<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Jerry Lu on Medium]]></title>
        <description><![CDATA[Stories by Jerry Lu on Medium]]></description>
        <link>https://medium.com/@thejerrylu?source=rss-242a8d3b145------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*p3lMN-MT_DLiVOsSiHl3mw.png</url>
            <title>Stories by Jerry Lu on Medium</title>
            <link>https://medium.com/@thejerrylu?source=rss-242a8d3b145------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 07:41:28 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thejerrylu/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Joining Advancit Capital]]></title>
            <link>https://medium.com/@thejerrylu/joining-advancit-capital-cf42cfb013b6?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/cf42cfb013b6</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[consumer]]></category>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Tue, 25 Feb 2020 19:03:27 GMT</pubDate>
            <atom:updated>2020-02-25T19:03:27.016Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uujt4KAyM3v8-eRu6n7aCw.jpeg" /><figcaption>Photo by <a href="https://www.pexels.com/@mikebirdy?utm_content=attributionCopyText&amp;utm_medium=referral&amp;utm_source=pexels">Mike</a> from <a href="https://www.pexels.com/photo/acoustic-amplifier-artist-audio-114820/?utm_content=attributionCopyText&amp;utm_medium=referral&amp;utm_source=pexels">Pexels</a></figcaption></figure><h4><strong>“Content is king.”</strong></h4><p>This statement became the mantra of the media industry back in the 1980s and it still rings true today. But in today’s content-heavy world, that mantra is being matched in equal measure by the critical influence of the digital consumer.</p><p>Looking back both personally and professionally, I see that I’ve always had a hand in content.</p><ul><li>From writing and curating music <a href="https://www.treehouse707.com/blog/author/By-JERRY-LU.">content</a> for a small collective in London, to …</li><li>Building premium <a href="https://www.youtube.com/premium">content</a> at YouTube with the launch and growth of YouTube Red and YouTube Originals, to…</li><li>Exploring the future of immersive <a href="https://medium.com/hackernoon/from-sci-fi-to-commercialization-the-rise-of-digital-avatars-ea164495c377">content</a> via the world of virtual avatars at Lux Capital.</li></ul><p>I’ve had a ringside seat as the importance of content grew everywhere I looked in our fragmented culture. As media has become democratized and consumers have become digital natives, content has evolved in great leaps, taking shape in various forms, being monetized through new business models, and being shared across different platforms.</p><p>Now, I’m excited to announce that I’ll be continuing to work with content, but in a new role: I’m joining the investment team at <a href="http://www.advancitcapital.com/">Advancit Capital</a>, where I’ll be investing in technology and the digital consumer as it impacts the media landscape.</p><p>From my first introduction to Advancit, I was enamored by their portfolio of companies, companies transforming the next generation of media. From their early investments in companies such as Maker Studios, <a href="https://www.headspace.com/">Headspace</a>, and <a href="https://www.masterclass.com/">Masterclass</a> to their most recent investments, Advancit’s story contains a narrative that defines how consumers are engaging with content today.</p><p>And as the definition of content has changed, the ability to engage an audience and monetize the content has expanded significantly, moving from traditional advertising revenue to new business models such as commerce, subscriptions, and community experiences.</p><p>Some of their recent investments exemplify this orientation:</p><ul><li><a href="https://100thieves.com/">100 Thieves</a> — a lifestyle brand and gaming organization built on the back of content and commerce</li><li><a href="https://theathletic.com/">The Athletic</a> — a subscription-based digital sports media company offering high-quality local journalism content</li><li><a href="https://wondery.com/">Wondery</a> — a community of storytellers that creates and curates immersive podcasts</li><li>…among many others</li></ul><p>It’s an exciting time to be involved with these cutting-edge companies, which span multiple categories like digital subscription services, gaming and e-sports, digital health and wellness, creative AI, future of the worker (content/training for the employee), and lifestyle brands. Content, of course, is at the center of it all, but the true underlying force is the engagement of the digital consumer.</p><p>More importantly, I’m incredibly excited to work with and learn from <a href="https://en.wikipedia.org/wiki/Shari_Redstone">Shari</a>, <a href="https://twitter.com/jost">Jason</a>, <a href="https://twitter.com/alligoldberg">Allison</a>, and <a href="https://en.wikipedia.org/wiki/Jonathan_Miller_(businessman)">Jon</a> who share deep industry knowledge and extensive, unparalleled access to industry relationships. I can’t think of a better place to support the entrepreneurs transforming the media landscape in an increasingly digital consumer world and take part in driving the mantra forward.</p><p>Because even in this digital landscape, content is still king.</p><pre>If you&#39;re an early-stage founder building what the next generation of media will look like - feel free to reach out!</pre><pre>Follow me: <a href="https://twitter.com/thejerrylu"><strong>@thejerrylu</strong></a><strong><br></strong>E-mail: <strong>jerry</strong> [at] <strong>advancitcap </strong>[dot]<strong> com</strong></pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cf42cfb013b6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Fake News Arms Race]]></title>
            <link>https://medium.com/hackernoon/the-fake-news-arms-race-448675592803?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/448675592803</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[fake-news]]></category>
            <category><![CDATA[the-fake-news-arms-race]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[medium]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Tue, 07 Aug 2018 13:27:52 GMT</pubDate>
            <atom:updated>2019-04-24T10:11:43.671Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*PknaAYSLSUux6ywMTTBPxQ.jpeg" /></figure><p>Since the 2016 US presidential election, the term “fake news” has become part of our daily vernacular. Fake news has become so prevalent in our society that even the most sophisticated of us fall for it sometimes. It has reached such epidemic proportions that French President Macron recently commented, “we must fight against the virus of fake news.”</p><p>In fact, the <a href="http://reports.weforum.org/global-risks-2013/section-seven-online-only-content/data-explorer/">World Economic Forum</a> ranks this massive digital spread of misinformation among the top future global risks, along with failure to adapt to climate change, organized crime, and the food shortage crisis.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fV-brFjl88ZzBpBk4IVcvQ.png" /><figcaption>World Economic Forum — <a href="http://reports.weforum.org/global-risks-2013/section-seven-online-only-content/data-explorer/">Global Risks 2013 Eighth Edition</a></figcaption></figure><p>We’re living in an information “<a href="https://hackernoon.com/tagged/arms-race">arms race</a>” where media and technology companies rush to develop new defenses as fake news techniques proliferate.</p><p>But who should fix these problems? The first thing to establish is who’s responsible for fake news. Some people like to blame social media and their lack of response to this new epidemic. Others like to add that it is in the nature of fake news to be extremely shareable — exactly like the legitimate content the news media has created its new distribution models around.</p><p>In my opinion, improving the ability of platforms and media to address the fake news phenomenon requires taking a more holistic approach, identifying areas in every part of the value chain — from creation all the way to circulation and discovery — that need to change.</p><p>As the problem has become more widely acknowledged, fighting fake news has proven to be fertile ground for <a href="https://hackernoon.com/tagged/startups">startups</a>. There are lots of different startups fighting fake news, and they’re going about it in many different ways. For instance, some aim to tackle the huge challenge of fighting falsehoods on social media platforms, while others aim to protect brands and their reputation from misinformation. As a result, there are a number of different markets in which startups have the opportunity to succeed.</p><p>In order to pinpoint where these opportunities and challenges lie, I’ve broken down fake news into four distinct stages: 1) creation, 2) publication, 3) circulation, and 4) discovery. I’ll discuss those stages below and highlight startups working to address various problems within each stage.</p><h3>Value Chain</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2Es6CclFQsSMZJqa4Cbm9A.png" /></figure><h3>Startup Map</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_h8Jg0xGyIw5U2YC" /></figure><h4>Creation</h4><p>An attacker looking to spread misinformation will need to create content that they can then distribute widely. To do this, they can employ two broad methods: 1) create fake content and/or 2) create fake accounts with which to spread fake content. An example of a company attacking the fake content issue is <a href="https://knowherenews.com/">Knowhere</a>, an AI powered news site, which generates unbiased news stories. Looking at all sites (left-leaning, right-leaning, etc.), Knowhere writes its own “impartial” version of a story based on what it finds. Tackling fake accounts, another company, <a href="https://www.perimeterx.com/">PerimeterX</a>, tracks the behavior of website visitors and uses artificial intelligence to determine which are bots and which ones should be blocked. Operating under the thesis that fake news has an army of distribution bots at its core, PerimeterX aims to stop malicious bots in their tracks before they attack.</p><h4>Publication</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-0GwasB0hOlv2InX" /></figure><p>Once content has been created, an attacker will need to publish their misinformation online. Most fake-news-fighting companies operate at this stage, looking to tackle the challenge of deciphering between fact or fiction. Although this is technologically a hard problem to solve given the nuances of human understanding, I view the startups operating solely in this stage as projects rather than ventures, as there is no clear business model and it’s very much a game of whack-a-mole that requires automation at scale. As a result, you see grassroots efforts like the <a href="http://www.fakenewschallenge.org/">Fake News Challenge</a>, a competition run by volunteers in the AI community, fostering the development of tools that help fact-checkers do their jobs more efficiently.</p><h4>Circulation</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/820/0*M7FO512bYnMMCJI4" /></figure><p>Like any good news story, fake content will be shared, liked, reposted, and distributed across many different platforms and channels. And while the impact on individual consumers might not be great, advertisers can lose business by hosting bad content and brands can be damaged by being displayed next to bad content. In fact, a June 2017 study from Yahoo’s Brightroll found that 96% of programmatic decision-makers are concerned about fake news as it relates to their business. Of those surveyed, 31% said they will respond to the issue by reducing spend with programmatic partners whose inventory associates brands with fake news. As such, startups like Israeli-based <a href="http://adverifai.com/">AdVerif.ai</a> are looking to help ad agencies not only identify fake stories to ensure their brand doesn’t appear alongside them, but also defund the bad actors who are producing low-quality content.</p><h4>Discovery</h4><p>The more widely a piece of misinformation can be spread, the better the chances that it will capture the public imagination and achieve its objective. In order to prevent the discovery of such content, some startups are taking a decentralized approach whereby an editorial board no longer determines what’s worth reading: instead, users, developers, and publishers run custom rankings to produce search results. One such company is <a href="http://userfeeds.io/">Userfeeds</a>, which is using Blockchain technology to create transparent and publicly auditable content networks and ranking algorithms. A decentralized platform gives publishers and application developers new business models, outside of advertising and subscriptions, and potentially improves “organic” content discovery for audiences that doesn’t rely on easily abused social media signals such as links, likes, and votes.</p><h3>Full-stack Solutions</h3><p>The companies I highlighted above were startups operating in singular stages, but if you take a look at the market map, you’ll notice three companies operating more as a full-stack solution, whereby the output of one stage is fed as inputs into another to create a data feedback loop, a competitive moat.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iTs4JeteMQwRZsxUtbOxBg.png" /></figure><p>It’s clear that, as attackers continue to evolve their techniques and innovate, so too will defenders, whether it’s:</p><ul><li><a href="https://factmata.com/">Factmata</a>’s approach of building a fact-checking platform to create content credibility scores that it uses to partner with ad networks or;</li><li><a href="https://robhat.com/">Robhat Labs</a>’ approach of using image data to protect consumers from fake news as they browse or;</li><li><a href="http://newknowledge.io/">New Knowledge</a>’s approach of building a platform to identify the coordination of attacks and providing tools to businesses to best respond.</li></ul><p>And as I think through the opportunities for why these companies are the most likely to succeed, I boil it down to three core themes:</p><h4>Dual-product business model</h4><p>Similar to the way a computer virus spreads from host to host, misinformation spreads like wildfire from person to person. For example, a recent MIT <a href="http://science.sciencemag.org/content/359/6380/1146">study</a> showed that fake news spreads six times faster than the truth on Twitter. Continuing this virus analogy, I see companies tackling fake news in the same way they responded to anti-virus software back in the 1990s and 2000s; Symantec is an example of an appropriate consumer-plus enterprise business model worth replicating today.</p><p>By providing a free consumer service, a company that roots out and identifies fake news can gather data at little to no cost and build up a base of knowledge about where all harmful content resides. This data can then be sold to enterprises, providing a layer of analytics and insights. The consumer product is thus a means to the end of selling to businesses.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/641/0*ZxEjXml2i6QUA9Xm" /></figure><p>Once the service compiles this data, one potential commercialization can be around brand security. The social media management / analytics market, of which reputation management is a sub-segment, is estimated to grow to $9.54B by 2022 and is driven largely by the increased access to information. Brands today are unaware of where they are going to appear and know that their reputation can be damaged in a very short time. Their brand may appear next to undesirable or inappropriate content, or other content may spread unfair, deceptive, or non-credible information about them. Companies are increasingly looking for solutions that not only protect their brands but give them insights into how their brands are being portrayed.</p><h4>Coordination over content</h4><p>At this point, I’ve seen too many startups building businesses around a fact-checking service that operates as some extension of Google Chrome. The problem with whitelisting news sources is that you’re handing the advantage to incumbent news sources — which means there can be very little innovation in this space.</p><p>What I’m much more interested in seeing are startups moving up the stack toward the source of the content and trying to identify the coordination of activities rather than the veracity of the data, which is a much smaller market.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*MsZtdkkcWAKvm2Uc" /></figure><p>If you have a network of things being shared and a network of people sharing them, there are orders of magnitude more people sharing “interesting” things than things being shared. The content itself doesn’t change nearly as fast as users do. Moreover, anybody identifying content runs into a slippery problem: despite its prevalence, identifying fake news is highly subjective. The term “fake news” has so many definitions and requires such nuances of human understanding that competent fact checking becomes a sort of AI-complete problem.</p><p>As such, the biggest room for improvement is in detecting different types of coordination, the signature of behavior. Due to the sheer scale of information online, people have yet to try all the possible ways to cluster accounts in a network. Imagine identifying clusters not only by the content but also by comparing stylometry, topicality, timing, and other characteristics. By analyzing the network propagation of stories and trying to use that heuristic to classify things as fake, companies can avoid having to determine what is ground truth.</p><h4>Pushing truth to power</h4><p>In my view, companies that aim to create a more holistic, trustworthy news platform have a higher chance of succeeding than those looking to build the next crowdsourced fact-checking service or bias-detection algorithm.</p><p>Pushing values of truth can take many forms. One example might be building a reputation score that indicates how much the reader can trust a piece of content.</p><p>Having a reputation system that identifies and differentiates trusted sources of information from bots and from misinformation could be an invaluable tool for consumers trying to make informed decisions.</p><p>I liken this to New York City’s requirement that restaurants post calorie counts on their menus. This not only encouraged New Yorkers to eat more healthfully, but — an unintended consequence — also forced restaurants to introduce lower-calorie items. Similarly, by having a reputation score on each piece of content, users would be able to see which sources and content they could believe, implicitly putting pressure on the content producers who care about their reputation.</p><p>Just as the availability of food metrics like calories and nutrients helps drive better decision-making and changes public health at a meta level, metrics that scale up to increase the authority of accurate news will drive more accurate thinking about a multiplicity of information sources.</p><p>Reputation leads to credibility, which leads to trust.</p><p>The internet has democratized publishing, but it has not made editing for the common good — that is, optimal consumer education and media literacy — a value or priority. Whether it’s coming up with new models of distribution or building out security products over time, I too resonate with French President Emmanuel Macron when he says, “We must fight against the virus of fake news.”</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FMkEesCaT2NU%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DMkEesCaT2NU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FMkEesCaT2NU%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/fb1fe01779a2106599d4baea734bc9af/href">https://medium.com/media/fb1fe01779a2106599d4baea734bc9af/href</a></iframe><p>Ultimately, I believe this space will be a blend of media and cybersecurity, in which companies will have rethought the distribution of true news while innovating technologically to stop the spread of fake news. If you think about it, cybersecurity attacks such as viruses and DDoS operate the same way, but instead now leverage the social media networks that have come to fruition over the past eight years. And startups that end up taking a multidisciplinary approach to prevention and look to tackle one or more of the three opportunities discussed above will triumph in the fake news arms race.</p><p>If you’re building a company that is looking to join the battle or have a startup to recommend in any of the above categories, drop me an email at <a href="/service/mailto:jerry.lu@luxcapital.com">jerry.lu@luxcapital.com</a>.</p><p>I’d love to chat.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fdde502%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2Fhackernoon%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/3c851dac986ab6dbb2d1aaa91205a8eb/href">https://medium.com/media/3c851dac986ab6dbb2d1aaa91205a8eb/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=448675592803" width="1" height="1" alt=""><hr><p><a href="https://medium.com/hackernoon/the-fake-news-arms-race-448675592803">The Fake News Arms Race</a> was originally published in <a href="https://medium.com/hackernoon">HackerNoon.com</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Sci Fi to Commercialization: The Rise of Digital Avatars]]></title>
            <link>https://medium.com/hackernoon/from-sci-fi-to-commercialization-the-rise-of-digital-avatars-ea164495c377?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/ea164495c377</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[digital-avatars]]></category>
            <category><![CDATA[avatar]]></category>
            <category><![CDATA[sci-fi-commercialization]]></category>
            <category><![CDATA[rise-of-digital-avatars]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Fri, 20 Jul 2018 14:34:08 GMT</pubDate>
            <atom:updated>2019-04-24T10:12:44.119Z</atom:updated>
            <content:encoded><![CDATA[<h4>Weaving digital characters into our real digital lives</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ntZz3C7T1BMS_DX2Qy2GZw.jpeg" /><figcaption>Mori Building Digital Art Museum in Tokyo by the Japanese studio teamLab (<a href="https://www.teamlab.art/">https://www.teamlab.art/</a>)</figcaption></figure><p>In the late 1990s, the Gorillaz — a virtual music band made up of four animated characters — introduced the concept of <a href="https://hackernoon.com/tagged/digital-avatars">digital avatars</a> to the general public. A decade later, in 2013, artistic director Marc Jacobs designed costumes for Hatsune Miku, a virtual pop singer who collaborated with musicians Lady Gaga and Pharrell Williams. More recently, the emergence of Lil Miquela, a 19-year old Instagram star, social influencer, and recording artist who happens not to be real, has reignited a broader conversation about digital avatars, so much so that <a href="http://time.com/5324130/most-influential-internet/"><em>Time</em></a> magazine featured a digital avatar in its 25 Most Influential People on the Internet line-up. Today this technology has begun to appear in a wider array of business and personal applications, beginning to blur the lines between the virtual and real worlds in a way that was envisioned by <a href="https://hackernoon.com/tagged/sci-fi">sci fi</a> writers of previous decades.</p><p>Much has been written about the origins and landscape of this phenomenon, and I won’t cover that ground in this post. For more on the digital avatar market, this blog <a href="https://medium.com/@mhdempsey/everything-to-know-about-digital-celebrities-and-how-they-could-change-the-world-7b3bf9806073">post</a> by Michael Dempsey stands out as an excellent overview and analysis. Instead, this piece seeks to outline what I see as the first instantiation of companies built around this emerging ecosystem. There are a myriad of potential applications for which digital avatars can be useful, but for investors, some applications are better suited for long-term sustainability, while others seem more akin to flash-in-the-pan fads.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gg5auNXvldO0rl851u_aTg.png" /></figure><p>In assessing this sector, I break the digital avatars ecosystem into three major areas (in chronological order of development): digital influencers, digital celebrities, and the digital self. For each, I will assess the opportunities to build a business and highlight companies working within that space, then discuss my expectations of how the market will evolve in its early stages.</p><h3>Digital Influencers</h3><p>We live in an age of authenticity whereby millennials and Generation Z consumers don’t want to feel marketed to. They regard authenticity as more important than the content itself, caring more about being able to relate to the source of the content than the information being presented. Especially with the rise of misinformation in social media platforms like Facebook, we’re starting to see increased skepticism in young people who worry that their personal information is being collected to sell them things. People want authenticity, but the advent of the digital age and social media has turned people into brands, and in many ways brands are becoming more like people. Just look at Instagram and Twitter respectively.</p><p>What has made Lil Miquela so successful, in my opinion, is that a team of people is crafting a narrative around her and her friends that makes her life feel authentic. The “influencers” we see online are curated carefully to garner attention and monetize likes; therefore they are inherently inauthentic. Lil Miquela, on the other hand, makes no claims to authenticity: by revealing herself as a robot, she is showing all that she has, without the Instagram filters and saturated colors.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8l_QiBb7i6pvD5yaWPx2Ww.png" /><figcaption><a href="https://www.instagram.com/p/BhwuJcmlWh8/?utm_source=ig_embed">Lil Miquela’s big reveal</a>, <a href="https://www.instagram.com/p/BhzSfUYnfC3/?utm_source=ig_embed">Brud’s response</a></figcaption></figure><p>Ironically, it is the fact that she is completely manufactured and managed that enables her to be an influencer. While she isn’t herself real, she’s created to touch on the authentic needs, desires, and interests of her target audience, so she represents an authentic expression of the group’s gestalt.</p><blockquote>“We’ve become bored with watching actors give us phony emotions. While the world he inhabits is in some respects counterfeit, there’s nothing fake about Truman himself. No scripts, no cue cards. It isn’t always Shakespeare, but its genuine. It’s a life.” — Christof (Ed Harris), the director and creator of The Truman Show</blockquote><p>Just like any great character in literature or opera or the stage, the character is made up, but the issues it touches on are not (i.e. Lil Miquela helped amplify and give a voice to Black Lives Matter, immigrants’ rights, and encouraged young people to vote). Indeed, as Dr. Tama Leaver, an associate professor of internet studies at Curtin University, notes, “She is all the influence without all the stuff we are not supposed to see.”</p><h4>Application</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v67bER7ERZkAss97_DZwZA.png" /><figcaption>Truman Show, WestWorld</figcaption></figure><p>The storyline of Lil Miquela unwinds on a path between reality and imagination, much like the Jim Carrey film “The Truman Show” and HBO’s recent hit “Westworld.” A command center-esque digital agency where humans plot out the storyline, introduce the story beats, and craft the characters’ personalities makes the characters feel real. Unlike chatbots today, which feel like thousands of people’s personalities are sandwiched into one, having a single virtual character with a consistent voice gives a digital entity a real feel, infusing it with a more genuine attractiveness in this age of authenticity.</p><p>Ultimately, I’d love to see how jobs might be created within this space. So much of AI has been about displacing jobs, e.g., “service jobs are going to evaporate because you can replace them with bots.” But what if, instead, those same people can become ghostwriters behind influencers or become part of a team that helps build these digital characters’ personalities to make them more human? These displaced workers might now take on the role of creating a Mechanical Turk for digital influencers.</p><h3><strong>Digital Celebrities</strong></h3><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FcQ54GDm1eL0%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DcQ54GDm1eL0&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcQ54GDm1eL0%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/002d8915c14edd7968069d71086104ba/href">https://medium.com/media/002d8915c14edd7968069d71086104ba/href</a></iframe><p>Have you seen the <a href="https://www.youtube.com/watch?v=cQ54GDm1eL0">video</a> of Obama calling Trump a “complete dipshit”? That video was created using a technique called <a href="https://hackernoon.com/exploring-deepfakes-20c9947c22d9">deepfake</a>, an AI-based human image synthesis technique that offers the ability to swap one face for another in an image or a video (please take a look at Gaurav Oberoi’s excellent <a href="https://hackernoon.com/exploring-deepfakes-20c9947c22d9">explainer</a> on the world of deepfakes). As the technology continues to improve, I see digital celebrities taking off as a forcing function of deepfakes. The original deepfake code is published on GitHub, so the implementation is relatively trivial; the greatest bottleneck occurs in the collection and preparation of training data (on the order of 300–2000 images).</p><p>But because there are millions of celebrity photos online, it’s easy to create your very own Brad Pitt deepfake that can replicate his voice and likeness. Celebrity is profitable because of its relative scarcity, but because there are usually plenty of images available for any given celebrity, the phenomenon of a celebrity may be digitally replicable, and thus celebrities risk losing the very uniqueness that built their brand. Celebrities are going to have to license their likeness with respect to certain properties or otherwise lose money. What’s going to happen when people start remaking movies with different actors or new dialogue? What will movie companies like Disney do? Smart companies that are able to tackle this issue will be at the forefront of preventing the destruction of their revenue stream like the ones caused by Napster and Bit-Torrent.</p><h4>Application</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hYwBkzELNsgtmoufTzUH-A.jpeg" /></figure><p>I see the initial path to productization being similar to the early growth of style transfer — the technique of recomposing images in the style of other images. When style transfer came out of academia, <a href="https://prisma-ai.com/">Prisma</a> emerged as the first company to productize it. In a similar fashion, I see the first widespread adoption of deepfake being a fun social app, an equivalent of a Snapchat filter whereby instead of you being Kim Kardashian, Kim K will be in the selfie with you, licensing her likeness so you can, for example, take a pic with her on the beach.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ruv0UtynAeicSQQjaxsfrw.png" /><figcaption>young Anthony Hopkins in Westworld</figcaption></figure><p>As adoption continues to grow, the biggest value is likely to be in the use of digital celebrities in the professional video / film production space. In a recent episode of “<a href="https://www.radiotimes.com/news/2016-10-18/heres-how-westworld-de-aged-anthony-hopkins/">Westworld</a>” there is a scene featuring a young Anthony Hopkins. According to visual effects supervisor Nick Worth, the de-aging was a mix of CGI and live-action filming. “We took a scan of Sir Tony and used that as our base; [we] then used a ton of photography and reference to figure out what we wanted him to look like.”</p><p>Imagine if you could do this much more cheaply and effectively, and if you could take older artists and allow them to portray younger versions of themselves. Or, even better, bring back popular artists of yesteryear, generating revenue for the estates of long-dead actors by bringing them back to life. A celebrity’s time and reach is finite, but a digitized version is not limited by the same physical constraints, thereby opening up new solutions and resources for content creation and monetization.</p><h3><strong>Digital Self</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uq4m1IEPoTawy4x4KADxDg.png" /></figure><p>Rather than just having Anthony Hopkins put out movies forever, what if we could extend relationships with people and family members? By creating a digital representation of someone we know to be real, we are essentially extending life as we know it. This can be accomplished by combining someone’s digital footprint with natural language processing and AI algorithms. Using these methods, we could theoretically extend someone’s life into a virtual eternity. Pullstring was most famous for making <a href="https://www.wired.com/story/a-sons-race-to-give-his-dying-father-artificial-immortality/">dadbot</a>. <a href="https://replika.ai/about/story">Replika</a> got started because its creator’s friend died in a car crash. One can imagine a future in which people use voice or conversational interfaces to create personalized relationships with virtual characters, whether that’s a replica of your dead relative or best friend.</p><h4>Application</h4><p>We’re already starting to see personalized relationships developing between humans and machines. Have you ever wondered why Google’s assistant doesn’t have a name? Unlike Siri or Alexa, you trigger Google Home by saying “OK, Google.” I don’t know the answer, but I’d speculate that by doing it this way, it might be easier for Google the company to personalize “OK, Google.” When everyone calls their digital assistant Alexa or Siri, it suggests that everybody’s Alexa or Siri is the same because it’s personified as the same person. My Google, on the other hand, could be very different from your Google, since it has developed different facets of its “personality” based on your interactions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*WaE19fbC-LFYYUDzvPEDfQ.jpeg" /></figure><p>The first inceptions of the digital self are going to be applications that I believe appeal to the needs at the top of Maslow’s hierarchy. Virtualize therapy and have it sound like a real person sympathizing at the other end of the phone line. A lot of the fear and friction that come with asking for help, or the need to build trust with a stranger, are alleviated via this method. Inhibitions melt away faster with a digital entity. In fact, an ongoing <a href="https://www.wired.com/story/virtual-therapists-help-veterans-open-up-about-ptsd/">study</a> within USC’s Institute for Creative Technologies (ICT) is using virtual therapists to help veterans open up about their PTSD. “People are very open to feeling connected to things that aren’t people,” says Gale Lucas, a psychologist at USC’s ICT and lead author of the study.</p><p>Once those higher needs are met, I can see the digital self gaining wider acceptance as it transitions from outside relationships to personal ones. In an era when talking to Siri and Alexa has become mainstream, interacting with a virtual dead relative may not seem such a radical idea. Hossein Rahnama, a visiting professor at the MIT Media Lab who is leading a project called <a href="https://www.media.mit.edu/projects/augmented-eternity/overview/">Augmented Eternity</a>, predicts that persona-based chatbots are two to five years away: “We are reaching a level of data maturity where we can represent someone’s identity, and reliably predict responses.” The remaining question is “how it gets rolled out in order to go mainstream.”</p><p>Technology and society have advanced immensely since the Gorillaz used a digital avatar to package videos and market music. This space has only now started to gain traction, and the recent success of Brud (the creator of Lil Miquela) may provide a window into why. First, Brud understands its audience and does a good job of connecting with humans by touching on things that are culturally at the top of people’s minds or part of the zeitgeist. Second, they use transmedia storytelling to develop brand “story worlds,” popularized by Disney, taking a holistic view that a story can take place on multiple channels and the digital realm is just one of those channels.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*0PvfCMWC8WXyhEIFehcBuw.png" /></figure><p>I’m excited about this space because I see the digital realm as one more channel where people can interact with their real-world experience. For this reason, I’m bullish on a growth trajectory that starts with the digital representation of what people know as real, beginning with trivial applications like fun social apps or very specific purposes like the virtual therapist described above. The gradual addition of concrete and successful applications will help build people’s comfort with the concept, just as Alexa and Siri have become daily tools for many of us. They will pave the way for later development of apps that identify something real, sustain it for longer, add emotional investment, and thus open up multiple channels for monetization.</p><p>This technology is challenging our understanding of reality, authenticity, and selfhood — and therefore has vast opportunities for commercialization as well as vast opportunities for misunderstanding and puzzlement among adopters. If we can preserve the core elements of an existing relationship, rooted in what’s real, then I believe these technologies will be adopted.</p><p>If you’re building the future of digital representation or have a startup to recommend in any of the above categories, drop me an email at <a href="/service/mailto:jerry.lu@luxcapital.com">jerry.lu@luxcapital.com</a>.</p><p>My digital avatar would love to chat.</p><p>Thank you to <a href="https://www.linkedin.com/in/brianrose/">Brian Rose</a>, <a href="https://www.linkedin.com/in/timfrietas/">Tim Frietas</a> and <a href="https://www.linkedin.com/in/goberoi/">Gaurav Oberoi</a>!</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fupscri.be%2Fdde502%3Fas_embed%3Dtrue&amp;dntp=1&amp;url=https%3A%2F%2Fupscri.be%2Fhackernoon%2F&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=upscri" width="800" height="400" frameborder="0" scrolling="no"><a href="https://medium.com/media/3c851dac986ab6dbb2d1aaa91205a8eb/href">https://medium.com/media/3c851dac986ab6dbb2d1aaa91205a8eb/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ea164495c377" width="1" height="1" alt=""><hr><p><a href="https://medium.com/hackernoon/from-sci-fi-to-commercialization-the-rise-of-digital-avatars-ea164495c377">From Sci Fi to Commercialization: The Rise of Digital Avatars</a> was originally published in <a href="https://medium.com/hackernoon">HackerNoon.com</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Wharton: Connecting Entrepreneurs Coast to Coast]]></title>
            <link>https://medium.com/startup-grind/why-wharton-connecting-entrepreneurs-coast-to-coast-cdeaf824941d?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/cdeaf824941d</guid>
            <category><![CDATA[financial-planning]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[university]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[education]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Mon, 07 May 2018 22:34:25 GMT</pubDate>
            <atom:updated>2020-09-07T18:59:18.087Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aXn5y-D7frYW0z6QpbDJjw.png" /></figure><h4>Earning an MBA at an intense business school like Wharton can be all-encompassing and exhausting, and when I’m in the middle of the grind, I sometimes don’t have a clear answer to that “why Wharton?” question a lot of people ask.</h4><p>As an engineer by training and profession, my decision to attend this business school certainly puzzled some of my friends and colleagues. But I am constantly reminded of the dynamic, talented, and diverse set of people drawn to Wharton, and what a fantastic opportunity it is to study here.</p><p>The first-ever Wharton Entrepreneur Summit in San Francisco just three weeks ago was a case in point. My friend <a href="https://www.linkedin.com/in/shawnxxu/"><strong>Shawn Xu</strong></a> and I organized the event, and it went even better than we expected, so I’d like to share some of my ideas and insights about the experience.</p><p>Although most people associate Wharton with the world of finance, I chose it for its increasingly impressive track record producing successful — and often visionary — <a href="https://www.inc.com/magazine/201805/tom-foster/direct-consumer-brands-middleman-warby-parker.html">entrepreneurs</a>. In fact, <a href="https://mba-inside.wharton.upenn.edu/semester-in-san-francisco/">Wharton’s San Francisco</a> campus is a testament to its commitment to keeping its finger on the pulse of the most promising and exciting trends and developments in business.</p><p>Its bicoastal footprint doubles the reach and scope of Wharton’s network, and also signals that the institution is not bound by staid thinking about what a university is, how it should act, and what it should aspire to become. On a more granular level, this expansion has allowed Wharton to develop additional subject-matter expertise in areas such as tech, healthcare, and data analytics.</p><p>The entrepreneur-oriented conference took place in New York City last year and, seeing the exciting things taking place in The Bay, Shawn and I began contemplating moving the conference to San Francisco.</p><p>We had some concerns about moving it away from New York because last year’s event was so well attended, but in the end, we elected to take this bold step and committed ourselves to rounding up a group of excellent speakers whose presence would entice people on the East Coast to fly out to California.</p><p>Leveraging our connections, we were fortunate enough to secure the Google San Francisco office as our venue. (Let me give an enthusiastic “thank you!” here to both Google CEO <strong>Sundar</strong> <strong>Pichai</strong> and Google CFO <strong>Ruth Porat,</strong> both of whom are Wharton alumni.)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-c4yGJP92F7rYAAo." /><figcaption>Wharton San Francisco <a href="https://mba-inside.wharton.upenn.edu/semester-in-san-francisco/">Campus</a></figcaption></figure><p>Our next major decision was whether to open the event to business schools from across the country. Business school students are often siloed when it comes to interfacing with other like-minded individuals.</p><p>There is so much talent across the country that we decided to use the conference to bring the community of tech-focused MBA students together to share and learn from each other, with the hope that maybe two people will form a friendship or professional partnership and have our conference to thank!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Y6AdB7KzBsG7VePl." /><figcaption>Students talking to <a href="https://www.linkedin.com/in/caseywinters/">Casey Winters</a> after moderating the growth panel</figcaption></figure><p>In order to promote our event, Shawn and I reached out to the entrepreneurship clubs, venture capital clubs, and various other student organizations at more than 20 business schools across the country. We were limited in terms of the number of attendees we could accept due to venue capacity constraints, but we had over 250 applications.</p><p>Having too much demand and figuring out what to do from there is not a bad problem to have! In the end, 150 students from 16 business schools attended. And, a special shout-out to INSEAD, who sent students from Singapore to the event! The demographics of the group included some interesting stats:</p><ul><li>53% of attendees were male, and 47% female</li><li>44% were first-years, 47% were second-years, and 9% other</li><li>27% of attendees had technical backgrounds</li><li>43% were active founders</li></ul><p>In addition to the amazing panels on growth hacking, early-stage AI investing, and a keynote speech from Wharton MBA alum <a href="https://www.linkedin.com/in/jzwillinger/"><strong>Joey Zwillinger</strong></a> (co-founder of Allbirds), a major component of our conference was a pitch competition in which we looked for the top business school startups nationwide.</p><p>With over 60 applications, we narrowed it down to the top 10; these were invited to the conference to pitch to partners and principals from <strong>Lightspeed</strong>, <strong>Floodgate</strong>, <strong>NEA</strong>, <strong>DFJ</strong>, and <strong>Felicis</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*hGAhOgFzMdbOtBoh." /><figcaption>Judges (from left to right): <a href="https://www.linkedin.com/in/arifj/">Arif Janmohamed</a>, <a href="https://www.linkedin.com/in/irischoi/">Iris Choi</a>, <a href="https://www.linkedin.com/in/speechu/">Sundeep Peechu</a>, <a href="https://www.linkedin.com/in/chiragchotalia/">Chirag Chotalia</a>, and <a href="https://www.linkedin.com/in/aaronejacobson/">Aaron Jacobson</a></figcaption></figure><p>These diverse startups ranged from a company tackling the nail salon industry to a company using blockchain technology to alleviate inefficiencies in banking systems.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/486284e9862666339cb6867d4af4cb2e/href">https://medium.com/media/486284e9862666339cb6867d4af4cb2e/href</a></iframe><p>The winner of the competition was Mon Ami, pitched by two female founders from Stanford looking to enter the dementia care market, a $55B market in the US that will only continue to grow as the population ages. By providing an online marketplace where people with dementia and their family caregivers can find and hire “activity companions,” Mon Ami makes it easy for customers — family members desperate to meet the human needs of their loved ones — to find their “mon amis.” This particularly hit home for Shawn, whose father has Parkinson’s Disease and recently suffered a stroke.</p><p>His father now needs around-the-clock care at home, and their family has found it particularly difficult to find the right caregivers. As Shawn puts it, a company like Mon Ami should exist in the world, and he’s hopeful that one day they might grow into a company that could help families like his.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Uw4OdtMF9NFWWqPP." /><figcaption>From left to right: <a href="https://www.linkedin.com/in/kelly-chen1/">Kelly Chen</a> (outgoing e-club president), <a href="https://www.linkedin.com/in/shawnxxu/">Shawn Xu</a> (incoming e-club president), <a href="https://www.linkedin.com/in/runmdc/">Madeline Dangerfield-Cha</a> and <a href="https://www.linkedin.com/in/joyzhang8/">Joy Zhang</a> (co-founders of Mon Ami), and I</figcaption></figure><p>Going into this conference, I had some concerns that the competitors would be merely “business school startups” that lacked technical sophistication. I feared that judges from the top VCs in the world might not be impressed. I couldn’t have been more wrong!</p><p>The pitch competition was very well-received, and the judges expressed delight at the quality of teams and ideas coming out of business schools today. Indeed, one judge in the pitch competition commented that Shawn and I “did a great job curating [the] companies. I was pleasantly surprised at the quality of the teams and ideas!”</p><p>And I found it most interesting that each judge had a different runner-up, an indicator of the diversity in thought among VCs, and the fact that different companies appeal to different investors for various reasons.</p><p>In addition to meeting other interesting and talented entrepreneurs, the conference also helped to make connections that strengthened the Wharton community. For instance, <a href="https://brianrothenberg.com/"><strong>Brian Rothenberg</strong></a>, VP and GM of Eventbrite, was one of the speakers on our panel on growth, and during our subsequent conversation, he expressed an interest in guest lecturing at Wharton.</p><p>He already speaks frequently at Stanford and Berkeley, and he’s keen to become more involved in the Wharton community because he sees it as a fertile ground for dynamic, promising companies. Relationships like this are, in fact, why Wharton sought to establish a robust presence on the West Coast.</p><p>Of course, although we view this experience as a huge success, we still want to make it better. With this in mind, we were eager to hear detailed feedback and specific ideas. Attendees indicated that they loved the Networking Happy Hour (not sure if it was the open bar or the networking itself), but that it was too short.</p><p>Given that one of the main strengths of such an event is the space it allows for relationship-building and networking, they suggested more fully featured opportunities for business students, successful entrepreneurs, and business leaders to interact. Thus, in the future, we will have a longer networking event that includes name tags indicating what schools people are from.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1023/1*bLVP4GmhakKfgWjL0ABPPQ.png" /><figcaption>Thank you to all of our amazing sponsors!</figcaption></figure><p>Finally, I’m still amazed that the entire event was all self-funded. We hustled to secure the partnerships that made it possible for the event to be entirely free for the business students who attended. We can’t thank all of our sponsors enough for their generosity. Had it not been for <strong>Google</strong>, we wouldn’t have had a venue to host our event. Had it not been for <strong>Accenture</strong>, we wouldn’t have had a $20K cash prize for the pitch competition, a clear incentive for startups to come pitch. Had it not been for <strong>WeWork</strong> and <strong>WSGR</strong>, we wouldn’t have had an open-bar networking happy hour and founder investor dinner, respectively, and been able to meet and connect with like-minded individuals across schools.</p><p>Our corporate sponsors showed their dedication to nurturing the next generation of tech companies, and we are massively grateful for their contributions.</p><p>As the VP of the conference and with Shawn as the Co-President of Wharton’s Entrepreneurship Club again next year, I can’t wait to do this again. It not only brings like-minded individuals from across the country together, but helps bring Wharton to the center of technological innovation and entrepreneurship. In the aftermath of that weekend, the answer to “why Wharton?” has never been clearer to me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*dErX2WDsMVgo8u7W." /></figure><p>If you’re a VC, startup, student, or person with a general interest in being a part of next year’s event, please feel free to reach out! We can’t wait to see you soon.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cdeaf824941d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/startup-grind/why-wharton-connecting-entrepreneurs-coast-to-coast-cdeaf824941d">Why Wharton: Connecting Entrepreneurs Coast to Coast</a> was originally published in <a href="https://medium.com/startup-grind">Startup Grind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The VC Playbook: An MBA’s Journey into Venture]]></title>
            <link>https://medium.com/startup-grind/mba-vc-playbook-8c5405290dcd?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c5405290dcd</guid>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[journey]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[strategy]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Fri, 13 Apr 2018 00:17:53 GMT</pubDate>
            <atom:updated>2020-05-12T16:23:39.036Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rK0jwo0-Zx4dectQ2uzZJg.jpeg" /></figure><h4>“At the end of the day, you have to prove that <em>your thinking </em>is stronger, <em>your network </em>is more extensive, <em>your personality</em> is more receptive, and <em>your deal flow </em>is more proprietary than the other 500 candidates that are applying.” –- David Selverian, Bessemer Venture Partners</h4><p>VC fuels most startups, small and large, today, and is of vital interest to any business student or entrepreneur seeking funding. But what about business students who want to be on the VC side of the table/equation? How does a new MBA learn about this space and break into a career path in venture?</p><p>Let me first start by saying that there is no standard path for breaking into venture. This playbook is just one example that students have followed in order to enter the industry. I’ve followed this playbook and the details I will share are a combination of personal anecdotes and conversations I’ve had with successful venture capitalists who have entered the industry post-graduation.</p><h4>This playbook is structured around:</h4><ol><li>Positioning your prior experience</li><li>Differentiating yourself from other candidates</li><li>Building relationships</li><li>Leveraging tools and resources</li></ol><p>Within each section, I will share stories from friends in the industry; in a future post, I will recount my own journey of breaking into VC by following this exact playbook.</p><h3>Position your prior experience.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aeTyonH8VhxCZqvKSjL9Ag.jpeg" /></figure><p>Looking at the prior work experience of my Wharton Class of 2019, I would bucket students into one of three main verticals, which to my surprise is quite evenly split: industry, finance, and consulting (excluding the “Others” bucket).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1CtnMlHGRLsXB-Zn." /><figcaption>Wharton Class of 2019 <a href="https://mba.wharton.upenn.edu/class-profile/">Profile</a></figcaption></figure><p>Each vertical has its own specialties, and a student wishing to work in VC can play up their strengths to match. In venture, having deep strengths is more important than being a jack of all trades, as is commonly thought. You have to have something that differentiates yourself, on which you can develop a personal brand.</p><h4>Industry</h4><p>For people who have worked in industry, it’s important to identify your skill set and your contributions to the company and the industry as a whole. For me, that was being a data scientist at Google. For someone else, it could be deep marketing experience in consumer-facing e-commerce. Identify your key differentiated skill and really play up its relevance to VC. For example, if you have a data science background, you could help with tech diligence on opportunities, or if you have a marketing background you could evaluate marketing-driven companies.</p><h4>Finance</h4><p>If you’ve worked in pure finance, your skills could be valuable to early-stage startups. There are very few bankers who leave banking to join an early-stage startup. A lot of these startups have difficulty with their books, so most of them engage outsourced CFOs since it’s hard to bring in solid finance talent early from outside the organization.</p><p>Your strategy is to try to find an internship and/or part-time school experience where you can provide your services to early-stage startups. If you’re really lucky, one of those startups will start to get hot during that window, and you can ride that reputation when you are job-searching.</p><h4>Consulting</h4><p>Similarly, for consultants, it’s important to find your connection to the startup space. Take time outside of classes to help early-stage startups with things like business development, market assessment, strategic partnerships. These opportunities not only give you exposure to the startup ecosystem but also allow you to hone your skill set while potentially improving the trajectory of a company.</p><h4>What do you do that is different or unique?</h4><p>Ultimately, you have to figure out what your unique background brings to the table and clearly demonstrate how it can be applied to the job. However, just spinning a prior experience isn’t going to suffice. It’s just as important to do all the stuff no one else is willing to do to indicate your deep passion for the field. The last thing you want is to be that person who wants to break into the industry but is half-hearted about it, and isn’t willing to put in the time and effort to demonstrate their commitment. The next section describes how to follow this path.</p><h3>Differentiate yourself from other candidates.</h3><blockquote>“You have to do things that other people aren’t willing to do in order to set yourself apart.” — Roger Chen, Silverton Partners</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5PV1CoHzVH1CfoeiZX3n6w.jpeg" /></figure><p>Although there is an element of luck when it comes to VC, you have to make your own luck. And the only way to do that is to just put in the work. For instance, most people haven’t done what Roger and I did before coming to business school: taking a pre-MBA internship. Most students see the summer before business school as their last months of freedom. We could have been on vacation, too, and no one counseled us to do a pre-MBA internship. This is the mentality you have to carry with you throughout the whole process.</p><h4>Thought Leadership</h4><blockquote>“Come in with examples of being scrappy and resourceful and prove that you have access to proprietary deals and markets. Demonstrate your skills and knowledge through subtle language throughout the interviews and conversations.” –- David Selverian, Bessemer</blockquote><p>If you know for sure that you want to break into venture after business school, you simply have to do things that other people aren’t willing to do. It takes a lot of work, no one’s going to tell you to do it, and you’re not going to get paid for it, but taking the initiative on some of these fronts will set you apart. Here are some examples.</p><p>Many people schedule coffee chats with VCs asking for advice. Asking for advice is good, but your goal is really to engage with the VC at a much deeper level, demonstrating value before you make a request. For example, with my background in data science, I could put together a deck that highlights my perspective on artificial intelligence (AI) and some companies that I find exciting within the space.</p><p>This presentation should be tailored to the individual and it should be directed to a specific partner’s roadmap. In fact, most partners at venture firms have specific investment areas they are deeply looking into that year; through building relationships with investors, you’ll come to learn the focus areas of specific investors.</p><p>However, don’t be superficial: simply doing a Google search is not enough. To make it credible, you need to have talked to people. Let’s take Silverton Partners’ Roger Chen, for example, and say his roadmap is AI. Since he’s based in Austin, he can’t easily talk to founders of AI companies at various stages in San Francisco.</p><p>If he can talk to a student like me, who not only did part-time work for one of the spoken-to startups but also reached out to all the companies when creating my presentation, that gives me a ton of credibility. Now I’m bringing something to the table that Roger can’t do himself.</p><p>Once you have this presentation, don’t be afraid to publish your content. Although it may be scary to think about publishing your ideas for the world to see (or, if you’re like me, you struggle with the process of writing), think of it this way: forcing yourself to create content also forces you to think deeply about an industry, practice concise communication, and develop a brand that people can associate with you and can, ultimately, set you apart.</p><h4>Extracurriculars / Classes</h4><p>Be selective and make your participation meaningful. Rather than simply signing up for every possible club available at your business school, choose extracurriculars that will help you build your network with VCs. Keep in mind that the goal isn’t to pad your resume: it’s to build relationships. In my opinion, the best positions in clubs are not on the executive team but the trek leader, the panel organizer, the person who interfaces with the VCs.</p><p>Ultimately, it’s about doing things that you enjoy doing. David Selverian of Bessemer did not start Scottie Ventures, a student investment group based in Pittsburgh, for the sole purpose of getting into venture. He did it because he’s always been passionate about entrepreneurship.</p><p>For him, the point of starting Scottie Ventures was to make sure that the next generation could get a well-organized experience with VC, so they could see what it’s like to work in the industry and speak knowledgeably about that experience.</p><p>In addition, he says, having to teach a weekly lesson on venture capital to a core group of 15 students while acting as the filter for the work they produced and the fund itself was a good exercise in management. As a byproduct, he was able to build on those relationships and broaden his network, bringing value to Bessemer as a firm.</p><p>Or take Jeff Ellington of Fuel Capital. Jeff was a history and philosophy major during college but took it upon himself during his MBA at Wharton to double major in statistics (not conventional for someone doing “business” school). It was the extra stats class that allowed him to be able to have a real conversation with a company using applied machine learning as a technology for a business application.</p><p>It was the extra stats class that allowed him to understand cohort analysis and benchmark CAC / LTV when evaluating the next Warby Parker. Most people that get into this business have found ways to build up skills in other places.</p><p>The reality of VC jobs is that VCs have way more really qualified people inbound than they can ever hire. So you have to do the extra work to show that you have something special. Faking it won’t make it in the venture world.</p><h3>Approach “Networking”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8j6d6ITTUepma7xGYiJ4bg.jpeg" /></figure><h4>Personally, I don’t like the term networking.</h4><p>Networking seems to imply that you’re building relationships for the sole purpose of advancing your career. It’s important to make this component more personal and be intentional about how you “network.”</p><ol><li>Before reaching out, figure out how to deliver value before it is asked (this ties back to our earlier section on how to differentiate yourself).</li><li>Coffee is better than a phone call, which is better than an email. Do your best to come in at the top of that food chain.</li><li>When sending emails, keep it short and to the point. Show, don’t tell, who you are, what you do, the value you can bring, and why you want to be a part of the firm.</li><li>Don’t ask for too much time (15–20 minutes).</li></ol><p>Finally, a lot of this process is about relationship <em>building</em>. You need to demonstrate that you’re smart and that you’re well networked.</p><p>You will want to show that you know how to get access to professional, efficient conversations with people in your network across the investment spectrum, across different stages, across different markets. Perhaps most importantly — demonstrate that people trust you. When you have a network built out, you can bring that value to any firm.</p><h3>Leverage Tools and Resources</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wdTDUDhzr1A1lGNm1lzA0w.jpeg" /></figure><h4>Daily newsletters</h4><p>Any person trying to get into VC should have certain specific resources delivered to their inbox every morning and should scan through every single one of them to see who’s getting funded, who’s doing the funding, what markets are exciting, and what models are coming to fruition.</p><p>Those should be no brainer automatic reads every single morning and it should be something you should do naturally because you want to. If you’re resisting doing this, give this career path some more thought. You may be forcing yourself into this role because it seems shiny to you from the outside, but if the day-to-day work isn’t appealing and doesn’t come naturally to you, you are probably not going to enjoy VC or do well in it.</p><h4>Here’s a short list of the resources you should be eagerly scanning every day:</h4><p><a href="http://fortune.com/gettermsheet/">Term Sheet</a></p><p><a href="https://www.axios.com/pro-rata-2349858369.html">Pro Rata</a></p><p><a href="http://avc.com/subscribe/">AVC</a></p><p><a href="http://www.feld.com/">Feld Thoughts</a></p><p><a href="https://pitchbook.com/subscribe">Pitchbook</a></p><p><a href="https://mattermark.com/newsletters/">Mattermark</a></p><p><a href="https://www.cbinsights.com/newsletter">CB Insights</a></p><h4>Job Tracking</h4><ol><li>One of the most valuable resources for tracking down jobs is John Gannon’s <a href="https://johngannonblog.com/">blog</a>, which includes a job board that’s updated quite frequently. It also includes dozens of links, which take you to insider resources about how to break into the industry.</li><li>Tracking who is raising funds and what firms are investing in — there are a few Google docs circulating in the venture community that identify firms that have recently raised funds (<a href="https://docs.google.com/spreadsheets/d/1ebGZ6-ivf_3woBGC4Kz_3217DhjGsefoRu_5iP3nuFQ/htmlview#gid=0">New funds raised</a>, <a href="https://docs.google.com/spreadsheets/d/1khvSN2pcnsc7O5dr9mq8TwJv0h_zJJYLzhiZZ_KDmfM/edit">Firms by thesis</a>).</li><li>These kinds of firms are probably looking to hire, because they now have more management fees coming in, especially if their fund size has increased fund-over-fund. Reach out to these firms with a warm intro. In addition, be on the lookout for funds that haven’t raised in a while and offer to work for them for free. Funds typically raise every 5–7 years, so it’s good to stay ahead of the curve.</li></ol><p>You can look for funds that are raising new capital, but keep in mind, it’s impossible to time hiring. What you want is to be top-of-mind for them when they’re ready to hire so you can jump on opportunities as they come available.</p><p>Again, you’ve got to make your own luck, and the only way to do that is to put in the work. If you’re willing to do that, people will recognize it. Yes, it’s hard, there’s no doubt about it, especially while you’re in business school. But I know that you can do it. You just have to make the time and hustle hard.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TSvZeEkT6fL_FTDsk2FfrA.jpeg" /></figure><h4>Two parting thoughts to leave you with:</h4><ol><li>You’re never too good for a job. If you really want to get into venture, it’s important to take on any experience you can get (unpaid, part-time, remote, etc.) and be appreciative of the opportunities that come your way.</li><li>Always be willing to help others. It sends a signal about your integrity, it will make you a better person, and you never know when an act of goodwill might end up benefiting you enormously later. It’s not a quid pro quo: don’t expect others to immediately pay you back in kind. But if I’ve learned anything in business school about the business world, it’s this: what goes around really does come around — when you help others, unexpected (positive) things can happen.</li></ol><p>If you’re in business school and facing another week packed with team meetings and problem sets, this may seem more like a play encyclopedia than a playbook. But if VC is the right path for you, you’ll find a lot of these components come naturally to you and are even enjoyable.</p><p>Start by taking some time to think seriously about how you can position your prior experience and differentiate yourself from other candidates. Once you’re clear on that big picture, leverage the tools and resources you need to be knowledgeable and credible, and get started building relationships.</p><figure><a href="https://about.startupgrind.com/membership/#membership-welcome"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xTGvpsG_lwJvGKaTlGc-yQ.png" /></a></figure><figure><a href="http://partners.startupgrind.com/"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*evD8OQSj8LDRlbGsYTqWyA.png" /></a></figure><figure><a href="https://www.oracle.com/startup/?pcode=&amp;source=:ad:ba:::MediumAd&amp;SC=ADV"><img alt="" src="https://cdn-images-1.medium.com/max/780/1*sdKS9u5ciPkHD0tjQfMeTw.png" /></a></figure><p>Thank you to <a href="https://www.linkedin.com/in/chenro/">Roger Chen</a>, <a href="https://www.linkedin.com/in/jeff-ellington-94779422/">Jeff Ellington</a>, and <a href="https://www.linkedin.com/in/davidselverian/">David Selverian</a> for your insights and perspectives!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c5405290dcd" width="1" height="1" alt=""><hr><p><a href="https://medium.com/startup-grind/mba-vc-playbook-8c5405290dcd">The VC Playbook: An MBA’s Journey into Venture</a> was originally published in <a href="https://medium.com/startup-grind">Startup Grind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ok, Alexa…how do you work?]]></title>
            <link>https://medium.com/data-science/ok-alexa-how-do-you-work-4365ddb588b4?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/4365ddb588b4</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[voice-recognition]]></category>
            <category><![CDATA[speech-recognition]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Tue, 23 Jan 2018 18:07:29 GMT</pubDate>
            <atom:updated>2018-02-09T14:48:15.097Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_mhgcRV2UY0N7CG1vDyvYw.jpeg" /></figure><blockquote>“ I have many components made by lots of people.” — Amazon Echo</blockquote><blockquote>“I can help you with lots of things. For example, you can say things like… To see more examples check out the Google Home app.” — Google Home</blockquote><p>Not quite the answer I was looking for.</p><h4>Ok Jerry, how <strong>do</strong> voice-controlled speakers work?</h4><p>I recently stumbled upon the YouTube channel for venture capitalists a16z (<a href="https://a16z.com/">Andreesen Horowitz</a>), and was immediately drawn in by the content of their videos, particularly their animated shorts <a href="https://www.youtube.com/playlist?list=PLM4u6XbiXf5rcgvD82RkfxIBy1jsn_C_6">series</a>. In my favorites, general partners <a href="https://twitter.com/vijaypande">Vijay Pande</a>, <a href="https://twitter.com/BenedictEvans">Benedict Evans</a>, and <a href="https://twitter.com/cdixon">Chris Dixon</a> take complex concepts of machine learning, S-curves, and the Innovator’s Dilemma, respectively, and break them down into digestible, easy-to-understand 5-minute videos.</p><p>Inspired by their example, I decided to create my own animated short summarizing my passion and research area, voice computing, as clearly and succinctly as I could. It’s an animated TL;DR of my written posts, if you will. Leveraging similar hand-drawn visuals and color animations, I wanted to create a short that captures the essence of</p><ol><li>why voice is the next big platform shift</li><li>how voice-centric devices work</li><li>how we measure the quality of an audio sensor and how we increase directionality</li><li>where there are frontiers for improvement that can drastically improve the directional microphones already included in the Amazon Echo and Google Home</li></ol><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F--wIoNRWuow%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D--wIoNRWuow&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F--wIoNRWuow%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/43c98b3ce510f9a395afad172190b495/href">https://medium.com/media/43c98b3ce510f9a395afad172190b495/href</a></iframe><p>Voice computing is already among us. In fact, Baidu’s chief scientist, Andrew Ng, predicts that by 2020, 50 percent of all searches will be voice- or image- based. Similarly, a report from Juniper Research has found that by that same year, voice-enabled smart speakers will reach 55 percent of US households. As we inch closer to that future, I’m excited to see startups continue to 1) develop services for the Amazon Echo and Google Home ecosystem, 2) find new opportunities to use voice across multiple industries, and 3) build out the voice-enabling infrastructure — the picks and shovels for the growing voice tech rush.</p><p>If you enjoyed the video, please like, comment, and share. And I’d love to hear from you about other areas of voice you would like me to take a deep dive into.</p><p>Also, I would highly recommend checking out and subscribing to a16z’s YouTube channel. In addition to its shorts, it also features great content like video recordings of their annual a16z Summit and <a href="https://twitter.com/withfries2">Frank Chen</a>’s exceptional <a href="https://www.youtube.com/playlist?list=PLM4u6XbiXf5pIWcUyOX3WkAb5Gv5kKwMY">walkthroughs</a> of all things frontier tech.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4365ddb588b4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/ok-alexa-how-do-you-work-4365ddb588b4">Ok, Alexa…how do you work?</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Principles of Artificial Intelligence & Machine Learning]]></title>
            <link>https://medium.com/data-science/principles-of-artificial-intelligence-machine-learning-122be27d33a4?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/122be27d33a4</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Tue, 31 Oct 2017 18:26:37 GMT</pubDate>
            <atom:updated>2017-10-31T19:03:00.707Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*voV0s4n6oewS7SH0x6ugzQ.jpeg" /></figure><p>Artificial intelligence has captivated me since I worked on projects at Google that ranged from detecting fraud on Google Cloud to predicting subscriber retention on YouTube Red. Looking to broaden my professional experience, I then entered the world of venture capital by joining <a href="http://bv.ai/">Baidu Ventures</a> as its first summer investment associate where I got to work with amazingly talented founders building AI-focused startups.</p><p>Now at the Wharton School at the University of Pennsylvania, I am looking for opportunities to meet people with interesting AI-related ideas and learn about the newest innovations within the AI ecosystem. Within the first two months of business school, I connected with <a href="https://www.linkedin.com/in/nicholasthorlind/">Nicholas Lind</a>, a second-year Wharton MBA student who interned at IBM Watson as a data scientist. Immediately recognizing our common passion for AI, we produced a lunch-and-learn about AI and machine learning (ML) for our fellow classmates.</p><p>Using the following deck, we sought to:</p><ul><li>define artificial intelligence and describe its applications in business</li><li>decode buzzwords such as “deep learning” and “cognitive computing”</li><li>highlight analytical techniques and best practices used in AI / ML</li><li>ultimately, educate future AI leaders</li></ul><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.slideshare.net%2Fslideshow%2Fembed_code%2Fkey%2FUZVMHN5E0hthR&amp;url=https%3A%2F%2Fwww.slideshare.net%2FJerryLu15%2Fprinciples-of-artificial-intelligence-machine-learning-81434105&amp;image=https%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Fprinciplesofaimlvflinkedinpreso-171031184059-thumbnail-4.jpg%3Fcb%3D1509475418&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=slideshare" width="600" height="500" frameborder="0" scrolling="no"><a href="https://medium.com/media/f56d7eea1da2888f0151182cb896cab0/href">https://medium.com/media/f56d7eea1da2888f0151182cb896cab0/href</a></iframe><p>The lunch-and-learn was well received. When it became apparent that it was the topic at hand and not so much the free pizzas that attracted the overflowing audience, I was amazed at the level of interest. It was reassuring to hear that classmates were interested in learning more about the technology and its practical applications in solving everyday business challenges. Nick and I are now laying a foundation to make these workshops an ongoing effort so that more people across the various schools of engineering, design, and Penn at large can benefit.</p><p>With its focus on quantitative rigor, Wharton already feels like a perfect fit for me. In the next two years, I look forward to engaging with like-minded people, both in and out of the classroom, sharing my knowledge about AI with my peers, and learning from them in turn. By working together to expand Penn’s reach and reputation with respect to this new frontier, I’m confident that we can all grow into next-generation leaders who help drive companies forward in an era of artificial intelligence.</p><p>I’d love to hear what you think. If you found this post or the deck useful, please recommend them to your friends and colleagues!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=122be27d33a4" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/principles-of-artificial-intelligence-machine-learning-122be27d33a4">Principles of Artificial Intelligence &amp; Machine Learning</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Mic Check 1, 2, 3]]></title>
            <link>https://medium.com/startup-grind/mic-check-1-2-3-1d756a17d5d9?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/1d756a17d5d9</guid>
            <category><![CDATA[voice-recognition]]></category>
            <category><![CDATA[speech-recognition]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Tue, 19 Sep 2017 12:27:56 GMT</pubDate>
            <atom:updated>2018-07-23T03:15:37.490Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hN6EpM8nLEh14EDxHNt-Mw.jpeg" /></figure><h4>Sound is the most natural way for human beings to communicate, and in the near future microphone technology will enable us to use it to control the world around us.</h4><p>As some of the world’s largest tech companies race to take advantage of ecosystems based on voice recognition, the billion-dollar mic industry is going to have to keep pace with advancements in technology.</p><p>In this article, I examine the current $1 billion global market for microelectromechanical systems (MEMS) microphones and consider potential frontiers for improvement, as well as some of the startups and research ideas disrupting this industry.</p><h3>The microphone market today</h3><p>The use of microphones is increasing rapidly. YoY growth is about 18% and IHS predicts that the use of MEMS microphones will rise to about 6 billion units in 2019.</p><p>The biggest reason for this fast growth curve is that smartphone, tablet, and even smart speaker makers are putting multiple microphones into their devices. For instance, the Amazon Echo has as many as seven MEMS mics; even in some of the latest smartphones, you’ll find up to four microphones.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/874/0*nPz9NdMyaayJVMNE." /></figure><p>Traditionally, the market has focused on using microphones in mobile applications. In contrast, we envision growth in areas such as IoT, automotive, industrial, and smart homes and buildings, driven largely by voice input and contextual awareness.</p><h3>Existing microphone technology</h3><p>A MEMS microphone is typically composed of a fixed back plate and a moveable diaphragm. When sound comes along, the moveable diaphragm is distorted by sound pressure, changing the gap (and capacitance) between its surface and the backplate. The change in capacitance is then measured and converted to an electrical signal by a digital ASIC. This electrical signal represents sound.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/513/0*6hoVjqIxP4M0zIis." /></figure><p>Since the gap between the diaphragm and the backplate is essential for the capacitive MEMS microphone to function, it can also cause many failure modes. The graphic below shows some of those failure modes: 1) normal, 2) dust / particle damage, 3) water entry, and 4) stiction failure, which may be caused by a sudden acceleration or loud sound (acoustic overload) that throws the diaphragm toward the back plate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/954/0*Y0GxKkUWtqUJC4DY." /></figure><p>As you can see, the capacitive MEMS microphones that have dominated the market for the past 10–15 years are easily damaged by common environmental contaminants such as water, dust, and particulate matter.</p><p>The poor reliability of capacitive MEMS microphones is a direct result of their architecture. The industry is keenly aware of these problems, and product designers are actively seeking a better option.</p><h3>Piezoelectric microphones</h3><p>Building on research conducted at the University of Michigan, engineers from <a href="http://vespermems.com/">Vesper Technologies</a> have built a microphone without a gap, called the VM1000 — the first commercially available piezoelectric MEMS microphone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/489/0*LzGKGDVHbxJ3E5Wq." /></figure><p>The VM1000 consists of a single layer of flexible plates for both the backplate and the diaphragm. Changes in sound pressure cause these plates to bend and experience stress. As the plates are built from a sandwich of piezoelectric materials, the stress generates an electrical charge, which allows for direct measurement of sound. This creates a microphone that does not need a backplate.</p><p>In a nutshell, piezoelectricity is the appearance of an electrical charge across the sides of certain solid materials when you subject them to mechanical stress. This material is not new: smartphones contain dozens of piezoelectric radio-frequency (RF) filters.</p><p>The RF filter industry is worth billions of dollars. So there is no shortage of reliable and replicable manufacturing tools to build piezoelectric MEMS. Vesper is capitalizing on this infrastructure to mass-produce the first piezoelectric MEMS microphone on the market.</p><p>Piezoelectricity also enables Vesper’s systems to draw incredibly low power and wake upon sound. Without requiring the push of a button to activate Siri with the AirPods or Alexa with the Echo Tap, Vesper’s technology could make truly always-listening interfaces possible even on the smallest of battery-powered devices.</p><h3>Optical microphones</h3><p>Historically, microphones have always been based on mechanically moving parts, whether it’s a capacitive membrane or a piezo. Because these devices are mechanical, they are prone to interference, mechanical disturbance, and the inert mass common in conventional microphones.</p><p>Israeli startup <a href="http://vocalzoom.com/">VocalZoom</a> has released a voice biometrics solution based on its optical sensor that measures voices using facial vibrations, eliminating the need for microphones and the noise reduction software of traditional acoustic solutions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ftrEdK96q5_8FcvV." /></figure><p>The laser is directed at the face of the person talking, and measures vibrations “in the order of tens and hundreds of nanometers — so small that nothing else can pick them up.” These micro-measurements of the skin are converted into audio.</p><p>Because of their precision, aim, and ability to sidestep sound waves, no other surrounding noise interferes with the clarity of the voice. In addition, by detecting the direction of arrival, the sensor can verify that the person of focus is in the right direction and at the right distance so the sensor only listens to a particular user.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*q1ujuCnf0nCukWDx." /></figure><p>The company is working with most voice-recognition software systems and headset manufacturers, and is also working on a car mirror integration approach and with MEMS manufacturers who are interested in combining VocalZoom’s technology with classic acoustic audio.</p><h3>Graphene microphones</h3><p>Most commercially available microphones today use nickel to make microphone membranes. However, following the development of graphene as a promising material for various electronics devices, it has also been studied for MEMS microphones.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*q5B0dMc97W1ZmSWV." /></figure><p>In November 2015, researchers at the University of Belgrade in Serbia built the world’s first graphene-based microphone, leveraging graphene’s ability to detect faint and high-frequency sound waves.</p><p>The team grew a multi-layer graphene membrane on a nickel foil substrate using a chemical vapor deposition process, then etched the nickel foil away. The resulting membrane can act as the vibrating membrane that converts sound to an electrical current in a microphone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/0*i52sInF-KmAgyEla." /><figcaption>Multilayer graphene condenser microphone <a href="https://www.researchgate.net/publication/277722494_Multilayer_graphene_condenser_microphone">paper</a></figcaption></figure><p>The team’s prototype graphene mic boasts 10 decibels higher sensitivity than commercial microphones, at frequencies of up to 11 kHz. And model simulations indicate that even greater sensitivity is theoretically possible.</p><p>With 300 layers rather than the prototype’s 60, a graphene vibrating membrane may be able to detect frequencies of up to 1MHz — approximately fifty times higher than the upper limit of human hearing.</p><p>Recently, several companies have expressed interest in using graphene membranes for speakers or microphones. In May 2017, Apple was a granted a new <a href="https://www.google.com/patents/US20170006382">patent</a> (filed in 2015) that details an audio device that uses a diaphragm made from a graphene-enhanced composite material. Apple’s graphene membrane can be used in a speaker, microphone, or headphone device.</p><p>As devices become smaller and lighter, it will be more and more challenging to provide high-quality audio using conventional materials and researchers are looking at innovative ways to improve the mechanical response of these audio devices — graphene being one.</p><h3>Frontiers for improvement.</h3><p>While development areas such as piezoelectric and optical are improving the performance of existing microphones today, I’m interested in learning more about “new” microphones — companies or research that are radically influencing the way audio capture is done.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4SPLx02H6to_15u8." /><figcaption>Abe Davis: New video technology that reveals an object’s hidden properties (TED <a href="https://www.youtube.com/watch?v=npNYP2vzaPo">talk</a>)</figcaption></figure><p>For instance, Abe Davis’s <a href="http://www.abedavis.com/thesis.pdf">research</a> from MIT uses a regular camera and a potato chip bag, the ultimate in a low-tech/dirty environment/open back chamber.</p><p>Requiring no additional sensors or detection modules, Davis’s cost-effective method of speech recognition is capable of producing accurate and efficient representations of surround-sound acoustics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*NbLXuLqgjxHUY8bQ." /><figcaption>Stack spectrograms of accelerometer, microphone, and EMI sensors. (<a href="http://www.gierad.com/projects/supersensor/">Carnegie Mellon University</a>)</figcaption></figure><p>Gierad Laput’s <a href="http://www.gierad.com/assets/supersensor/supersensor.pdf">research</a> from CMU uses a single small sensor board to detect dozens of phenomena of interest in a room, including sounds, vibration, light, heat, electromagnetic noise, and temperature.</p><p>One small device can function as an all-purpose super sensor, which can be plugged in and deployed for any sensing application. You can then program next-level applications, for example, turning on a warning light when the faucet is being turned on or calculating wear and tear on a forklift.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1014/0*wHcDaVxm9WN4Ikp0." /><figcaption>Remotely detecting a faucet being turned on. (<a href="http://www.gierad.com/projects/supersensor/">Carnegie Mellon University</a>)</figcaption></figure><p>The ability to communicate over a wide range of modalities could drive many new applications, potentially leading to drastic improvements over the directional microphones included in the Amazon Echo or Google Home.</p><p>I’ll be watching advances in adjacent industries (like piezos, lasers, and cameras) and if you’re working on innovative ideas within this space, I would love to hear from you!</p><figure><a href="https://www.startupgrind.com/conference/#/"><img alt="" src="https://cdn-images-1.medium.com/max/720/1*H-MKIYdxN3heLbg3HYYBAw.png" /></a></figure><figure><a href="https://www.startupgrind.com/start/"><img alt="" src="https://cdn-images-1.medium.com/max/720/1*hu7Gme84g9vnGitjqGIsxA.png" /></a></figure><figure><a href="https://www.startupgrind.com/startup/"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6rGIjWP3uRpMd9mZ74dlMA.png" /></a></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1d756a17d5d9" width="1" height="1" alt=""><hr><p><a href="https://medium.com/startup-grind/mic-check-1-2-3-1d756a17d5d9">Mic Check 1, 2, 3</a> was originally published in <a href="https://medium.com/startup-grind">Startup Grind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can you hear me now? Far-field voice]]></title>
            <link>https://medium.com/data-science/can-you-hear-me-now-far-field-voice-475298ae1fd3?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/475298ae1fd3</guid>
            <category><![CDATA[voice-recognition]]></category>
            <category><![CDATA[towards-data-science]]></category>
            <category><![CDATA[speech-recognition]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Tue, 01 Aug 2017 17:15:18 GMT</pubDate>
            <atom:updated>2017-08-05T22:53:01.437Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9eak7qvds76J5U6yozS-HQ.jpeg" /></figure><h4>In my previous <a href="https://medium.com/startup-grind/the-network-effect-of-voice-ba47f08ad2de">post</a>, I made the case that successful AI companies will develop a moat within the “computational speech” value chain by either creating a network effect around data or developing proprietary algorithms. In this post, I will examine the first step in that value chain, the initial pickup of voice signals, and identify opportunities where startups can succeed.</h4><h4>Zooming in on voice</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*-fsj14f4eQo2EZEI." /></figure><p>Voice control is an intuitive method of interacting with hardware and associated services. It’s far more natural than connecting up a keyboard and mouse, tapping an on-screen keyboard, or clicking a remote control.</p><p>As we see more voice-activated devices, it is important to understand how a device is capable of listening to and understanding voice commands. There are many components in this process, but the two we will examine are the microphone array (the hardware) and the deep learning architecture (the software) that make such a complex system work.</p><h4>Hardware | Microphone arrays</h4><p>Speech recognition systems often use multiple microphones to reduce the impact of reverberation and noise. With each generation of the iPhone, there has been an increase in the number of microphones, from one in the first iPhone to four in the iPhone 5 and 6S.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/0*U9hk9Bd2G1AbZ_So." /></figure><p>Even smart speaker devices like the Amazon Echo are using up to seven microphones. In fact, the Echo mics are arranged in a hexagonal layout, with one microphone at each vertex and one in the center. The delay between each microphone receiving the signal enables the device to identify the source of the voice and cancel out noise coming from other directions. This is a phenomenon known as beamforming.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*H46QgE24Q0WsU8ZJ." /></figure><h4>Software | Deep learning</h4><p>Deep learning has played a fundamental role in voice pickup. The ability to recognize spoken language was established a few years ago, but learning-based techniques like Deep Neural Networks (DNNs) have allowed language processing to equal or surpass human performance in many test cases.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ySoFtH83q4tmZAqr." /><figcaption>AI Progress Measurement | Electronic Frontier Foundation: <a href="https://www.eff.org/ai/metrics">https://www.eff.org/ai/metrics</a></figcaption></figure><p>Only a strong combination of hardware and voice recognition algorithms can lead to product success. With bad microphones, recognition accuracy degrades no matter how intelligent of a deep learning model is employed. On the flip side, having excellent microphones with sub-optimal machine learning does not produce the necessary accuracy.</p><h4>The problem of far-field voice pickup</h4><p>While state-of-the-art speech recognition systems perform reasonably well in close-talking microphone conditions, performance degrades in conditions where the microphone is far from the user.</p><p>Imagine a common scenario in which one person is indoors, speaking to an Amazon Echo.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/0*fOqLKtmkfKiH2mM6." /></figure><p>The audio captured by the Echo will be influenced by 1) the speaker’s voice against the wall of the room, 2) the background noise from outside, 3) the acoustic echo coming from the device’s loudspeaker, and 4) the output audio against the wall of the room.</p><p>These factors all contribute to low signal-to-noise ratio (SNR), room reverberation, and unknown direction of speech and noise — all important challenges that need to be addressed. As the user moves further away from the product’s microphones, the speech level decreases while the background noise level remains the same. Beyond noise and reverberation, other challenges include a lack of large-scale far-field data and unexplored efficient deep learning architectures. The bottom line is that there is still a huge gap between speech recognition and human performance in these far-field scenarios.</p><h4>How does far-field voice pickup work?</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*iI0ZFqdWYmc_hu4g." /></figure><p>Speech recognition systems often use separate modules to perform speech recognition.</p><ol><li>The audio input is sent to an acoustic sensor,</li><li>which converts acoustic signals to electronic, then to digital signals.</li><li>Then, it goes to a digital signal processing chip, where speech enhancement is applied with fixed embedded algorithms. These embedded algorithms perform traditional signal processing techniques: source localization (locating the direction of sound) and beamforming (suppressing background noise).</li><li>The resulting enhanced signal goes to a conventional acoustic model for speech recognition.</li></ol><p><em>Microphone technology aims to replace fixed embedded algorithms with a deep-learning-based trainable algorithm.</em></p><p>The drawback of fixed algorithms and components is that they cannot adapt to the rest of the trainable machine learning systems built on top of them. When you put a trainable deep learning system on a set of fixed chips, that deep learning model has to learn what the embedded algorithm is doing, undo it, and perform its own computation on top of it. This complicates things for far-field speech recognition, because the audio input is already distorted and along each step of 1) converting the signal from acoustic to electric to digital and then 2) pre-processing, you lose information along the way.</p><p>Google has been at the forefront of this <a href="https://drive.google.com/file/d/0ByfBg9vVsBJ-OHpycWV6VllxVGM/view">research</a>, demonstrating the use of raw waveforms as the output of acoustic sensors, and thereby avoiding any of the pre-processing done by the built-in chips (i.e., localization, beamforming) of today’s systems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/448/0*3ItCBDAtxD44nlUe." /><figcaption>Google’s neural network adaptive beamforming model architecture</figcaption></figure><p>In essence, Google is looking to combine steps 3 and 4 in the process. The idea is to give the microphone array more degrees of freedom to optimize the algorithms based on the data. With the acoustic sensor, you just hope that it detects correctly and doesn’t add too much noise or distortion, and that the rest of the system can retrieve its information using trainable deep learning architectures.</p><h4>Opportunities for startups</h4><p>Typically, fixed algorithms are developed for general cases based on heuristics, while deep-learning-based algorithms are trained with the actual data itself for that specific task. Microphones that consist of fixed chips with embedded algorithms are probably not what you want when you have a deep-learning-based system.</p><p>Given that scope, an ideal hardware setup for any speech recognition application should do minimal digital signal processing on those fixed embedded chips. It should take in raw waveforms and build complicated deep learning algorithms on top that are both trainable and flexible.</p><p>The question then becomes whether or not you can collect a lot of training data to train a deep learning algorithm for that specific hardware, or use other deep learning techniques to compensate for the differences in acoustic sensors. The problem today is that deep learning speech recognition models are trained on hundreds of thousands of hours of data that were collected on different hardware.</p><p>In order for startups to compete against the big tech giants who own the full design capability, they need to look for opportunities to disrupt the value chain through data, software, or automation.</p><p><em>The hardware space has huge potential, as sophisticated deep learning techniques must be deployed on highly-customized inexpensive hardware.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kjUw8TZ8x2hFjXPq." /><figcaption>the battle of voice-enabled hardware, @ziwang ’s toys</figcaption></figure><p>All existing major data sets will have been utilized, and access to these models is widespread. Companies that build the infrastructure to gather, annotate, and train new models for data sets that don’t exist yet will succeed.</p><p>Take a company like <a href="http://vespermems.com/">Vesper</a>. They are changing the way microphones are designed by exploiting the physical property of piezoelectric material (I’ll explain the underlying technology in a future post). As a result, their microphone technology doesn’t suffer from dust and environmental degradation like traditional mics do, so their microphones are much higher quality and therefore deliver higher fidelity results, which serve as the signal transmitted for further processing.</p><p><em>The key to production-level recognition is the extensive use of data, advanced techniques in data augmentation, and model architecture.</em></p><p>The weakness of fixed algorithms and components is that they cannot adapt to the trainable machine learning systems built on top of them. But, if your end-to-end model is like those discussed above, the deep learning model can actually learn which information features should be extracted by relating them to the overall goal, which is the decoded characters or words in speech recognition.</p><p><em>Automation will comprise of personalized devices and continuous learning.</em></p><p>For example, with data collected from a particular user, the whole pipeline can be optimized to understand that user better.</p><p>But personalization still poses many problems. After a good initial system has been deployed, the deep learning algorithm needs continuous training based on usage, as behavior changes.</p><p>Both excellent hardware and intelligent software is needed, whether improving performance for a user (e.g., using their language, speech, etc.) or providing user-specific solutions (e.g., an assistant with personalized voice, personality, etc.). Personalized hardware and deep learning systems can bring a competitive advantage in both these areas.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*eFKUzrJU4Qzynfdy." /></figure><p>Baidu is at the forefront of this research with the recent announcement of their <a href="http://research.baidu.com/deep-voice-2-multi-speaker-neural-text-speech/">Deep Voice 2</a> system. Constructed entirely from deep neural networks, the system can learn the nuances of a person’s voice with just half an hour of audio and can learn to imitate hundreds of different speakers. This system gives machines new diversity of speech and will be huge in bringing personalization and familiarity to speech recognition.</p><p>One interesting personalization opportunity for startups is to be the first to adapt and market their speech recognition systems to other languages. Although a company like Baidu may adapt Deep Voice 2 to a language like Chinese in the near future, such companies are ignoring the problem today.</p><p>Another interesting domain and verticalized niche for personalization might be accessibility. For example, <a href="https://www.ava.me/">Ava</a> is using voice recognition software to translate conversations into text for people with hearing impairments. Similarly, an interesting startup in Turkey called <a href="https://arikovani.com/projeler/akilli-baston/detay">WeWalk</a> is building a smart cane for the visually impaired and leveraging Nuance communications for their speech technology.</p><p>Personalization and continuous learning will expedite the voice first revolution, allowing machines to replicate human-like interaction, smoothing awkward conversations, and enabling customization.</p><p><a href="http://www.fluent.ai/">Fluent.AI</a>, a company based in Montreal, is tackling improvements in both software and personalization. This company’s solution is trying to bridge the gap directly from audio input to NLP without the speech-to-text step in between, extracting intent directly from spoken commands and context. In doing so, their intent recognition system becomes truly personalized, learning from user-dependent accents, varying contexts, and acoustic behaviors to recognize frequent phrases across different environmental effects.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/992/0*3WauplNB0FpaDy3h." /></figure><p>By further collaborating with OEMs and suppliers, the Fluent.AI solution is able to have more control over the data that is collected. Through this partnership, Fluent.AI can offer OEMs a faster, more cost-efficient road to market as well as broader reach than conventional technology.</p><p>With all that said, there is plenty of space for a smaller, nimble player to get traction if they focus on one or more of those three key areas — data, software, and automation. Ultimately, startups who are able to own the entire voice stack — both the data collection via the hardware and the trainable deep learning algorithm in the software — will be at a competitive advantage, providing better performance in speech recognition and language understanding.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=475298ae1fd3" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/can-you-hear-me-now-far-field-voice-475298ae1fd3">Can you hear me now? Far-field voice</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Network Effect of Voice]]></title>
            <link>https://medium.com/startup-grind/the-network-effect-of-voice-ba47f08ad2de?source=rss-242a8d3b145------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba47f08ad2de</guid>
            <category><![CDATA[speech-recognition]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[voice-recognition]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <dc:creator><![CDATA[Jerry Lu]]></dc:creator>
            <pubDate>Sun, 09 Jul 2017 19:08:43 GMT</pubDate>
            <atom:updated>2017-07-15T17:07:58.289Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gD4GwO1zV33IIgoeYLVrzA.jpeg" /></figure><h4>In today’s voice ecosystem, the large platform players of Amazon, Google, Apple, and Baidu are controlling the entire voice-stack, all the way from the hardware to the software. How, if ever, are startups going to be able to compete?</h4><p>Having researched this space and mapped the voice market <a href="https://medium.com/startup-grind/the-voice-enabled-revolution-f35f751c2f77">landscape</a>, I’d like to share my thoughts about where the greatest value creation opportunities lie for voice-positioned startups.</p><blockquote><strong>In short, I believe that companies who focus on disrupting one or more of the three key areas of data, software, and automation within the “computational speech” value chain will thrive in this competitive ecosystem.</strong></blockquote><p><strong>What is the “computational speech” value chain?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vL02d2Gqczbdent_." /></figure><p>If you’ve seen my market <a href="https://medium.com/@thejerrylu/the-voice-enabled-revolution-f35f751c2f77">map</a>, you’ll see that this value chain closely maps to the “platforms / core technologies stack” there. This horizontal layer enables voice computing across a vast array of problem types.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*U0y35qV8je5WvO51." /><figcaption>my voice <a href="https://medium.com/@thejerrylu/the-voice-enabled-revolution-f35f751c2f77">landscape</a></figcaption></figure><p>The idea is to find players who are disrupting this value chain by focusing on one or more of those three key areas — data, software, and automation. The most influential combination of the three creates a positive feedback loop.</p><p>The collection of data leads to improvements in software, leading to the automation of the learning process. Thus, systems will continue to learn and improve from experience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Gskz8CIKV-R0s7wK." /></figure><p>By identifying disruptive feedback loops within the speech value chain, we find voice applications with potential for big businesses. With this in mind, let’s look at why data, software, and automation are so key to AI startups positioned for voice.</p><h3>Data | Access to unique or proprietary datasets</h3><blockquote><strong>Time and time again, we’ve heard that “data is the new oil.”</strong></blockquote><p>As the AI talent pool grows, and new tools enable people to engage with the technology, access to unique and proprietary datasets become key to building differentiated products.</p><p>According to InsideBigData, the size of the digital universe will double every two years, a 50-fold growth from 2010 to 2020.</p><p>Human and sensor (i.e., machine-generated) data combined is experiencing an overall 10x faster growth rate than traditional business data, with sensor data increasing even more rapidly at 50x the business data growth rate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pg432kntjJiciw_U." /><figcaption><a href="https://insidebigdata.com/2017/02/16/the-exponential-growth-of-data/">InsideBigData</a>, The Exponential Growth of Data, February 16, 2017</figcaption></figure><p>It’s worth noting that the benefit of data may become saturated, as the learning capacities of the practically deployable models are limited. For this reason, application-related data can be more valuable than generic data.</p><p>If you consider the big problem of converting a random audio to a random text, it’s very hard for a startup to compete, because it requires a lot of data and a lot of compute power and engineering resources.</p><p>To compete against the big tech companies, startups must figure a limited domain to be their beachhead. They must find solutions to problems that do not require a lot of data and computing resources, in industries that have been so far overlooked by the tech giants.</p><p>Take <a href="https://www.digitalgenius.com/">Digital Genius</a>. This company is automating conversations between big brands and their consumers using artificial intelligence (AI) and natural language processing (NLP).</p><p>Working with numerous Fortune 500 customers as clients, Digital Genius has exclusive access to millions of historical chat records. These records help train their deep learning algorithms, learning the most appropriate responses and reducing the training period before having to interact with new, live customers.</p><p>Having access to a unique dataset and leveraging AI and deep learning techniques enables Digital Genius to enter a niche where the likes of Amazon and Google are not yet investing; working smarter not harder.</p><h3>Software | Differentiated approaches to machine learning</h3><p><strong>To make sense of the data, you need to have the right software tools</strong>.</p><p>Advances in deep learning have allowed people to find ways to work on unlabeled data and use unsupervised learning to figure out a topology to get a basic structure (i.e., learning features automatically).</p><p>You can then use traditional back propagation, where you propagate errors back through your weightings to classify and predict what you have labeled.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ajlkuc1LI42h2MI9oUUcJg.png" /><figcaption>(Lee, Largman, Pham &amp; Ng, NIPS 2009) (Lee, Grosse, Ranganath &amp; Ng, ICML 2009)</figcaption></figure><p>Startups that find new ways to undermine the competitive value of large datasets through innovative machine learning concepts will provide value in this market.</p><p>For instance, a company in Paris called Snips.AI is using a <a href="https://medium.com/snips-ai/hey-snips-announcing-the-first-private-by-design-voice-platform-bf23b8a843fd">method</a> called training set synthesis (TSS) to generate artificial training data to train machine learning models, instead of relying on real user data.</p><p>Similar to <a href="https://en.wikipedia.org/wiki/Generative_adversarial_networks">GANs</a>, which generates synthetic datasets when you don’t have enough primary data, Snips.AI is showing that these artificial training datasets can be used to train other discriminative models.</p><p>This approach has potential in applications where audio data is scarce (e.g., in specific domains like healthcare).</p><p>Where user flexibility is desired (e.g., by enabling specific keywords).</p><p>Where enhancing the dataset would improve performance.</p><blockquote><strong>The problem is challenging, though, as the generated datasets must be high quality and diverse.</strong></blockquote><h3>Automation | Continuous incremental improvement</h3><p><strong>The success of an AI solution depends on continuous learning and tuning of the AI model.</strong></p><p>Various AI related technologies such as natural language processing (NLP) and speech recognition have substantially progressed over the years to coalesce into systems that learn-unlearn-relearn.</p><p>Especially for voice interfaces, user personalization is key. Each person has a unique voice with individual characteristics.</p><p>Identifying a user by their voice enables a company to provide a personalized experience by knowing the users frequently used services, the music they prefer, their past purchases, and more.</p><p>Personalization requires technologies that learn these individual behaviors. As a result, startups working on true personalization will need to continuously learn from past and present behavioral data.</p><p>Not only does behavior differ among individuals, but individual behavior changes over time. Systems that rely only on large sets of historical data cannot effectively characterize these changes.</p><p>The success of an AI solution largely depends on the continuous learning that follows every voice interaction.</p><p>I can imagine a system that combines machine learning with human operators, ensuring the availability and accuracy of information is fed into the AI-based solution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LY9t-sMEr7i3za34." /></figure><p>For example, Seattle-based <a href="https://www.definedcrowd.com/en-us/">definedCrowd</a> is automating and accelerating enterprise data training and modeling with this blend of human and machine, combining crowdsourcing and machine learning for speech and NLP technology.</p><p>With people in 53 countries across 42 languages, they are able to help data scientists and developers build natural language applications faster and with higher quality, without requiring deep knowledge of speech science or natural language understanding.</p><blockquote>A true AI system needs to be able to learn continuously. It needs to combine high quality training data with state-of-the-art machine learning algorithms and automation in order to deliver on the enormous potential of personalization, faster and at scale.</blockquote><h3>Holding an Advantage Over the Tech Giants</h3><p>Nimble startups working on detecting and producing vast amounts of data, using AI techniques to represent data as knowledge, or automating knowledge discovery and data mining along the speech value chain, will hold an advantage over the tech giants.</p><p>In theory, finding this positive feedback loop in any AI niche presents an opportunity for a startup to thrive. Machine learning enabled companies connect product and data at the hip.</p><p>The main driver of quality in a machine learning enabled product is the accuracy of the underlying algorithms that form any given task. And that largely comes back to the quantity, quality, and uniqueness of the data on which those algorithms have been trained.</p><p><strong>This is what people describe as data network effects.</strong></p><p>However, because I’ve been so interested in the voice enabled ecosystem, I’m applying this value proposition to “computational speech” first.</p><p>In my next post, I’ll primarily pick apart the first step of the value chain (i.e., voice pickup) and highlight some areas that I’m excited about.</p><p>In particular, I’m watching startups disrupting this domain, powered by proprietary data, machine learning models, and continuous learning.</p><p>If you’re building a company focused on one of these three key areas within the industry, I’d love to chat.</p><figure><a href="http://eepurl.com/bBbrFX"><img alt="" src="https://cdn-images-1.medium.com/max/479/1*11YPy_7jhKXEHVv2tW9xBw.jpeg" /></a></figure><figure><a href="http://startupgrind.com/conference"><img alt="" src="https://cdn-images-1.medium.com/max/480/1*tpiTwqg_k1Az7Qw3QXJYCg.jpeg" /></a></figure><figure><a href="http://startupgrind.com/chapters"><img alt="" src="https://cdn-images-1.medium.com/max/480/1*ZArCdGaGEURWL6erukHAcQ.jpeg" /></a></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba47f08ad2de" width="1" height="1" alt=""><hr><p><a href="https://medium.com/startup-grind/the-network-effect-of-voice-ba47f08ad2de">The Network Effect of Voice</a> was originally published in <a href="https://medium.com/startup-grind">Startup Grind</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>