title
Andrew Ng - The State of Artificial Intelligence
description
Professor Andrew Ng is the former chief scientist at Baidu, where he led the company's Artificial Intelligence Group. He is an adjunct professor at Stanford University. In 2011 he led the development of Stanford University’s main MOOC (Massive Open Online Courses) platform and also taught an online Machine Learning class that was offered to over 100,000 students, leading to the founding of Coursera.
November 7, 2017
detail
{'title': 'Andrew Ng - The State of Artificial Intelligence', 'heatmap': [{'end': 1196.281, 'start': 1177.016, 'weight': 1}], 'summary': "Explores ai's transformative potential across industries, comparing it to the impact of electricity, highlighting its economic value, lucrative applications in technology, and its potential for automating tasks and impacting job nature. it also discusses tech's responsibility in addressing job displacement, the correlation between neural network size, data availability, and the rise of deep learning, the economic value, and deployment challenges of ai algorithms, and leading ai organizations' data acquisition strategies and workflow challenges.", 'chapters': [{'end': 384.642, 'segs': [{'end': 113.195, 'src': 'embed', 'start': 66.752, 'weight': 0, 'content': [{'end': 67.733, 'text': 'Healthcare was transformed.', 'start': 66.752, 'duration': 0.981}, {'end': 72.597, 'text': 'But today, all of these industries, you have a hard time imagining how to run these things without electricity.', 'start': 67.753, 'duration': 4.844}, {'end': 77.401, 'text': 'I think that AI technology, especially deep learning, as Bo was saying,', 'start': 73.277, 'duration': 4.124}, {'end': 83.266, 'text': 'has now advanced to the point where we see a surprisingly clear path for it to also transform every major industry.', 'start': 77.401, 'duration': 5.865}, {'end': 95.859, 'text': 'Now, despite all the hype about AI and there is a lot of hype it turns out that 99% of the economic value created by AI today is through one type of AI,', 'start': 84.335, 'duration': 11.524}, {'end': 104.723, 'text': 'which is learning A to B or input to output mappings.', 'start': 95.859, 'duration': 8.864}, {'end': 113.195, 'text': 'For example, AI is getting really good at inputting a picture and outputting, is it you? You know, 01.', 'start': 104.803, 'duration': 8.392}], 'summary': 'Ai technology, especially deep learning, is set to transform every major industry, with 99% of the economic value created by ai today coming from learning a to b or input to output mappings.', 'duration': 46.443, 'max_score': 66.752, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs66752.jpg'}, {'end': 157.544, 'src': 'embed', 'start': 133.068, 'weight': 3, 'content': [{'end': 139.35, 'text': "if you can figure this out, this is a decent business in terms of figuring out whether or not to approve someone's loan application.", 'start': 133.068, 'duration': 6.282}, {'end': 144.193, 'text': 'Possibly the most lucrative application of this today is online advertising.', 'start': 140.051, 'duration': 4.142}, {'end': 148.795, 'text': 'If you look at all the large online web serving platforms Google, Facebook, Baidu,', 'start': 144.233, 'duration': 4.562}, {'end': 157.544, 'text': 'Twitter and so on Almost all the launch ad platforms have a piece of AI technology that takes this input and adds some information about you,', 'start': 148.795, 'duration': 8.749}], 'summary': 'Ai technology is used for online advertising and loan approvals, with platforms like google and facebook using it.', 'duration': 24.476, 'max_score': 133.068, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs133068.jpg'}, {'end': 228.947, 'src': 'embed', 'start': 189.241, 'weight': 2, 'content': [{'end': 195.665, 'text': 'Both Google and Baidu have said publicly that over 10% of the searches on mobile devices now come through voice search.', 'start': 189.241, 'duration': 6.424}, {'end': 197.526, 'text': "So there's a large volume of revenue.", 'start': 195.685, 'duration': 1.841}, {'end': 207.412, 'text': 'And the reason users are willing to accept voice input is because speech recognition is finally accurate enough for you to speak to your cell phone and have it transcribe that pretty accurately.', 'start': 198.067, 'duration': 9.345}, {'end': 212.095, 'text': "So that's enabled by phone VR that inputs an audio clip and outputs a text transcript.", 'start': 207.853, 'duration': 4.242}, {'end': 220.441, 'text': 'You know, machine translation, input an English sentence, output a French sentence, and self-driving cars.', 'start': 213.717, 'duration': 6.724}, {'end': 228.947, 'text': 'I guess I recently joined the board of self-driving car company Drive.ai, and a key component of every self-driving car,', 'start': 220.461, 'duration': 8.486}], 'summary': 'Over 10% of mobile searches are through voice; accurate speech recognition enables voice input for users; phone vr and machine translation are key technologies.', 'duration': 39.706, 'max_score': 189.241, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs189241.jpg'}, {'end': 328.705, 'src': 'embed', 'start': 283.681, 'weight': 5, 'content': [{'end': 305.052, 'text': 'I think has been helpful, which is anything a typical person can do with less than a second of thinking we can probably now or soon automate.', 'start': 283.681, 'duration': 21.371}, {'end': 323.221, 'text': 'If you look at a lot of the examples I had on the previous list, many of them, such as look at the picture and figure out is this a picture of you?', 'start': 316.775, 'duration': 6.446}, {'end': 326.924, 'text': 'Many of those are tasks that a human can do in less than one second.', 'start': 323.281, 'duration': 3.643}, {'end': 328.705, 'text': 'And this is an imperfect rule.', 'start': 327.324, 'duration': 1.381}], 'summary': 'Many tasks that take less than a second can now be automated.', 'duration': 45.024, 'max_score': 283.681, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs283681.jpg'}], 'start': 5.885, 'title': "Ai's impact across industries", 'summary': 'Delves into the transformational potential of ai, likening it to the impact of electricity on industries, highlighting its current economic value and potential to influence all major industries, while also exploring lucrative applications in technology, such as loan approval and speech recognition, and discussing its impact on various applications and the potential for automation of tasks, impacting the nature of jobs.', 'chapters': [{'end': 113.195, 'start': 5.885, 'title': "Ai's impact on industries", 'summary': 'Discusses the transformative potential of ai, comparing it to the impact of electricity on various industries, emphasizing the current economic value created by ai and its potential to impact every major industry.', 'duration': 107.31, 'highlights': ['AI is positioned to transform every major industry, similar to how electricity revolutionized various sectors such as agriculture, communications, manufacturing, and healthcare, with a clear path for widespread impact.', 'Currently, 99% of the economic value created by AI is through learning A to B or input to output mappings, showcasing its significant influence in various applications such as image recognition.', 'The speaker expresses excitement about the potential of AI and its transformative impact on businesses, offering advice on how to navigate and thrive in the evolving technological landscape.']}, {'end': 207.412, 'start': 113.195, 'title': "Ai's impact in technology", 'summary': 'Discusses the lucrative applications of ai, such as loan approval, online advertising, and speech recognition, with examples like google and baidu processing over 10% of mobile searches through voice search.', 'duration': 94.217, 'highlights': ['Speech recognition by Google and Baidu accounts for over 10% of mobile searches, indicating significant user adoption and revenue volume.', 'Online advertising platforms use AI to analyze user data and enhance ad targeting, with the aim of increasing click-through rates and revenue.', "AI's applications in loan approval processes present a lucrative business opportunity, demonstrating the potential for significant financial gains.", 'The use of AI in unlocking doors and cell phone security highlights the growing integration of AI in everyday devices for enhanced user experience and security measures.']}, {'end': 384.642, 'start': 207.853, 'title': 'Ai applications and impact', 'summary': 'Discusses the applications of ai, such as phone vr, machine translation, and self-driving cars, and highlights the potential for automation of tasks that a typical person can do in less than a second, impacting the nature of jobs.', 'duration': 176.789, 'highlights': ['AI applications like phone VR, machine translation, and self-driving cars are discussed, showcasing the potential for automation of tasks.', 'The potential for automating tasks that a typical person can do in less than a second is highlighted, with examples such as analyzing a picture and identifying known individuals.', 'The impact of AI on the nature of jobs is discussed, emphasizing the potential automation of jobs that involve a sequence of one-second tasks.']}], 'duration': 378.757, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs5885.jpg', 'highlights': ['AI is positioned to transform every major industry, similar to how electricity revolutionized various sectors.', 'Currently, 99% of the economic value created by AI is through learning A to B or input to output mappings.', 'Speech recognition by Google and Baidu accounts for over 10% of mobile searches, indicating significant user adoption and revenue volume.', "AI's applications in loan approval processes present a lucrative business opportunity, demonstrating the potential for significant financial gains.", 'AI applications like phone VR, machine translation, and self-driving cars showcase the potential for automation of tasks.', 'The potential for automating tasks that a typical person can do in less than a second is highlighted.']}, {'end': 636.066, 'segs': [{'end': 412.708, 'src': 'embed', 'start': 385.082, 'weight': 4, 'content': [{'end': 388.044, 'text': 'And so there are jobs that are at significant risk of automation.', 'start': 385.082, 'duration': 2.962}, {'end': 392.906, 'text': 'because these jobs are a sequence of one-second tasks that we string together.', 'start': 388.744, 'duration': 4.162}, {'end': 398.108, 'text': 'And so I think to the extent that those of us in tech are causing some of these job displacement problems.', 'start': 393.366, 'duration': 4.742}, {'end': 401.41, 'text': 'I think we have a responsibility to help address these issues as well.', 'start': 398.108, 'duration': 3.302}, {'end': 403.631, 'text': "And we can talk more about that later if you're interested.", 'start': 401.47, 'duration': 2.161}, {'end': 412.708, 'text': 'Now, Will mentioned the rise of deep learning, the rise of neural networks driving a lot of the recent excitement in AI.', 'start': 404.639, 'duration': 8.069}], 'summary': 'Some jobs are at significant risk of automation, with a focus on addressing job displacement and the rise of deep learning and neural networks in ai.', 'duration': 27.626, 'max_score': 385.082, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs385082.jpg'}, {'end': 474.651, 'src': 'embed', 'start': 449.893, 'weight': 2, 'content': [{'end': 457.24, 'text': 'then what happened with the digitization of society over the last 20 years is for many vertical industries, we start to collect more and more data.', 'start': 449.893, 'duration': 7.347}, {'end': 462.004, 'text': 'And because our smartphones, computers, digital activities generate data.', 'start': 458.081, 'duration': 3.923}, {'end': 468.509, 'text': 'And also, your health record is more likely 10 years ago, it would have been a physical piece of X-ray film.', 'start': 462.425, 'duration': 6.084}, {'end': 469.989, 'text': "Today, it's more likely a digital image.", 'start': 468.549, 'duration': 1.44}, {'end': 474.651, 'text': 'The state of supply chains used to be pieces of paper in filing cabinets all over the globe.', 'start': 470.37, 'duration': 4.281}], 'summary': 'Digitization has led to increased data collection across industries, with health records and supply chains transitioning to digital formats.', 'duration': 24.758, 'max_score': 449.893, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs449893.jpg'}, {'end': 566.288, 'src': 'embed', 'start': 538.723, 'weight': 1, 'content': [{'end': 545.811, 'text': 'First, this means that if you want the best levels of performance, you want to hit that level of performance, it means two things.', 'start': 538.723, 'duration': 7.088}, {'end': 550.475, 'text': "First, it really helps if you have a lot of data, if you're far to the right of this axis.", 'start': 546.291, 'duration': 4.184}, {'end': 555.54, 'text': "And, second, really helps if you're able to train a very large neural network.", 'start': 550.836, 'duration': 4.704}, {'end': 558.842, 'text': 'so, um, there has been a lot of algorithmic innovation.', 'start': 555.54, 'duration': 3.302}, {'end': 560.784, 'text': 'uh, in fact you you hear from actually, i think,', 'start': 558.842, 'duration': 1.942}, {'end': 566.288, 'text': "ian goodfellow later one of the tr35 awardees that's been driving some of the algorithmic innovation as well.", 'start': 560.784, 'duration': 5.504}], 'summary': 'To achieve high performance, abundant data and large neural networks are crucial, along with significant algorithmic innovation.', 'duration': 27.565, 'max_score': 538.723, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs538723.jpg'}, {'end': 595.631, 'src': 'embed', 'start': 572.152, 'weight': 0, 'content': [{'end': 579.878, 'text': 'as well as the availability of huge data sets, i think, have been two of the biggest drivers of this um recent rise in deep learning.', 'start': 572.152, 'duration': 7.726}, {'end': 585.062, 'text': "Now it turns out there's a lot of.", 'start': 581.799, 'duration': 3.263}, {'end': 589.626, 'text': 'there are also a lot of different algorithms being explored,', 'start': 585.062, 'duration': 4.564}, {'end': 595.631, 'text': "and I wanted to share with you my own perspective in terms of what's really most valuable today.", 'start': 589.626, 'duration': 6.005}], 'summary': 'Huge data sets and diverse algorithms drive recent rise in deep learning.', 'duration': 23.479, 'max_score': 572.152, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs572152.jpg'}], 'start': 385.082, 'title': "Tech's role in job displacement and neural network performance", 'summary': "Discusses the tech industry's responsibility in addressing job displacement due to automation and the impact of digitalization on data collection. it also explores the correlation between neural network size, data availability, and the rise of deep learning, emphasizing the significance of training large neural networks and algorithmic innovation.", 'chapters': [{'end': 449.893, 'start': 385.082, 'title': "Tech's responsibility in job displacement", 'summary': 'Discusses the responsibility of the tech industry in addressing job displacement due to automation, highlighting the rise of deep learning and neural networks as key factors driving recent excitement in ai, linked to the quantity of data and task performance.', 'duration': 64.811, 'highlights': ['The rise of deep learning and neural networks driving recent excitement in AI, with a focus on the amount of data and task performance.', "The tech industry's responsibility in addressing job displacement due to automation, emphasizing the need to help solve these issues.", 'Explanation of why neural networks and deep learning are taking off recently, linked to the quantity of data available for a task and its impact on performance.']}, {'end': 510.279, 'start': 449.893, 'title': 'Impact of digitalization on data collection', 'summary': 'Discusses the impact of digitalization on data collection in various industries, highlighting the transition from physical to digital data storage and the challenges faced by traditional ai algorithms in processing the increasing volume of data over the last 20 years.', 'duration': 60.386, 'highlights': ['The digitalization of society has led to the collection of increasing amounts of data from various vertical industries, including the transition of health records from physical to digital formats.', 'Supply chains, which were previously managed using physical documents, have now shifted to digital storage, resulting in a significant increase in data availability.', 'Traditional AI algorithms, such as support vector machines and logistic regression, have faced challenges in effectively processing the growing volume of data generated by digital activities over the past three to four years.']}, {'end': 636.066, 'start': 510.279, 'title': 'Impact of neural network size on performance', 'summary': 'Discusses the impact of neural network size on performance, highlighting the correlation between larger network size, data availability, and the rise of deep learning, with emphasis on the significance of training large neural networks and the role of algorithmic innovation and data availability.', 'duration': 125.787, 'highlights': ['Training larger neural networks leads to improved performance, emphasizing the importance of data availability and the ability to train very large neural networks.', 'Algorithmic innovation and the availability of huge datasets have been significant drivers of the recent rise in deep learning.', 'The rise of GPU computing and the availability of massive datasets have contributed to the ability to train very large neural networks, further enhancing performance.', 'Transfer learning, a type of machine learning, enables leveraging knowledge from one problem domain to another with less available data, exemplifying its application in recognizing X-ray radiology images.', 'Unsupervised learning is also mentioned as a type of machine learning, indicating the diverse algorithms being explored in the field of deep learning.']}], 'duration': 250.984, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs385082.jpg', 'highlights': ['The rise of deep learning and neural networks driving recent excitement in AI, with a focus on the amount of data and task performance.', 'Training larger neural networks leads to improved performance, emphasizing the importance of data availability and the ability to train very large neural networks.', 'The digitalization of society has led to the collection of increasing amounts of data from various vertical industries, including the transition of health records from physical to digital formats.', 'Algorithmic innovation and the availability of huge datasets have been significant drivers of the recent rise in deep learning.', "The tech industry's responsibility in addressing job displacement due to automation, emphasizing the need to help solve these issues."]}, {'end': 1315.422, 'segs': [{'end': 699.086, 'src': 'embed', 'start': 676.044, 'weight': 2, 'content': [{'end': 685.175, 'text': 'I think reinforcement learning is one class of technology where the PR excitement is vastly disproportional relative to the actual deployments today.', 'start': 676.044, 'duration': 9.131}, {'end': 689.6, 'text': 'Turns out that supervised learning has a huge hunger for data.', 'start': 686.096, 'duration': 3.504}, {'end': 693.803, 'text': 'Reinforcement learning has an even greater hunger for data.', 'start': 690.701, 'duration': 3.102}, {'end': 696.765, 'text': "so it turns out to be great for playing video games, because if you're playing video games,", 'start': 693.803, 'duration': 2.962}, {'end': 699.086, 'text': 'you can play an infinite number of video games on the computer.', 'start': 696.765, 'duration': 2.321}], 'summary': 'Reinforcement learning has greater hunger for data than supervised learning, advantageous for playing infinite video games.', 'duration': 23.042, 'max_score': 676.044, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs676044.jpg'}, {'end': 766.324, 'src': 'embed', 'start': 738.651, 'weight': 1, 'content': [{'end': 744.553, 'text': 'In particular, supervised learning requires a lot of data with both the input A and the output B.', 'start': 738.651, 'duration': 5.902}, {'end': 749.074, 'text': 'So for speech recognition, A was the audio clip and B was the text transcript.', 'start': 744.553, 'duration': 4.521}, {'end': 754.476, 'text': 'So today, leading speech recognition systems are built using north of 50, 000 hours.', 'start': 749.514, 'duration': 4.962}, {'end': 760.32, 'text': "there's about five years of audio data, which is a very expensive and very difficult asset to collect.", 'start': 754.476, 'duration': 5.844}, {'end': 766.324, 'text': 'You have 50, 000 hours of data together with transcripts for all five years of continuous audio, all 50, 000 hours of data.', 'start': 760.82, 'duration': 5.504}], 'summary': 'Leading speech recognition systems use over 50,000 hours of audio data with transcripts.', 'duration': 27.673, 'max_score': 738.651, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs738651.jpg'}, {'end': 891.47, 'src': 'embed', 'start': 840.449, 'weight': 0, 'content': [{'end': 846.174, 'text': 'to build a competitive, merely decent web search engine without access to that data asset.', 'start': 840.449, 'duration': 5.725}, {'end': 851.438, 'text': 'So this is one example of where that data asset, even more than the specific algorithms are, I think,', 'start': 846.234, 'duration': 5.204}, {'end': 855.322, 'text': 'makes the web search business incredibly defensible.', 'start': 851.438, 'duration': 3.884}, {'end': 863.409, 'text': 'Having said that, there is also maybe a little bit of excessive PR about the importance of data.', 'start': 856.023, 'duration': 7.386}, {'end': 868.594, 'text': 'Because for new vertical categories, for new applications, maybe a new product idea that you have,', 'start': 863.489, 'duration': 5.105}, {'end': 874.298, 'text': 'it is often possible for you to get started even with a modest amount of data.', 'start': 869.915, 'duration': 4.383}, {'end': 884.025, 'text': 'So when I plan out and launch new products, I often plan out this type of virtuous circle.', 'start': 874.799, 'duration': 9.226}, {'end': 891.47, 'text': "So you often see if you're launching a new product, get a little bit of data that allows you to launch a product.", 'start': 884.225, 'duration': 7.245}], 'summary': 'Data asset is crucial for web search, but new products can start with modest data.', 'duration': 51.021, 'max_score': 840.449, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs840449.jpg'}, {'end': 944.528, 'src': 'embed', 'start': 915.541, 'weight': 4, 'content': [{'end': 917.163, 'text': 'So I knew them when they were Stanford students.', 'start': 915.541, 'duration': 1.622}, {'end': 919.445, 'text': 'They were actually taking my class when they were Stanford students.', 'start': 917.183, 'duration': 2.262}, {'end': 926.391, 'text': 'And so they have built these big rigs that go up and down fields, agricultural fields,', 'start': 920.165, 'duration': 6.226}, {'end': 934.458, 'text': 'and takes pictures of lettuce and makes life and death decisions about which head of lettuce shall live and which head of lettuce shall die.', 'start': 926.391, 'duration': 8.067}, {'end': 937.981, 'text': 'So when they were students at Stanford, I saw them.', 'start': 934.938, 'duration': 3.043}, {'end': 944.528, 'text': 'They were going out to these fields and using their cameras and just taking a bunch of pictures of these heads of lettuce in the field.', 'start': 938.021, 'duration': 6.507}], 'summary': 'Stanford students built rigs for lettuce evaluation.', 'duration': 28.987, 'max_score': 915.541, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs915541.jpg'}, {'end': 1004.914, 'src': 'embed', 'start': 977.217, 'weight': 5, 'content': [{'end': 980.758, 'text': 'of pictures of heads of lettuce kind of half buried in the dirt.', 'start': 977.217, 'duration': 3.541}, {'end': 987.98, 'text': 'And this actually makes their business, in my opinion, pretty defensible, because even the giant tech companies, the global giant tech companies,', 'start': 981.378, 'duration': 6.602}, {'end': 991.04, 'text': 'as far as I know, do not have this particular data asset.', 'start': 987.98, 'duration': 3.06}, {'end': 999.482, 'text': 'And this actually makes their business difficult, even for the very large tech companies, at least challenging for them to enter.', 'start': 991.7, 'duration': 7.782}, {'end': 1004.914, 'text': 'This is one example.', 'start': 1003.712, 'duration': 1.202}], 'summary': "Company's unique data assets make it difficult for even giant tech companies to enter their business.", 'duration': 27.697, 'max_score': 977.217, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs977217.jpg'}, {'end': 1156.039, 'src': 'embed', 'start': 1131.126, 'weight': 6, 'content': [{'end': 1138.808, 'text': "for example, internet companies engage in pervasive a b testing because you can on the internet, and if you don't,", 'start': 1131.126, 'duration': 7.682}, {'end': 1141.488, 'text': "your competitors will and they'll learn faster than you.", 'start': 1138.808, 'duration': 2.68}, {'end': 1144.009, 'text': 'um, internet companies tend to have very short cycle times.', 'start': 1141.488, 'duration': 2.521}, {'end': 1152.198, 'text': 'We are used to shipping a product every day, or maybe shipping a product every week because we can on the internet,', 'start': 1145.576, 'duration': 6.622}, {'end': 1156.039, 'text': 'whereas a traditional retailer might redesign their stores every six months.', 'start': 1152.198, 'duration': 3.841}], 'summary': 'Internet companies conduct frequent a/b testing and have shorter cycle times, such as shipping products daily or weekly, compared to traditional retailers.', 'duration': 24.913, 'max_score': 1131.126, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1131126.jpg'}, {'end': 1201.503, 'src': 'heatmap', 'start': 1177.016, 'weight': 1, 'content': [{'end': 1183.478, 'text': 'and so you know, in a traditional retailer maybe a CEO can make all the decisions and then everyone just does what the CEO says.', 'start': 1177.016, 'duration': 6.462}, {'end': 1188.079, 'text': 'but Internet companies only the engineers and the product managers know a lot of the best decisions.', 'start': 1183.478, 'duration': 4.601}, {'end': 1190.22, 'text': 'you have to push decision-making power down to them,', 'start': 1188.079, 'duration': 2.141}, {'end': 1196.281, 'text': "and I've been actually heavily influenced by some of Jeff Bezos' thinking speaking on some of these.", 'start': 1190.22, 'duration': 6.061}, {'end': 1201.503, 'text': 'But to me, these are the things that truly define an internet company, not whether or not you run a website.', 'start': 1196.681, 'duration': 4.822}], 'summary': "Internet companies empower engineers and product managers for decision-making, influenced by jeff bezos' thinking.", 'duration': 24.487, 'max_score': 1177.016, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1177016.jpg'}, {'end': 1274.144, 'src': 'embed', 'start': 1249.304, 'weight': 7, 'content': [{'end': 1256.65, 'text': 'I think Google and Baidu are truly phenomenal and among the foremost thought leaders in figuring out what it really means to be a truly AI company.', 'start': 1249.304, 'duration': 7.346}, {'end': 1259.113, 'text': "I don't think any of us fully know what this is yet.", 'start': 1257.071, 'duration': 2.042}, {'end': 1266.098, 'text': "Just as maybe 20 years ago, We didn't really know 20 years ago that A-B testing was going to be such a big thing.", 'start': 1260.194, 'duration': 5.904}, {'end': 1267.699, 'text': "I think we're still figuring out one of these things.", 'start': 1266.158, 'duration': 1.541}, {'end': 1271.822, 'text': 'But these are some things that I think might turn out to be true.', 'start': 1268.179, 'duration': 3.643}, {'end': 1274.144, 'text': 'One is strategic data acquisition.', 'start': 1272.002, 'duration': 2.142}], 'summary': 'Google and baidu are leading in ai, still figuring it out, strategic data acquisition is important.', 'duration': 24.84, 'max_score': 1249.304, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1249304.jpg'}], 'start': 636.066, 'title': "Ai's economic value and corporate strategy", 'summary': 'Delves into the economic value and deployment challenges of ai algorithms, emphasizing the disproportionate excitement about reinforcement learning, the significance of data assets in web search technology, and the changing corporate strategy in the ai era, including the uncertainties surrounding the definition of a truly ai company and the importance of strategic data acquisition.', 'chapters': [{'end': 914.401, 'start': 636.066, 'title': 'Ai algorithms and economic value', 'summary': 'Discusses the economic value and deployment challenges of ai algorithms, highlighting the disproportionate excitement about reinforcement learning, the hunger for data in supervised and reinforcement learning, the shifting corporate strategy due to open-source algorithms, and the significance of data assets in web search technology.', 'duration': 278.335, 'highlights': ['Leading web search companies utilize data assets to provide more relevant results, making the web search business incredibly defensible. Data assets in leading web search companies enable them to offer more relevant results based on user location and preferences, making the web search business highly defensible.', 'Supervised learning requires a significant amount of data, with leading speech recognition systems built using over 50,000 hours of audio data, which is expensive and difficult to collect. Leading speech recognition systems require over 50,000 hours of audio data and transcripts, making data collection expensive and challenging.', 'Reinforcement learning has a disproportionate PR excitement relative to its actual deployments, and it has a greater hunger for data than supervised learning. Reinforcement learning receives disproportionate PR excitement compared to its actual deployments and has a greater hunger for data than supervised learning.', 'New vertical categories, applications, or product ideas can often be launched with a modest amount of data, leading to a positive feedback loop and the potential for a defensible business over time. New vertical categories, applications, or product ideas can be launched with modest data, leading to a positive feedback loop and the potential for a defensible business over time.']}, {'end': 991.04, 'start': 915.541, 'title': 'Lettuce farming innovation', 'summary': "Highlights the journey of two stanford students who utilized their data asset of lettuce images to build a successful agricultural decision-making product, possibly leading to the world's largest data asset of its kind.", 'duration': 75.499, 'highlights': ['The founders, former Stanford students, developed rigs to capture images of lettuce in agricultural fields, enabling life-and-death decisions for lettuce crops.', 'They initially built a modest data asset from lettuce images, sufficient for farmers to adopt their product.', "As they expanded their user base, their data asset grew, potentially becoming the world's largest of its kind.", 'Their unique data asset provides a defensible advantage, as even global tech giants lack this particular type of data.']}, {'end': 1315.422, 'start': 991.7, 'title': 'Corporate strategy in the ai era', 'summary': 'Discusses how corporate strategy is changing in the ai era, emphasizing the difference between traditional and internet companies, and the uncertainties surrounding the definition of a truly ai company. it also highlights the importance of strategic data acquisition in building a defensible business.', 'duration': 323.722, 'highlights': ['Difference between traditional and internet companies Internet companies engage in pervasive A/B testing, have short cycle times, and push decision-making power down to engineers and product managers, whereas traditional retailers have longer cycle times and centralized decision-making. This is exemplified by the difference between a shopping mall with a website and a truly internet company like Amazon.', 'Defining a truly AI company The speaker emphasizes that adding neural networks or deep learning to a traditional tech company does not make it an AI company. The definition of a truly AI company is still uncertain, with Google and Baidu considered thought leaders in this space. Strategic data acquisition, such as planning a path for data acquisition resulting in a defensible business, is highlighted as a key criterion.']}], 'duration': 679.356, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs636066.jpg', 'highlights': ['Leading web search companies utilize data assets to provide more relevant results, making the web search business incredibly defensible.', 'Supervised learning requires over 50,000 hours of audio data, making data collection expensive and challenging.', 'Reinforcement learning receives disproportionate PR excitement compared to its actual deployments and has a greater hunger for data than supervised learning.', 'New vertical categories, applications, or product ideas can be launched with modest data, leading to a positive feedback loop and the potential for a defensible business over time.', 'The founders developed rigs to capture images of lettuce in agricultural fields, enabling life-and-death decisions for lettuce crops.', 'Their unique data asset provides a defensible advantage, as even global tech giants lack this particular type of data.', 'Internet companies engage in pervasive A/B testing, have short cycle times, and push decision-making power down to engineers and product managers, whereas traditional retailers have longer cycle times and centralized decision-making.', 'The definition of a truly AI company is still uncertain, with Google and Baidu considered thought leaders in this space.']}, {'end': 1753.029, 'segs': [{'end': 1336.661, 'src': 'embed', 'start': 1315.562, 'weight': 0, 'content': [{'end': 1326.774, 'text': 'And so the leading AI organizations are starting to think through these frankly sort of multi-year chess games to acquire the data asset that allows you to build a defensible business.', 'start': 1315.562, 'duration': 11.212}, {'end': 1329.997, 'text': 'And this data acquisition strategy is really complicated.', 'start': 1327.515, 'duration': 2.482}, {'end': 1333.12, 'text': "Here's one you can go home and implement.", 'start': 1331.579, 'duration': 1.541}, {'end': 1336.661, 'text': 'AI organizations tend to build unified data warehouses.', 'start': 1333.64, 'duration': 3.021}], 'summary': 'Leading ai organizations are strategizing multi-year chess games to acquire data assets for building defensible businesses, often through unified data warehouses.', 'duration': 21.099, 'max_score': 1315.562, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1315562.jpg'}, {'end': 1402.798, 'src': 'embed', 'start': 1362.737, 'weight': 1, 'content': [{'end': 1365.699, 'text': 'Based on your web search, we know what type of food you like.', 'start': 1362.737, 'duration': 2.962}, {'end': 1367.761, 'text': "And based on your GPS, we know you're in Boston.", 'start': 1365.779, 'duration': 1.982}, {'end': 1370.763, 'text': 'Well, that could be put together to recommend a new restaurant to you for tonight.', 'start': 1367.801, 'duration': 2.962}, {'end': 1376.848, 'text': 'So the only way for those things to happen, either for the engineer or for the software to look at the data,', 'start': 1371.524, 'duration': 5.324}, {'end': 1384.291, 'text': "is if it's in the unified data warehouse which which, which significantly increases the odds of Your teams, be able to connect the dots.", 'start': 1376.848, 'duration': 7.443}, {'end': 1388.553, 'text': 'So this is one and this is often a multi exercise for companies to implement that.', 'start': 1384.311, 'duration': 4.242}, {'end': 1400.997, 'text': 'but and then I think what internet companies tend to be very good at spotting pervasive automation opportunities and and And also, you know,', 'start': 1388.553, 'duration': 12.444}, {'end': 1402.798, 'text': 'new job descriptions right?', 'start': 1400.997, 'duration': 1.801}], 'summary': 'Unified data warehouse enhances recommendation accuracy for restaurant selection.', 'duration': 40.061, 'max_score': 1362.737, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1362737.jpg'}, {'end': 1486.455, 'src': 'embed', 'start': 1454.573, 'weight': 3, 'content': [{'end': 1460.036, 'text': "And this is part of the workflow we've figured out in tech in Silicon Valley or in the tech world,", 'start': 1454.573, 'duration': 5.463}, {'end': 1463.298, 'text': 'in order to have product managers and engineers work together.', 'start': 1460.036, 'duration': 3.262}, {'end': 1465, 'text': 'But this is breaking down.', 'start': 1463.919, 'duration': 1.081}, {'end': 1469.162, 'text': 'But these old workflows and processes are breaking down in the AI era.', 'start': 1465.12, 'duration': 4.042}, {'end': 1478.889, 'text': "So, for example, if you're trying to build a chatbot, if a product manager draws, you know, I don't know this picture right.", 'start': 1469.563, 'duration': 9.326}, {'end': 1486.455, 'text': "so so actually, one of the companies i've been helping out with is a chatbot cognitive behavioral therapy chatbot.", 'start': 1478.889, 'duration': 7.566}], 'summary': 'Old tech workflows breaking down in ai era, impacting product manager and engineer collaboration.', 'duration': 31.882, 'max_score': 1454.573, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1454573.jpg'}, {'end': 1634.278, 'src': 'embed', 'start': 1605.003, 'weight': 4, 'content': [{'end': 1609.765, 'text': 'and then to matrix in the AI talent into your various different business units.', 'start': 1605.003, 'duration': 4.762}, {'end': 1612.066, 'text': "So why do I recommend this? Let's take an example.", 'start': 1610.145, 'duration': 1.921}, {'end': 1616.028, 'text': "Let's say this business unit is your gift card division.", 'start': 1612.507, 'duration': 3.521}, {'end': 1625.993, 'text': "Your head of gift card might be a great leader, but there's a good chance that they don't have the capability to select and retain the top AI talent.", 'start': 1617.489, 'duration': 8.504}, {'end': 1629.735, 'text': 'And why would the top AI people want to work for your gift card division?', 'start': 1626.013, 'duration': 3.722}, {'end': 1634.278, 'text': 'So I think that if you can build a centralized AI team,', 'start': 1630.775, 'duration': 3.503}], 'summary': 'Integrate ai talent into business units for enhanced capabilities and retention.', 'duration': 29.275, 'max_score': 1605.003, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1605003.jpg'}, {'end': 1753.029, 'src': 'embed', 'start': 1727.718, 'weight': 5, 'content': [{'end': 1740.803, 'text': 'we now have tools for your chief learning officer or your central AI unit to provide very broad-based employee training to just level up your whole organization to better understand how to incorporate these AI processes in the businesses.', 'start': 1727.718, 'duration': 13.085}, {'end': 1742.864, 'text': 'I hope that was helpful.', 'start': 1741.943, 'duration': 0.921}, {'end': 1744.344, 'text': "I'm definitely running over time.", 'start': 1743.064, 'duration': 1.28}, {'end': 1746.085, 'text': 'Let me say thank you very much.', 'start': 1744.644, 'duration': 1.441}, {'end': 1753.029, 'text': 'Thank you.', 'start': 1752.483, 'duration': 0.546}], 'summary': "Tools for ai training to level up organization's understanding.", 'duration': 25.311, 'max_score': 1727.718, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1727718.jpg'}], 'start': 1315.562, 'title': 'Ai strategies and workflow challenges', 'summary': "Discusses leading ai organizations' data acquisition strategies and complexities, emphasizing the importance of unified data warehouses and personalized recommendations. it also addresses challenges in adapting workflows to the ai era, emphasizing the need for new processes and centralized ai teams to leverage ai capabilities.", 'chapters': [{'end': 1376.848, 'start': 1315.562, 'title': 'Ai data acquisition strategy', 'summary': 'Discusses how leading ai organizations are strategically acquiring data assets for building defensible businesses, emphasizing the complexity of data acquisition strategies and the importance of unified data warehouses, with an example of leveraging data for personalized recommendations.', 'duration': 61.286, 'highlights': ['AI organizations are strategically planning multi-year data acquisition strategies to build defensible businesses.', 'Unified data warehouses are crucial for combining data from multiple silos, especially in large companies with numerous databases.', 'Example of leveraging data for personalized recommendations based on web search and GPS data for recommending a new restaurant.']}, {'end': 1753.029, 'start': 1376.848, 'title': 'Challenges in adapting workflows for ai era', 'summary': 'Discusses the challenges in adapting traditional job descriptions and workflows to the ai era, emphasizing the need for new language and processes to effectively collaborate between product managers and engineers, along with the recommendation to form a centralized ai team and matrix ai talent into various business units to leverage ai capabilities.', 'duration': 376.181, 'highlights': ['The need for new language and processes to effectively collaborate between product managers and engineers, particularly in adapting traditional job descriptions and workflows to the AI era. Adapting workflows and processes to the AI era, emphasizing the need for new language and processes to collaborate effectively between product managers and engineers.', 'Recommendation to form a centralized AI team and matrix AI talent into various business units to leverage AI capabilities, with the aim of establishing consistent hiring and promotion practices in the AI era. Recommendation to form a centralized AI team and matrix AI talent into various business units to leverage AI capabilities, establishing consistent hiring and promotion practices in the AI era.', 'The suggestion to provide broad-based employee training through digital content platforms to enhance the understanding and incorporation of AI processes in businesses. Suggestion to provide broad-based employee training through digital content platforms to enhance the understanding and incorporation of AI processes in businesses.']}], 'duration': 437.467, 'thumbnail': 'https://coursnap.oss-ap-southeast-1.aliyuncs.com/video-capture/NKpuX_yzdYs/pics/NKpuX_yzdYs1315562.jpg', 'highlights': ['AI organizations strategically plan multi-year data acquisition strategies for defensible businesses.', 'Unified data warehouses are crucial for combining data from multiple silos, especially in large companies.', 'Leveraging data for personalized recommendations based on web search and GPS data is exemplified.', 'Adapting workflows and processes to the AI era requires new language and collaboration between product managers and engineers.', 'Forming a centralized AI team and matrix AI talent into various business units is recommended to leverage AI capabilities.', 'Providing broad-based employee training through digital content platforms enhances understanding and incorporation of AI processes.']}], 'highlights': ['AI is positioned to transform every major industry, similar to how electricity revolutionized various sectors.', 'Currently, 99% of the economic value created by AI is through learning A to B or input to output mappings.', 'Speech recognition by Google and Baidu accounts for over 10% of mobile searches, indicating significant user adoption and revenue volume.', "AI's applications in loan approval processes present a lucrative business opportunity, demonstrating the potential for significant financial gains.", 'AI applications like phone VR, machine translation, and self-driving cars showcase the potential for automation of tasks.', 'The potential for automating tasks that a typical person can do in less than a second is highlighted.', 'The rise of deep learning and neural networks driving recent excitement in AI, with a focus on the amount of data and task performance.', 'Training larger neural networks leads to improved performance, emphasizing the importance of data availability and the ability to train very large neural networks.', 'The digitalization of society has led to the collection of increasing amounts of data from various vertical industries, including the transition of health records from physical to digital formats.', 'Algorithmic innovation and the availability of huge datasets have been significant drivers of the recent rise in deep learning.', "The tech industry's responsibility in addressing job displacement due to automation, emphasizing the need to help solve these issues.", 'Leading web search companies utilize data assets to provide more relevant results, making the web search business incredibly defensible.', 'Supervised learning requires over 50,000 hours of audio data, making data collection expensive and challenging.', 'Reinforcement learning receives disproportionate PR excitement compared to its actual deployments and has a greater hunger for data than supervised learning.', 'New vertical categories, applications, or product ideas can be launched with modest data, leading to a positive feedback loop and the potential for a defensible business over time.', 'The founders developed rigs to capture images of lettuce in agricultural fields, enabling life-and-death decisions for lettuce crops.', 'Their unique data asset provides a defensible advantage, as even global tech giants lack this particular type of data.', 'Internet companies engage in pervasive A/B testing, have short cycle times, and push decision-making power down to engineers and product managers, whereas traditional retailers have longer cycle times and centralized decision-making.', 'The definition of a truly AI company is still uncertain, with Google and Baidu considered thought leaders in this space.', 'AI organizations strategically plan multi-year data acquisition strategies for defensible businesses.', 'Unified data warehouses are crucial for combining data from multiple silos, especially in large companies.', 'Leveraging data for personalized recommendations based on web search and GPS data is exemplified.', 'Adapting workflows and processes to the AI era requires new language and collaboration between product managers and engineers.', 'Forming a centralized AI team and matrix AI talent into various business units is recommended to leverage AI capabilities.', 'Providing broad-based employee training through digital content platforms enhances understanding and incorporation of AI processes.']}