WEBVTT 00:00:00.000 --> 00:00:02.600 You must have noticed, right? Generative AI seems 00:00:02.600 --> 00:00:06.040 to be absolutely everywhere now. Oh, yeah. Creating 00:00:06.040 --> 00:00:09.539 images, writing text. It feels, I don't know, 00:00:09.640 --> 00:00:11.599 almost like magic sometimes. It does have that 00:00:11.599 --> 00:00:14.859 feel. But then you remember, AI isn't exactly 00:00:14.859 --> 00:00:16.920 new. It's been working away in the background 00:00:16.920 --> 00:00:19.820 for years, hasn't it? For a long time, yes. Often, 00:00:19.879 --> 00:00:23.679 quite invisibly. So it got me thinking, what 00:00:23.679 --> 00:00:28.300 is the actual difference? between this flashy 00:00:28.300 --> 00:00:30.739 generative stuff and the traditional AI we've 00:00:30.739 --> 00:00:32.899 had for ages. You probably wondered that too. 00:00:33.380 --> 00:00:35.539 Absolutely. The buzz around generative AI is 00:00:35.539 --> 00:00:39.850 huge and, well, for good reason. Yeah, to really 00:00:39.850 --> 00:00:42.030 get what makes it tick, you need to understand 00:00:42.030 --> 00:00:45.409 what came first, the traditional AI. Okay. And 00:00:45.409 --> 00:00:46.929 that's exactly what we're diving into today. 00:00:47.049 --> 00:00:49.030 We're using some insights from a really clear 00:00:49.030 --> 00:00:51.590 article over on hackscience .education. It's 00:00:51.590 --> 00:00:53.789 called AI, Generative, and Traditional. Right. 00:00:53.850 --> 00:00:55.429 So the mission today is pretty straightforward. 00:00:55.670 --> 00:00:57.429 We want to get clear on what these two types 00:00:57.429 --> 00:01:00.250 of AI are, how they differ, and crucially, why 00:01:00.250 --> 00:01:02.210 you should care about the difference. So let's 00:01:02.210 --> 00:01:04.810 start with the baseline, this traditional AI, 00:01:05.150 --> 00:01:07.530 the workhorse, you called it. What's its story? 00:01:07.719 --> 00:01:09.760 Well, like the article points out, it's been 00:01:09.760 --> 00:01:12.099 powering a lot of things we use every single 00:01:12.099 --> 00:01:16.359 day, often without us noticing. Like what? Think 00:01:16.359 --> 00:01:19.659 about search engine results or when you're shopping 00:01:19.659 --> 00:01:21.680 online and get those product recommendations. 00:01:21.780 --> 00:01:24.739 Yeah. Or your music playlist, how they seem to 00:01:24.739 --> 00:01:27.359 know what you like. Even predictive text on your 00:01:27.359 --> 00:01:30.140 phone. That's often traditional AI doing its 00:01:30.140 --> 00:01:33.629 job. OK, so it's the quiet helper tech, making 00:01:33.629 --> 00:01:36.170 things smoother. But how does it actually work? 00:01:36.590 --> 00:01:38.489 The article mentioned classical data science 00:01:38.489 --> 00:01:41.629 and a structured process. Sounds a bit formal. 00:01:42.010 --> 00:01:44.030 It is quite methodical. It usually starts with 00:01:44.030 --> 00:01:45.969 gathering data. Then you have to prepare that 00:01:45.969 --> 00:01:48.250 data, clean it up. Then there's a step called 00:01:48.250 --> 00:01:51.430 feature engineering, which in traditional AI 00:01:51.430 --> 00:01:53.810 often means humans have to figure out and tell 00:01:53.810 --> 00:01:56.569 the AI which bits of the data are most important 00:01:56.569 --> 00:01:59.390 for the task. OK, so humans guide it quite a 00:01:59.390 --> 00:02:02.099 bit there. Exactly. Then the model gets trained 00:02:02.099 --> 00:02:04.400 on that data. And finally, you check how well 00:02:04.400 --> 00:02:07.680 it performs, validate it. The main goal here 00:02:07.680 --> 00:02:10.759 is usually prediction or classification. Meaning 00:02:10.759 --> 00:02:13.300 it uses the data it learned from to answer questions 00:02:13.300 --> 00:02:16.860 like, what category does this belong to? Or what's 00:02:16.860 --> 00:02:19.800 likely to happen next based on past patterns? 00:02:20.060 --> 00:02:22.800 Got it. Spotting patterns to make guesses or 00:02:22.800 --> 00:02:25.560 sort things. That makes sense. The article also 00:02:25.560 --> 00:02:28.389 said it operates within predefined boundaries. 00:02:28.509 --> 00:02:31.069 What's that about? That's key. Traditional AI 00:02:31.069 --> 00:02:33.330 works based on the rules and instructions programmed 00:02:33.330 --> 00:02:36.210 into it. It learns from a specific data set for 00:02:36.210 --> 00:02:39.270 a specific job. And its abilities are pretty 00:02:39.270 --> 00:02:41.490 much limited to what it was taught. The results 00:02:41.490 --> 00:02:44.629 are usually deterministic, same input, same output, 00:02:45.009 --> 00:02:46.969 and predictable. It's like a tool built for one 00:02:46.969 --> 00:02:49.530 specific purpose. So the example in the article. 00:02:49.800 --> 00:02:52.219 A traditional customer service chatbot. How does 00:02:52.219 --> 00:02:54.560 that fit? Right. That kind of bot probably has 00:02:54.560 --> 00:02:57.020 a set list of questions it understands and prepared 00:02:57.020 --> 00:02:59.620 answers it gives out. OK. It matches keywords, 00:02:59.919 --> 00:03:02.699 finds the right script. It can be good for common 00:03:02.699 --> 00:03:05.340 questions within its zone. But if you ask something 00:03:05.340 --> 00:03:08.099 weird. Exactly. If you ask something outside 00:03:08.099 --> 00:03:10.360 its programming, it'll likely just say, sorry, 00:03:10.479 --> 00:03:12.699 I don't understand, or give a default response. 00:03:13.060 --> 00:03:15.919 It can't really adapt or learn on the fly beyond 00:03:15.919 --> 00:03:18.219 its rules. It's like a very complex flowchart 00:03:18.219 --> 00:03:20.599 in a way. Right, a sophisticated flowchart. That's 00:03:20.599 --> 00:03:23.000 a good way to put it. And what about complexity? 00:03:23.979 --> 00:03:26.340 Resources. The article hinted traditional AI 00:03:26.340 --> 00:03:31.319 is maybe simpler, less demanding. Generally, 00:03:31.599 --> 00:03:34.610 yes. Compared to the big generative models, traditional 00:03:34.610 --> 00:03:37.150 AI models are often less complex. They don't 00:03:37.150 --> 00:03:39.949 need quite as much computing power. And like 00:03:39.949 --> 00:03:41.650 the article says, they're usually trained on 00:03:41.650 --> 00:03:44.250 smaller data sets, and importantly, labeled data 00:03:44.250 --> 00:03:46.330 sets, data where someone has already marked the 00:03:46.330 --> 00:03:48.449 right answer. And they don't really have ways 00:03:48.449 --> 00:03:50.750 to adapt to new stuff unless you completely retrain 00:03:50.750 --> 00:03:53.289 them with new labeled data. OK, solid picture 00:03:53.289 --> 00:03:56.409 of traditional AI then. Focused, rule following, 00:03:56.830 --> 00:04:00.330 good at predicting within its limits. Now, generative 00:04:00.330 --> 00:04:03.009 AI. The exciting newcomer. What's the fundamental 00:04:03.009 --> 00:04:06.509 shift? Well, the article calls it a unique and 00:04:06.509 --> 00:04:08.409 fascinating advancement. And I think that's fair. 00:04:09.189 --> 00:04:12.009 The absolute core difference is what it does. 00:04:12.270 --> 00:04:14.550 Instead of just analyzing or predicting from 00:04:14.550 --> 00:04:18.269 existing data, it can actually create new data, 00:04:18.670 --> 00:04:21.129 new content, stuff that wasn't explicitly in 00:04:21.129 --> 00:04:24.720 its training data. Whoa. OK. Creation. That's 00:04:24.720 --> 00:04:27.860 a big leap from just prediction. Huge. It's not 00:04:27.860 --> 00:04:29.759 just about identifying something, it's about 00:04:29.759 --> 00:04:33.040 producing something novel. This ability to generate, 00:04:33.279 --> 00:04:35.100 though, it comes with uncertainty. Sometimes 00:04:35.100 --> 00:04:37.680 it creates amazing things, sometimes less so. 00:04:38.000 --> 00:04:40.220 That needs checking, which isn't really a factor 00:04:40.220 --> 00:04:42.920 for deterministic traditional AI. Making new 00:04:42.920 --> 00:04:45.379 stuff versus recognizing old stuff, how does 00:04:45.379 --> 00:04:47.480 it manage that? The article said something about 00:04:47.480 --> 00:04:50.259 replicating patterns, but also imagining new 00:04:50.259 --> 00:04:52.800 ones. That sounds, well... It does, doesn't it? 00:04:52.920 --> 00:04:54.779 And that's the exciting part. Basically, it learns 00:04:54.779 --> 00:04:57.040 the deep patterns, the structures, the relationships 00:04:57.040 --> 00:05:00.379 within truly massive amounts of data. And then 00:05:00.379 --> 00:05:02.899 it uses that learned understanding to generate 00:05:02.899 --> 00:05:05.480 new examples that follow those same kinds of 00:05:05.480 --> 00:05:08.110 patterns but are still original. It's crafting 00:05:08.110 --> 00:05:10.490 scenarios, maybe even synthesizing knowledge 00:05:10.490 --> 00:05:12.889 in a way. It's an evolution of deep learning, 00:05:13.149 --> 00:05:15.110 as the article notes. Deep learning. Where the 00:05:15.110 --> 00:05:17.490 AI figures out the important features itself 00:05:17.490 --> 00:05:20.129 from all that data, unlike the manual feature 00:05:20.129 --> 00:05:23.209 engineering in traditional AI. And the article 00:05:23.209 --> 00:05:27.970 contrasted the deterministic, rule -based approach 00:05:27.970 --> 00:05:30.829 with a probabilistic approach. What does that 00:05:30.829 --> 00:05:32.790 mean in practice? Go ahead, think of it like 00:05:32.790 --> 00:05:35.069 this. Traditional AI is like flipping a light 00:05:35.069 --> 00:05:37.720 switch. You flip it, the light comes on, same 00:05:37.720 --> 00:05:41.620 action, same result. Deterministic. Gotcha. Generative 00:05:41.620 --> 00:05:43.839 AI is more like rolling dice. You roll them, 00:05:43.920 --> 00:05:45.879 you know you'll get numbers between 1 and 6, 00:05:46.000 --> 00:05:47.740 but you don't know exactly which numbers until 00:05:47.740 --> 00:05:50.600 they land. It's based on probabilities. OK. It 00:05:50.600 --> 00:05:52.920 looks at the input, considers the patterns it 00:05:52.920 --> 00:05:55.040 knows, and then generates what it calculates 00:05:55.040 --> 00:05:57.279 as the most likely next thing, word, pixel, note, 00:05:57.420 --> 00:06:00.180 whatever. But there's an element of chance of 00:06:00.180 --> 00:06:03.279 variability. So it's not rigidly fixed. Exactly. 00:06:03.519 --> 00:06:05.459 The outcome isn't predetermined in the same way. 00:06:05.879 --> 00:06:08.459 That probability aspect is what allows for the 00:06:08.459 --> 00:06:11.360 novelty, the variation, the outputs that weren't 00:06:11.360 --> 00:06:13.579 explicitly programmed line by line. So that's 00:06:13.579 --> 00:06:15.959 the magic the article mentioned. Generating original 00:06:15.959 --> 00:06:18.540 content. Stuff we used to think only humans could 00:06:18.540 --> 00:06:21.480 do. It's a big mental shift. It really is. But 00:06:21.480 --> 00:06:23.939 crucially, the article also said generative AI 00:06:23.939 --> 00:06:26.540 doesn't just kick classical data science to the 00:06:26.540 --> 00:06:30.089 curb, they work together. How? Yeah, that's super 00:06:30.089 --> 00:06:32.810 important. Generative AI isn't replacing everything. 00:06:33.389 --> 00:06:36.110 Classical data science is still vital. For one, 00:06:36.189 --> 00:06:38.110 you need those techniques to manage and prepare 00:06:38.110 --> 00:06:40.810 the enormous data sets these models learn from. 00:06:41.569 --> 00:06:43.790 And maybe even more critically, you need traditional 00:06:43.790 --> 00:06:46.189 methods to evaluate the output of generative 00:06:46.189 --> 00:06:49.029 AI. Is it accurate? Is it biased? Is it safe? 00:06:49.129 --> 00:06:51.730 Is it useful? We need ways to measure that. So 00:06:51.730 --> 00:06:53.610 they complement each other. Definitely. It's 00:06:53.610 --> 00:06:55.889 a partnership. OK, the article helpfully summarized 00:06:55.889 --> 00:06:58.069 the key differences to make it clear. Let's run 00:06:58.069 --> 00:07:01.170 through those first. Creation versus prediction. 00:07:02.149 --> 00:07:05.350 More examples. Sure. Traditional AI, identifying 00:07:05.350 --> 00:07:08.290 if an email is spam. That's classification, a 00:07:08.290 --> 00:07:11.629 type of prediction. OK. Or forecasting stock 00:07:11.629 --> 00:07:14.550 market trends based on past data. Prediction 00:07:14.550 --> 00:07:17.970 again. Right. Generative AI, writing a poem. 00:07:18.220 --> 00:07:21.339 composing music, generating an image from a text 00:07:21.339 --> 00:07:24.040 prompt, like a cat riding a bicycle on the moon, 00:07:24.180 --> 00:07:27.339 or writing software code, designing a new molecule. 00:07:27.699 --> 00:07:30.399 It's producing something new. Got it. Understanding 00:07:30.399 --> 00:07:33.959 versus making. Next difference. Approach and 00:07:33.959 --> 00:07:37.680 outcome. Rule -based deterministic versus probabilistic 00:07:37.680 --> 00:07:40.680 varied. You mentioned the outputs can vary with 00:07:40.680 --> 00:07:42.920 generative AI. Can you control that? Yes, often 00:07:42.920 --> 00:07:45.410 you can. The article mentions temperature. It's 00:07:45.410 --> 00:07:47.790 like a creativity knob on many models. A creativity 00:07:47.790 --> 00:07:50.110 knob. Sort of. Lower temperature means that AI 00:07:50.110 --> 00:07:52.709 sticks to the most likely common patterns, more 00:07:52.709 --> 00:07:55.230 predictable, maybe safer outputs. OK. Higher 00:07:55.230 --> 00:07:57.470 temperature lets it take more chances, explore 00:07:57.470 --> 00:08:00.250 less probable paths, can lead to more creative, 00:08:00.470 --> 00:08:03.129 surprising results, but maybe also more nonsensical 00:08:03.129 --> 00:08:05.209 ones. Interesting. So you can dial the randomness 00:08:05.209 --> 00:08:07.649 up or down? Pretty much. That non -deterministic 00:08:07.649 --> 00:08:11.930 nature is core to generative AI. Next. Training 00:08:11.930 --> 00:08:14.629 data and method. Sounds like a massive difference 00:08:14.629 --> 00:08:17.750 here. Oh, absolutely huge. Traditional AI often 00:08:17.750 --> 00:08:21.269 uses smaller, carefully curated, labeled data 00:08:21.269 --> 00:08:23.829 sets. Maybe thousands or tens of thousands of 00:08:23.829 --> 00:08:26.529 examples. Right, labeled. Generative AI, especially 00:08:26.529 --> 00:08:29.610 the big models like LLMs, train on vast amounts 00:08:29.610 --> 00:08:32.309 of data. Think internet scale text and images. 00:08:32.629 --> 00:08:35.529 Billions, trillions of data points. And the methods 00:08:35.529 --> 00:08:37.710 are different too. Things like self -supervision, 00:08:37.990 --> 00:08:40.629 where the AI learns from unlabeled data by, say, 00:08:40.750 --> 00:08:43.649 predicting missing words and sentences. It takes 00:08:43.649 --> 00:08:45.909 way more computer power, way more time. Which 00:08:45.909 --> 00:08:48.269 leads directly to the next point, model complexity 00:08:48.269 --> 00:08:50.690 and resources. Yeah. Generative AI needs more, 00:08:50.950 --> 00:08:53.269 muscle. Way more. Traditional models can often 00:08:53.269 --> 00:08:55.789 run on regular computers, even your phone sometimes. 00:08:55.850 --> 00:08:57.990 These big generative models, they have billions, 00:08:58.149 --> 00:09:00.590 sometimes trillions of parameters, the internal 00:09:00.590 --> 00:09:03.230 variables they learn. They almost always need 00:09:03.230 --> 00:09:06.490 powerful cloud computing setups, accessed via 00:09:06.490 --> 00:09:09.350 APIs usually. That means higher costs too. Makes 00:09:09.350 --> 00:09:12.559 sense. Okay, adaptation. Traditional AI sounds 00:09:12.559 --> 00:09:15.120 pretty fixed after training. It mostly is. To 00:09:15.120 --> 00:09:17.440 adapt, you usually have to retrain it with new 00:09:17.440 --> 00:09:20.529 labeled data. which is a whole process. But generative 00:09:20.529 --> 00:09:23.149 AI. It's more flexible. The article mentions 00:09:23.149 --> 00:09:25.389 prompt engineering. Just changing your instructions 00:09:25.389 --> 00:09:27.690 can change the output dramatically. Right, the 00:09:27.690 --> 00:09:30.049 prompt. Then there's ARGI retrieval augmented 00:09:30.049 --> 00:09:33.210 generation. That lets the AI pull in current 00:09:33.210 --> 00:09:36.029 information from outside sources to answer questions. 00:09:36.049 --> 00:09:38.649 Oh, clever. And fine -tuning, where you can tweak 00:09:38.649 --> 00:09:41.690 a pre -trained model for a specific task without 00:09:41.690 --> 00:09:44.450 starting from scratch. So yeah, more ways to 00:09:44.450 --> 00:09:47.379 adapt its behavior after the main training. And 00:09:47.379 --> 00:09:50.779 the last difference listed, modality of interaction, 00:09:51.000 --> 00:09:54.320 how we talk to them. Yeah. Traditional AI often 00:09:54.320 --> 00:09:57.059 uses structured input, filling forms, clicking 00:09:57.059 --> 00:10:00.580 buttons, specific data formats, generative AI, 00:10:00.820 --> 00:10:03.200 particularly LLMs. You interact using natural 00:10:03.200 --> 00:10:05.960 language, just typing or speaking what you want, 00:10:06.019 --> 00:10:08.440 prompts. Much more conversational. Exactly. It's 00:10:08.440 --> 00:10:10.740 more expressive. And that's why, as the article 00:10:10.740 --> 00:10:13.220 says, prompt engineering itself is becoming a 00:10:13.220 --> 00:10:15.580 whole field, learning how to ask the right way 00:10:15.580 --> 00:10:18.460 to get the best results. Okay, these distinctions 00:10:18.460 --> 00:10:20.720 really help paint the picture. And you mentioned 00:10:20.720 --> 00:10:23.980 LLMs. Large language models are central to the 00:10:23.980 --> 00:10:26.179 generative AI buzz. Tell us more about them. 00:10:26.399 --> 00:10:29.340 Right. LLMs are a specific, very powerful type 00:10:29.340 --> 00:10:32.340 of generative AI. They're trained on truly immense 00:10:32.340 --> 00:10:35.159 amounts of text data. Who's text? Primarily text, 00:10:35.200 --> 00:10:37.039 which allows them to understand and generate 00:10:37.039 --> 00:10:39.759 human language incredibly well. What's amazing 00:10:39.759 --> 00:10:42.220 is their versatility. How so? Because their general 00:10:42.220 --> 00:10:44.980 purpose. Give one the right prompt, and it can 00:10:44.980 --> 00:10:47.580 answer questions, summarize articles, translate 00:10:47.580 --> 00:10:50.779 between languages, write code. all from one model. 00:10:51.100 --> 00:10:54.139 Often, yes, without specific retraining for each 00:10:54.139 --> 00:10:56.539 task. The article names the big ones you hear 00:10:56.539 --> 00:11:00.279 about OpenAI's GPT models, Anthropics Claude, 00:11:00.460 --> 00:11:03.919 Google's Polym, and Gemini, foundational models. 00:11:04.059 --> 00:11:06.759 Incredible flexibility. But the article also 00:11:06.759 --> 00:11:09.019 warned it's not a silver bullet. There are catches, 00:11:09.500 --> 00:11:11.980 right? Considerations. Definitely. It raises 00:11:11.980 --> 00:11:15.080 important issues, financial costs, legal questions 00:11:15.080 --> 00:11:17.860 like copyright, technical challenges, and significant 00:11:17.860 --> 00:11:21.019 ethical concerns. risks of generating incorrect 00:11:21.019 --> 00:11:23.980 or biased information, sometimes called hallucinations, 00:11:24.480 --> 00:11:26.899 potential for misuse, creating deep fakes, or 00:11:26.899 --> 00:11:29.220 spreading disinformation. Yeah, serious stuff. 00:11:29.379 --> 00:11:31.539 So the article stresses that organizations need 00:11:31.539 --> 00:11:34.860 a clear strategy. They need goals, good governance, 00:11:35.240 --> 00:11:37.960 responsible AI policies. You need to think about 00:11:37.960 --> 00:11:41.000 costs, reliability, especially if you're depending 00:11:41.000 --> 00:11:43.779 on an API. Plan for problems, basically. Exactly. 00:11:43.799 --> 00:11:46.139 And think about how to use your own unique data. 00:11:46.519 --> 00:11:48.559 Manage the costs. It's not just plug and play. 00:11:48.399 --> 00:11:51.659 magic. Right. Requires careful thought. But the 00:11:51.659 --> 00:11:54.399 article finished on a key point. These two types 00:11:54.399 --> 00:11:57.860 of AI, traditional and generative, they're not 00:11:57.860 --> 00:12:00.240 enemies. Not at all. That's a crucial takeaway. 00:12:00.519 --> 00:12:03.039 They are not mutually exclusive. Often they work 00:12:03.039 --> 00:12:05.710 best together. How would that look? Think of 00:12:05.710 --> 00:12:08.330 a sophisticated chatbot. It might use a generative 00:12:08.330 --> 00:12:10.649 LLM for the fluid natural conversation part, 00:12:10.889 --> 00:12:13.289 but rely on traditional models behind the scenes 00:12:13.289 --> 00:12:15.870 to maybe detect customer frustration from the 00:12:15.870 --> 00:12:18.149 text or route the query to the right department 00:12:18.149 --> 00:12:20.750 based on classification. Combining strengths. 00:12:21.029 --> 00:12:23.710 Precisely. And remember what we said about evaluation. 00:12:24.009 --> 00:12:27.169 You absolutely need classical data science methods 00:12:27.169 --> 00:12:30.149 to check the outputs of generative AI to make 00:12:30.149 --> 00:12:32.350 sure it's accurate and responsible. So the future 00:12:32.350 --> 00:12:34.970 isn't one or the other. No, the future is almost 00:12:34.970 --> 00:12:37.629 certainly a blend, using the right tool for the 00:12:37.629 --> 00:12:40.889 right job. Traditional AI for reliable prediction 00:12:40.889 --> 00:12:43.970 and classification where rules work well. Generative 00:12:43.970 --> 00:12:47.409 AI for creating, brainstorming, tackling new 00:12:47.409 --> 00:12:49.649 kinds of problems that need that flexibility. 00:12:49.830 --> 00:12:51.649 That makes perfect sense, a synergy. Exactly. 00:12:51.870 --> 00:12:54.070 And that's really the core message today, I think. 00:12:54.490 --> 00:12:57.049 Traditional AI predicts and classifies within 00:12:57.049 --> 00:13:00.629 known rules. Generative AI creates novel things 00:13:00.629 --> 00:13:04.169 using probabilities learned from vast data. Understanding 00:13:04.169 --> 00:13:06.330 that difference feels really important now. It's 00:13:06.330 --> 00:13:09.259 becoming essential. As AI gets woven deeper into 00:13:09.259 --> 00:13:11.620 everything, knowing what kind of AI is involved 00:13:11.620 --> 00:13:14.440 helps you understand its power and also its limits. 00:13:15.200 --> 00:13:17.059 Absolutely. So something for you, the listener, 00:13:17.639 --> 00:13:20.159 to think about as you go about your day. Consider 00:13:20.159 --> 00:13:24.059 this. How might? this blend, this increasing 00:13:24.059 --> 00:13:26.620 integration of both traditional and generative 00:13:26.620 --> 00:13:29.799 AI actually reshape things, not just tech, but 00:13:29.799 --> 00:13:32.539 maybe creativity, problem solving, even how we 00:13:32.539 --> 00:13:35.039 think about intelligence itself. When you encounter 00:13:35.039 --> 00:13:37.980 AI, a recommendation, a summary, an image, try 00:13:37.980 --> 00:13:40.240 and ask yourself, is this mostly predicting something 00:13:40.240 --> 00:13:43.039 based on rules and past data, or is it generating 00:13:43.039 --> 00:13:45.159 something fundamentally new? It's a fascinating 00:13:45.159 --> 00:13:46.799 lens to view the world through right now.