is a dedicated build section available. You can 00:09:49.799 --> 00:09:53.419 create an entire complex app using just plain 00:09:53.419 --> 00:09:55.740 English instructions. You do not need to write 00:09:55.740 --> 00:09:58.419 a single line of traditional code. You just describe 00:09:58.419 --> 00:10:01.019 exactly what you want the assistant to do. The 00:10:01.019 --> 00:10:03.759 source provides a very specific and interesting 00:10:03.759 --> 00:10:06.220 prompt example for us. You tell the system to 00:10:06.220 --> 00:10:08.919 be a strict but encouraging language coach. You 00:10:08.919 --> 00:10:11.440 give it very specific behavioral guidelines to 00:10:11.440 --> 00:10:13.919 follow during the chat. You instruct it to correct 00:10:13.919 --> 00:10:16.419 your grammar out loud immediately when speaking. 00:10:16.700 --> 00:10:18.659 You tell it to ask follow -up questions to keep 00:10:18.659 --> 00:10:21.399 the conversation flowing smoothly. Once it is 00:10:21.399 --> 00:10:24.039 live, it stays perfectly locked in that specific 00:10:24.039 --> 00:10:26.320 character. It does not randomly break character 00:10:26.320 --> 00:10:28.620 and sound like a generic robot again. It's kind 00:10:28.620 --> 00:10:31.399 of like stacking Lego blocks of data and behavioral 00:10:31.399 --> 00:10:33.860 instructions together. You can tweet the personality 00:10:33.860 --> 00:10:36.190 until it... Feels exactly right for you. Getting 00:10:36.190 --> 00:10:38.789 it absolutely perfect usually takes two or three 00:10:38.789 --> 00:10:41.490 quick iterative adjustments. Let's dive into 00:10:41.490 --> 00:10:43.289 some of the advanced settings available in the 00:10:43.289 --> 00:10:46.250 studio. There is a very cool thinking level toggle 00:10:46.250 --> 00:10:48.809 you can easily use. You can set this specific 00:10:48.809 --> 00:10:51.909 parameter to low. medium or high. Low thinking 00:10:51.909 --> 00:10:55.129 provides extremely fast, instantaneous replies 00:10:55.129 --> 00:10:58.289 for simple, casual daily chats. High thinking 00:10:58.289 --> 00:11:00.590 means the model takes a bit more time to process. 00:11:00.830 --> 00:11:03.769 It uses that extra computational time to be highly 00:11:03.769 --> 00:11:06.529 accurate and logical. You use this advanced setting 00:11:06.529 --> 00:11:09.710 for complex math or deep coding help. You can 00:11:09.710 --> 00:11:12.009 also toggle on Google search as a live tool. 00:11:12.169 --> 00:11:14.710 This means the AI is never stuck in the past 00:11:14.710 --> 00:11:17.509 with outdated data. It pulls real -time information 00:11:17.509 --> 00:11:19.789 from the web to answer complex news questions. 00:11:20.009 --> 00:11:22.090 Let's clearly discuss the global availability 00:11:22.090 --> 00:11:24.509 and the specific pricing structure now. You can 00:11:24.509 --> 00:11:26.549 use Gemini Live on your mobile phone starting 00:11:26.549 --> 00:11:28.690 today. Search Live is currently available in 00:11:28.690 --> 00:11:31.190 over 200 different global countries. Testing 00:11:31.190 --> 00:11:33.889 everything inside Google AI Studio is completely 00:11:33.889 --> 00:11:36.509 free right now. It's a fantastic sandbox for 00:11:36.509 --> 00:11:38.750 experimentation and learning the ropes. But once 00:11:38.750 --> 00:11:41.830 you publish to Google Cloud, Standard API costs 00:11:41.830 --> 00:11:46.070 begin immediately. The backend pricing is entirely 00:11:46.070 --> 00:11:49.250 based on your total token usage. Let's define 00:11:49.250 --> 00:11:51.529 that technical term for our listeners right now. 00:11:51.950 --> 00:11:55.649 Tokens are tiny chunks of words the AI uses to 00:11:55.649 --> 00:11:57.730 read and write. For the flashlight model, the 00:11:57.730 --> 00:11:59.929 pricing is very straightforward and affordable. 00:12:00.250 --> 00:12:03.129 It is 25 cents for input per million tokens used. 00:12:03.289 --> 00:12:07.529 It is $1 .50 for output per million tokens generated. 00:12:07.950 --> 00:12:10.809 The standard flash model has a slightly higher 00:12:10.809 --> 00:12:13.809 overall cost structure. It runs at $0 .50 for 00:12:13.809 --> 00:12:16.909 input per million tokens used. And it costs $3 00:12:16.909 --> 00:12:19.769 for output per million tokens generated. If real 00:12:19.769 --> 00:12:22.309 -time conversational speed is not strictly required, 00:12:22.529 --> 00:12:24.870 there is another option. Batch pricing is exactly 00:12:24.870 --> 00:12:27.899 half the cost of those standard API rates. What 00:12:27.899 --> 00:12:30.700 happens if my custom app gets unexpectedly popular 00:12:30.700 --> 00:12:33.299 and expensive? Set a Google Cloud spending cap 00:12:33.299 --> 00:12:35.940 so you absolutely never overpay. We have covered 00:12:35.940 --> 00:12:38.500 a massive amount of technical ground today together. 00:12:38.639 --> 00:12:41.039 We looked at latency. tonal awareness, vision 00:12:41.039 --> 00:12:43.659 capabilities, and custom voice apps. Let's summarize 00:12:43.659 --> 00:12:46.080 the core big idea for you to take away. Google 00:12:46.080 --> 00:12:49.279 Gemini 3 .1 Flash is not just a simple processing 00:12:49.279 --> 00:12:52.139 upgrade. It's not just doing the exact same old 00:12:52.139 --> 00:12:54.940 things slightly faster. It is a fundamental shift 00:12:54.940 --> 00:12:58.600 in the entire daily user experience. It transforms 00:12:58.600 --> 00:13:01.820 AI from a sterile distant database that you simply 00:13:01.820 --> 00:13:05.120 query. It becomes a highly responsive, emotionally 00:13:05.120 --> 00:13:08.070 aware, virtual collaborator in your life. It 00:13:08.070 --> 00:13:10.750 works at the actual natural speed of human thought 00:13:10.750 --> 00:13:13.929 now. That profound lack of friction changes how 00:13:13.929 --> 00:13:16.669 you approach complex problems entirely. It removes 00:13:16.669 --> 00:13:19.230 the barrier between having an idea and executing 00:13:19.230 --> 00:13:21.590 that idea. We strongly encourage you to try it 00:13:21.590 --> 00:13:23.809 out for yourself today. You do not need to be 00:13:23.809 --> 00:13:26.429 an AI expert to start experimenting safely. Just 00:13:26.429 --> 00:13:28.710 go to Google AI studio and turn on talk mode 00:13:28.710 --> 00:13:31.320 immediately. Ask it a simple question and experience 00:13:31.320 --> 00:13:34.139 the incredible speed firsthand. Start with small, 00:13:34.399 --> 00:13:36.340 manageable tasks and let the tool grow with you. 00:13:36.399 --> 00:13:38.360 It takes time to build the muscle memory for 00:13:38.360 --> 00:13:40.879 this new interface. Before we go, I want to leave 00:13:40.879 --> 00:13:43.700 you with a lingering thought to sex silence. 00:13:43.879 --> 00:13:46.840 If AI can perfectly mimic the rhythm, tone, and 00:13:46.840 --> 00:13:48.860 empathy of human conversation without skipping 00:13:48.860 --> 00:13:51.500 a single beat, what happens when we start preferring 00:13:51.500 --> 00:13:54.419 its company over actual people for our day -to 00:13:54.419 --> 00:13:56.950 -day brainstorming? That is a deeply fascinating 00:13:56.950 --> 00:13:59.389 and complex question to chew on. Thank you for 00:13:59.389 --> 00:14:01.789 joining us on today's Deep Dive. Out to your 00:14:01.789 --> 00:14:02.090 music.