<?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 kevin novak on Medium]]></title>
        <description><![CDATA[Stories by kevin novak on Medium]]></description>
        <link>https://medium.com/@novakkm?source=rss-5cd09200323e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KhewV8zL5dJVBrkOHeMboQ.png</url>
            <title>Stories by kevin novak on Medium</title>
            <link>https://medium.com/@novakkm?source=rss-5cd09200323e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 04:51:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@novakkm/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[Is Now the Time to Start an AI Business — Why I’m Feeling Optimistic]]></title>
            <link>https://medium.com/@novakkm/is-now-the-time-to-start-an-ai-business-why-im-feeling-optimistic-eb5afbd84fc0?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/eb5afbd84fc0</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[founding]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[starting-a-business]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Mon, 24 Oct 2022 15:01:43 GMT</pubDate>
            <atom:updated>2022-10-24T15:01:43.104Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Is Now the Time to Start an AI Business — Why I’m Feeling Optimistic</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z1aS8Jp2fcashrWb1arNPA.jpeg" /></figure><p>While on my travels to visit Rackhouse founders, Rackhouse LPs, and (hopefully) prospective LPs last quarter, I finally had the chance to make some headway on my reading list…okay, my Audible list…and was able to finish <a href="https://www.amazon.com/Richer-Wiser-Happier-Greatest-Investors/dp/1501164856"><em>Richer, Wiser, Happier</em></a> by William Green.</p><p>In his book, Green interviews a series of great public markets investors to learn what makes them tick and how they developed their unique edges — all while trying to distill any nuggets of wisdom he can pry loose for managers and investors who are eager to continue to hone their investing craft. So, basically me.</p><p>In one chapter, Green spoke with <a href="https://en.wikipedia.org/wiki/Howard_Marks_(investor)">Howard Marks</a>, of Oaktree Capital fame. While walking through London between meetings, I heard Marks share the following quote:</p><blockquote><em>Skepticism calls for pessimism when optimism is excessive. But it also calls for optimism when pessimism is excessive.</em></blockquote><p>After hearing this, I literally had to stop and sit down on a park bench to think. First think about what I had been seeing, then what our founders were sharing, and finally, what the future might look like. It crystallized something I had been grappling with instinctively over these past few months.</p><p>I believe that sometime soon, an investing window is going to open that will have the potential to generate outsized returns. Precisely when it’s going to happen and how long the opportunity will persist are famously hard to predict, but I’ve taken a posture of watchfulness, readiness, and most importantly, optimism.</p><p>So, now you might be asking what makes me so optimistic about the future? Or more specifically, why do I have such optimism for AI startups on the cuff of an economic downturn?</p><p>Well, for one thing, we’ve been through this before.</p><h4><strong>A Brief Reflection on 2001</strong></h4><p>Back when the bubble popped in 2001, my dad lived in a house on Skyline Boulevard in the Santa Cruz Mountains. That summer, I remember driving home most nights and seeing a myriad of Lexuses and Mercedes sedans parked to the side on some quiet stretch of road.</p><p>It wasn’t until my dad told me that those cars were owned by the tech workers, made recently homeless by the crash, that I realized how serious and extreme the cycle of boom and bust could be in Silicon Valley.</p><p>Every night, startup employees, who until just recently had been enjoying penthouse suites and steak dinners, would drive up into the mountains to sleep in their cars, before showering the next morning at the YMCA and going back to work to try to turn their now-worthless stock into something worthwhile.</p><p>Things eventually stabilized, and the cars slowly stopped appearing, but the lesson on how quickly luck can turn and catch the unwary stuck with me. And, I’ve carried that same thinking and wariness to how I approach AI investing.</p><h3><strong>Staying Optimistic During Uncertain Times</strong></h3><p>From the lessons we’ve learned from 2001, I’m also seeing multiple trends in the market that give me optimism about future investment opportunities in AI.</p><p>Let’s review my top four.</p><h4><strong>Pre-Seed/Seed hiring is becoming easier</strong></h4><p>Compared to prior years, it’s becoming significantly easier for early-stage companies to attract and hire talent, including engineering and data talent.</p><p>Layoffs are obviously part of it. In fact, <a href="https://layoffs.fyi/">Layoffs.fyi</a> reports over 80k employees were laid off from startups this year. But an even more high-quality channel for pre-seed and seed companies with experienced talent are Series B/C startups that are now holding equity packages — which are significantly less valuable than when they were first hired (sometimes even worthless!).</p><p>These are generally employees with great value to early-stage companies, as well as relevant experience. They’re also fundamentally excited about “running it back” and finding something new.</p><h4><strong>AI does well in recessionary environments</strong></h4><p>While I’ll let others debate whether we’re in a recession or not, I can say with certainty that AI/ML, and its adjacent technologies, tend to provide business value propositions that resonate strongly with customers in recessionary environments.</p><p>Whether it’s maximizing ROI on any form of spend, mitigating complex risk factors in a portfolio of assets, or key task/process automation, AI founders will likely see increased enthusiasm and customer uptake rates.</p><p>These tailwinds can be massive for an early AI company just getting started.</p><h4><strong>Lots of VC dry powder</strong></h4><p>According to recent NVCA surveys, there is an estimated $290B in dry powder in the VC industry at the moment. For non-industry LPs, “dry powder” is slang for capital that has been allocated to a fund to invest but has not been invested yet.</p><p>Whereas fundraising for new funds is a slower and more deliberate process than in years past, a significant number of firms preemptively raised new funds, and fundraisers are still completing successfully.</p><p>All of this is enough of a reason for optimism. Though investors may be more skeptical and choosy than before, they do have the capital they can (and expect to) deploy.</p><p>Deals will get done once again, eventually.</p><h4><strong>Deal terms/pricing (probably) moving in favor of VCs</strong></h4><p>People can (and have) written entire dissertations on the effects of macro-economic trends on venture capital terms, but generally speaking, we should expect the pricing of companies to move in favor of investors during uncertain times.</p><p>Some important mitigating factors should be considered here (e.g. pre-seed pricing is quite insulated from macro trends generally, Zoom-ification of venture is shifting prices in lower-price markets as investors bring Bay Area biases elsewhere, etc.).</p><h4><strong>“Founders gonna found”</strong></h4><p>More simply, but most importantly, I’m optimistic for this next period of investing because the smartest people I know (and the smartest people they know) are still dreaming about being founders.</p><p>I’ve seen no fundamental shift in baseline enthusiasm for the entrepreneurial journey and for how these individuals want to build new companies. I’m seeing a lot of founders who are currently in heads-down “build mode”, planning to come to market with ideas this fall, and I expect it to be a busy autumn on the pitching front.</p><p>Now, bringing this full circle to my original assertion about the upcoming window of opportunity, while I believe these trends are real, important, and material to investing prospects, I also think they’re going to be hard to measure objectively and time — with any amount of precision.</p><p>On the same note, I further believe that they will be durable; the window will remain open for long enough that we won’t only get a few shots at capturing value. To that end, it’s probably better to be a little late to aggressive deal-making than too early.</p><h3><strong>A Final Thought</strong></h3><p>I can’t say definitively if the economy or market will turn unfavorably against tech companies or early stage investors, and if it does, I won’t be able to tell you when or for how long.</p><p>Just remember this: that same market crashed in 2001 and bounced back, creating massive value for another wave of entrepreneurs and investors, and some of the best returning VC funds ever were started in that same year.</p><p>We should hope to be so lucky to invest in such uncertain times.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eb5afbd84fc0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Hire a Data Scientist: Lessons from Uber Data’s Early Days]]></title>
            <link>https://medium.com/@novakkm/how-to-hire-a-data-scientist-lessons-from-uber-datas-early-days-66236e3959c1?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/66236e3959c1</guid>
            <category><![CDATA[data-scientist]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[hiring-data-scientists]]></category>
            <category><![CDATA[uber]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Thu, 04 Aug 2022 18:32:14 GMT</pubDate>
            <atom:updated>2022-08-04T18:32:14.311Z</atom:updated>
            <content:encoded><![CDATA[<p>It’s obvious to anyone who’s looked closely at Uber specifically, or ridesharing generally, that data (and data science) are core to the company’s ongoing mission of providing reliable ground transportation to everyone around the world.</p><p>What’s less obvious was that data science was comparatively late to develop in Uber’s journey — from finding an initial product-market fit to eventually scaling up to a global decacorn enterprise.</p><p>I was fortunate enough to play a key role in the founding of Uber Data, and I wanted to share the story of how Uber launched its data science function, with a few lessons I learned along the way. My hope is that this provides some insight and guidance to the next generation of applied data companies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U5AZSs0DjG8aGnHyE5ecqw.jpeg" /></figure><h3><strong>Early Days: The “Team Science” Era</strong></h3><p>In 2011, the job title of “data scientist” was years away from existing on an Uber organizational chart. Indeed, the entire data science trend was still in its infancy. Yet, despite all of that, it was clear that Ryan, Garret, and Travis were making data a priority in how they operated and in how they hired.</p><p>From the <a href="https://medium.com/@gc/the-beginning-of-uber-7fb17e544851">very first pitch deck</a> highlighting their goal to do live demand forecasting to <a href="https://techcrunch.com/2011/06/15/uber-out-maths-google-on-nyc-etas/">press releases</a> highlighting how much better they were at forecasting ETAs than Google Maps, it was clear that for Uber, the transportation problem was a data problem.</p><p>Organizationally, the plan to “do data” was less clear. It was clear, however, that the no-dispatcher-in-the-loop product vision required that all data models be deeply integrated into the codebase. Uber’s aggressive and complex operational scaling strategy would also need to benefit from current operations data at any employee’s fingertips, and new ideas would regularly spawn analysis projects.</p><p>Uber’s first attempt at addressing this was to hire a Director of Business Analytics. This individual’s responsibilities included</p><ul><li>Fielding data tasks from the young company and being the central receiving point for data demands.</li><li>For analytics tasks, attempting to answer the company’s questions in our (very incomplete) SQL database.</li><li>For product requests, writing MVP code or product specifications and handing it off to Engineering for implementation</li><li>Submitting any requests for data infrastructure and tooling to Engineering</li></ul><p>Can you spot the issue? They were being asked to solve problems that they were neither capable nor empowered with being able to solve. After 6 months of little progress, and even less support from the already stretched-too-thin Engineering team, the director quit. This led to key insight #1 for launching a data team:</p><blockquote><strong><em>Insight #1: </em></strong><em>Whenever possible, hire people who can do it all, from analysis all the way down to the infrastructure.</em></blockquote><p>Attempt #2 involved Uber hiring a recent neuroscience PhD who had planned to pursue a career in academia, but was in a period of uncertainty regarding his career path.</p><p>Output-wise, this skillset was much more productive — the PhD was able to perform the following tasks:</p><ul><li>Write enough SQL to get the necessary input information from Uber’s primitive data infrastructure</li><li>Have the quantitative ability to build useful models in Python/Jupyter that could be copy/pasted into our codebase.</li><li>Use past experience in writing research papers to write white papers and public-facing content materials, if needed</li></ul><p>Unfortunately, however, this person ultimately decided to return to academia, stepping back from Uber to become a professor.</p><blockquote><strong><em>Insight #2: </em></strong><em>Data Science is a calling, not a phase — hire people who are 100% committed to their career path.</em></blockquote><h3><strong>The 4 Principles: What to Look for in a Data Scientist</strong></h3><p>Factoring in these learnings, Uber eventually found a recipe for a data team that seemed to work. It consisted of four simple principles:</p><ol><li>Everyone must know SQL, regardless of job title.</li><li>We followed the Principle of Most Privilege — every user is given access to as much information as safely possible.</li><li>Data hires must be able to code well enough to commit code to the codebase and be the best in the company at math and/or statistics concepts (later generalized to include operations research and machine learning.) When in doubt, prioritize an engineering skill set that learned machine learning over a theorist who took a coding bootcamp.</li><li>The single most important skills to possess are grit, adaptability, and a bias to completion. “Ugly, Fast, and Done” trumps “Pretty, Slow, and In Progress”.</li></ol><p>Each of these principles served a key function in making data successful. Principles #1 and #2 were critical in enabling self-service analysis. This involved knowing how to get information (via a SQL query) and being organizationally empowered to access a broad basket of information, which enabled anyone at the company to answer their own mission-critical questions.</p><p>It also allowed the team members to run their own lightweight exploratory analysis and delegated a large portion of high-urgency/low-complexity work back to users (that would have otherwise been fielded by data hires).</p><p>Principle #3 grew out of Uber’s learnings from its first analytics hire. Having a key blocking dependency on another function (in this case, a data analyst waiting on an engineer) makes it very hard to scale or execute on problems quickly. And the single most common dependency for most of Uber’s projects was either one of the following:</p><ol><li>Committing a finished project to a production codebase downstream</li><li>Making a key change to infrastructure upstream to secure the necessary information to do the project.</li></ol><p>This requirement was non-negotiable and taken quite seriously. In fact, all data hires were interviewed by at least one engineer, and, without question, failing the coding interview was by far the most common reason someone didn’t pass an Uber Data interview.</p><p>Principle #4 worked hand in hand with Principle #3. If #3 required a data hire to be technically competent enough to write adequately good code, Principle #4 meant you also had to be bold enough to make changes to other people’s work. You were responsible for doing this as necessary and in enough of a hurry that you didn’t spend too much time agonizing over whether to do it at all.</p><p>It was also pragmatic to the pace of Uber’s growth in those days, as it was common for Uber to launch several cities a month (often with little notice) and/or require a major engineering push to support a new growth idea or marketing campaign.</p><p>It was also one of the hardest skills to interview for.</p><h3><strong>How Uber interviewed for my own grit</strong></h3><p>For context, I interviewed at Uber after deciding to leave my role in nuclear physics at Michigan State University. As part of my interview process, I was asked to write an API that parsed a simple dataset of historic observations, built a predictive model, and served up that prediction through a series of API endpoints (if none of that made sense to you, the point being: it was a meaty, multi-part problem).</p><p>Given that C++ was my most proficient language at the time, I spent several weeks prepping this assignment in this language for my final interview, which took place on a Friday at Uber HQ in San Francisco (I lived in Michigan at the time).</p><p>After my interview, I planned to spend the weekend visiting with family in the Bay Area, before heading home on a flight Monday. After a full day of interviewing, including a discussion of my take-home assignment that seemed to go well, Travis and Curtis approached me and said, “We really enjoyed the conversation today, but we noticed you wrote your code in C++. Uber does its code in Python. Why don’t you try rewriting this assignment in Python over the weekend, and we can meet again on Monday?”</p><p>Needless to say, I had never written Python before, but a long weekend of Stack Overflow, and some trial and error (not to mention, a lot of patience from my family!), I had an ugly but functional Python copy of my code by Monday morning.</p><p>While I’m not 100% sure my interview experience generalizes what your process should involve, and I rarely advise founders to put their interview candidates through weekend all-nighters, I can honestly say that I cannot think of a better way I could have proved my grit to the company than that.</p><p>And once I was hired and placed into a position to bring in others to join the team, I was able to implement these same four principles into Uber’s recruiting practices and successfully hire its first 100 data scientists.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=66236e3959c1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Articulate the Value of Your AI Company To Investors]]></title>
            <link>https://medium.com/@novakkm/how-to-articulate-the-value-of-your-ai-company-to-investors-4d8c32c2546a?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/4d8c32c2546a</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[fundraising]]></category>
            <category><![CDATA[founder-advice]]></category>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Tue, 19 Jul 2022 18:05:15 GMT</pubDate>
            <atom:updated>2022-07-19T18:05:15.693Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*SMCn_RCfsKVnt7Sd" /></figure><p>Recently, I met with one of my founders, and during our discussion, the topic of value came up. Or more specifically, how to articulate value to AI investors.</p><p>This got me thinking about how we all have our own definitions of what it means to have “value”, and what it takes for an AI (artificial intelligence) business to be considered “valuable.”</p><p>For starters, it’s prudent to acknowledge that aside from you, your team and investors probably have different scorecards for how they assess the value of some part of your company.</p><p>Let’s consider a few examples:</p><p>A strong mission-oriented team will, of course, place value on the pieces of the company that help them achieve their mission. But your engineers may place a higher value on the parts of the codebase they worked hardest on — or think has the coolest architectural patterns.</p><p>Your research team will likely value the work they did that allows them to write a white paper while your operational team will value the playbook they wrote at the tail-end of back-to-back all-nighters more than other playbooks they’ve written previously. After all, it’s only natural for humans to correlate this vague idea of “value” with the time, energy, people, money, etc. needed to make the product work.</p><p>Candidly speaking, investors usually don’t care about those types of “value” metrics, or at least they don’t put as much weight on them as you do. At the most basic economic level of their thought process, your AI company is thought of in more financial terms.</p><p>This often includes:</p><ol><li>How much revenue does it generate?</li><li>What are its profit margins?</li><li>At what rate is its customer base growing?</li><li>At what rate are customers churning?</li></ol><p>From those four metrics (revenue, margin, growth rate, and churn rate) and a little market research, investors can forecast the outcome of the company. The rest of it (e.g., unique tech, cool brand, hours invested) has some value, but it’s much smaller when compared to those basic facts and not sufficient enough to generate investor positivity.</p><h3><strong>Why do I share this reductive context?</strong></h3><p>There are two points I want to hit on, which I believe can help AI companies understand their investors’ thought process, so they can successfully articulate their product and/or company’s value:</p><p>1) <em>Since your team and investors have different scorecards, they won’t be motivated by the same pitch. </em>A team-focused brand story may be very uninteresting to an investor, or vice-versa. If you feel like you’re telling a compelling story, but the audience isn’t excited, make sure you’re customizing your pitch to each audience’s interests — not your own.</p><p>2) <em>When in doubt about how to articulate the value of something to an investor, tie it back to revenue, margin, growth rate, or churn rate.</em> I’ve found that you cannot go wrong with this strategy.</p><p>For example, the founder I spoke to had a long list of features the company needed to build in order to attract the right customers and convince them to buy the product.</p><ul><li>Since they were the only company offering all the features their target audience asked for (i.e., checked all the boxes), the customers are less likely to <strong>churn</strong>.</li><li>At the same time, their technology is so modularized that they are able to complete complex sales without massive engineering rewrites, protecting their <strong>margins.</strong></li><li>Finally, their technology allowed companies to easily integrate the software into their own infrastructure, decreasing the likelihood of <strong>churn</strong> while generating more <strong>revenue</strong>.</li></ul><p>On both a personal and professional level, I’ve found that you can’t go wrong with articulating your company’s value through these four key metrics.</p><h3><strong>A Final Thought</strong></h3><p>When I was working as an entrepreneur at startups, I held the all-to-common view of something like, “investors don’t get it” or “all investors are &lt;bleep&gt;ing morons.” <em>(sound familiar?)</em></p><p>As I got further in my entrepreneurial journey and developed more awareness of the craft of investing, I learned that most investors are not, in fact, morons. (This was disappointing in a sense — if the problem was simply intellect, the fix is easy and obvious!) Most of these apparent disconnects in understanding between founders and investors stem from the differences in incentives that are structural to the investing process.</p><p>This, in my opinion, gives reason for some <em>optimism</em> — If you recognize that the problem is structural and not one of competency, the solution becomes simply one of mapping your worldview to that of the investors. While there’s much that can be improved about the venture capital experience, as long as there are competent people working to improve on both sides of the negotiating table, I’m going to remain hopeful.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4d8c32c2546a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Will AI Bridge the Gap between Web2 and Web3?]]></title>
            <link>https://medium.com/@novakkm/will-ai-bridge-the-gap-between-web2-and-web3-99892a6e4628?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/99892a6e4628</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[web3]]></category>
            <category><![CDATA[venture-capital]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Fri, 10 Jun 2022 18:18:47 GMT</pubDate>
            <atom:updated>2022-06-10T18:18:47.060Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cr3H5yLSrXdr4jvfFNAhKA.jpeg" /></figure><p>If you’ve followed my series on Web3 and AI investing, you’ve probably noticed that I haven’t talked about NFTs (non-fungible tokens). I don’t want to get too caught up in the details of how they work. Instead, I want to bring them into a broader discussion on how AI can help scale consistency in blockchains, using the large amount of disparate information that’s available.</p><p>First, let me backup a moment to provide a little more clarity on why this topic matters. If you do any type of research on NFTs and Web3, you’ll quickly find a lot of enthusiasm from people who want to use a blockchain — and exploit the immutable nature of it — to verify the provenance of goods.</p><p>Again, NFTs are the most obvious example of this now. But imagine if you could make the title of your car or home live on a blockchain ledger. Having digital access to this type of physical object would allow you to transfer ownership on your own vs. going down to your country treasurer’s office to do so.</p><p>Most people would probably prefer this type of scenario. But there’s a problem with trading on a decentralized network like Web3. In order for blockchains to work successfully, everyone who is involved in the transfer of a particular good needs to be on Web3.</p><p>Consider this example. The company <a href="https://superareshop.com/">Superrare</a> specializes in limited edition streetwear. To help fight counterfeit products, they tie their merchandise to a unique address that lives on a blockchain. If they sell one of their products to someone on the chain, the transaction is trackable. However, if that same buyer sells the same product to someone else who isn’t on the chain — and doesn’t log the transfer of ownership — the physical and crypto realities of this object are no longer in sync.</p><p>The result is you end up with a lot of elaborate Proof of Stake (PoS) schemes, where people are trying to come up with these contrived ways to make sure the physical world and crypto world remain in sync.</p><p>This strikes me as an opportunity to develop an automated AI-decisioning layer to help authenticate this type of disparate information. I’m also intrigued by whether we can use an approach around data science or Machine Learning (ML) to build a type of Web3-to-Web2 authenticator link.</p><p><strong>Here’s an example of how this might work</strong></p><p>A lot of people like to pitch me a Web3 version of Uber, particularly doing rideshares as smart contracts. If you think about it, a person who is in one location, and needs to get to another destination, can send out a contract to drivers to be transported at a fixed rate. And once a driver agrees, they’ve initiated the contract and will receive payment after the work is completed.</p><p>However, this situation gets tricky, as it’s hard to decipher whether or not Uber drivers would need to be contractors or employees of the company. And since many want to qualify for health insurance — which is traditionally a benefit given to salaried employees who are responsible for completing a set amount of hours — it becomes even more difficult to track this information and decide who receives health benefits this month and who doesn’t.</p><p>The question then becomes: how do employees receive medical coverage when their work is rewarded in tokens and PoS? Or more broadly speaking, how do we map these necessary physical concepts, like healthcare benefits and authentication, that have physical ramifications to Web3’s “no gatekeeper” type of ecosystem?</p><p>Again, I believe this is a problem that a very intelligent decisioning layer can solve by deciphering the information available, based on a given set of criteria that’s created by the company. These just so happen to be the hallmarks of where AI/ML tend to be very strong.</p><p><strong>Final Thoughts</strong></p><p>There seems to be a ML-shaped hole that makes the transition from Web2 to Web3 more challenging and less seamless. This is going to require us to figure out what needs to go inside that space to ensure the two worlds (physical objects and crypto) sync consistently and seamlessly.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=99892a6e4628" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Can AI Intersect with Web3?]]></title>
            <link>https://medium.com/@novakkm/how-can-ai-intersect-with-web3-ffad7b861c70?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/ffad7b861c70</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[web3]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Mon, 06 Jun 2022 18:08:15 GMT</pubDate>
            <atom:updated>2022-06-06T18:08:15.146Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lHdvq7HhIB3WTZIPB2yQIg.jpeg" /></figure><p>Welcome to part II of AI investing in a decentralized digital ecosystem. In part I, I shared my reactions to the question, “does <a href="https://medium.com/@novakkm/does-web3-spell-the-end-of-ai-investing-b6ab324a8ff8">Web3 spell the end of AI investing?</a>”</p><p>To recap, I see AI playing a bigger role in Web3 for three reasons:</p><ol><li>The development of public blockchains will generate a preponderance of perfect information that will be both accessible and consumable.</li><li>Decentralization will open up the playing field for greater AI/ML innovation.</li><li>Data analytic developers will be uniquely positioned to find ways to accurately analyze and relay the information that’s available on Web3.</li></ol><p>Another dimension to Web3 that I find fascinating, and believe will intersect with AI, is smart contracts. You’re probably familiar with how these work, but to provide some background, smart contracts are self-executing contracts written as lines of code that exist across a decentralized blockchain.</p><p>There’s no gatekeeper or third parties to carry out transactions or enforce the terms and conditions of the agreement. Instead, the code itself in the contract controls the execution of the contract and triggers the next action of the contract, once conditions are met. All activities and transactions are recorded and made public.</p><p>As previously stated, a decentralized network allows us access to perfect information. With smart contracts, this would imply that we’d have perfect information about what’s required to fulfill each contract, as well, since it would be published in a decentralized way.</p><p>Broadly speaking, because smart contracts are typically unrestrained and not opinionated regarding who can actually fulfill its terms and requirements, to me, this presents an interesting opportunity for automation via AI.</p><p><strong>Leveraging Smart Contracts with AI</strong></p><p>To give you a real-world example, I know the founders of a company who are working to reinvent how contract software engineering is being performed. Essentially, their focus is turning common practices used for writing and testing software (i.e., unit testing and test-driven development) into smart contracts.</p><p>So, rather than the software development team sitting around in meetings and creating scope documents, the smart contract would function as the series of unit tests that they could send out to have someone else write the code. For example, the company would post a set of unit tests as a contract and if a software developer who is in the network could submit software that can make them all pass, he or she will receive an agreed upon payment.</p><p>What’s really fascinating about using this idea to me is that, if you think about it, there is nothing about the workflow that requires an actual human to write the code.</p><p>This framework presents an opportunity where Web3 and smart contracts might be the perfect business model for companies to gracefully leverage AI technologies, especially if they’re new to utilizing this type of infrastructure.</p><p>Again, with an abundance of perfect information available, AI has the capability to complete smart contracts and perform the required software unit testing with minimal human involvement.</p><p>Though there are many other examples of how smart contracts can intersect with AI, to me, this particular example provides a clearer picture of how we can leverage AI technologies to improve current software engineering practices.</p><p>Next Up: The Challenges of Web3</p><p>I’ve spent some time talking about how I see Web3 providing opportunities for AI investing. In the last part of this 3-part series, I’ll hit on a few obstacles it may have to overcome first.</p><p>Stay tuned!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ffad7b861c70" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Does Web3 Spell the End of AI Investing?]]></title>
            <link>https://medium.com/@novakkm/does-web3-spell-the-end-of-ai-investing-b6ab324a8ff8?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6ab324a8ff8</guid>
            <category><![CDATA[web3]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Wed, 01 Jun 2022 16:14:17 GMT</pubDate>
            <atom:updated>2022-06-01T16:14:17.340Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3MtFp_HCvIdMcA_jejScmw.jpeg" /></figure><p>When people talk about Web3, I can’t help but notice an overarching theme of “extreme decentralization.” Since AI (artificial intelligence) has been so front-and-center in an era of centralized data and models, it’s not too surprising why people would skip ahead a few chapters and automatically assume that AI won’t have a prominent place in this new open-source digital ecosystem.</p><p>You’re probably already familiar with Web3, but in case you need a little refresher course, let me offer a quick synopsis.</p><p><strong>What Is Web3?</strong></p><p>Broadly speaking, Web3 is a theory about the next generation of the internet, and it will be built as a decentralized network, using technologies like blockchain and smart contracts. Now, there are many aspects of this digital infrastructure that we could cover, but for the sake of this discussion, let’s focus on three key takeaways:</p><ol><li><strong>Web3 will be decentralized.</strong> Data and content will be built, operated, and owned by its users, not corporations, who will be running their own peer-to-peer networks.</li><li><strong>Web3 will be permissionless.</strong> There won’t be any control systems or “gatekeepers” in place to censor, block, or remove users from participating in the network.</li><li><strong>Web3 will be trustless.</strong> Any interactions or transactions will take place between two parties without requiring that they trust each other (or even know each other!). Third party platforms will no longer be necessary to establish mutual trust, as they are in web2.</li></ol><p><em>For a broader explanation on Web3, check out this article from the </em><a href="https://hbr.org/2022/05/what-is-web3"><em>Harvard Business Review</em></a><em>. I also recommend reading this WIRED interview with </em><a href="https://www.wired.com/story/web3-gavin-wood-interview/"><em>Gavin Wood</em></a><em>, the man who coined the term.</em></p><p><strong>The Centralization of AI</strong></p><p>Now let’s get back to the concept of centralization — particularly how it relates to AI. Most people have their own reasons for why they believe centralization is a problem — one being that it leaves little room for collaboration, and therefore, data becomes siloed.</p><p>But I’d argue that a bigger issue with centralized AI is the amount of money and effort it requires.</p><p>Consider that the innovative and emerging use cases for AI come from models that are trained in insanely massive datasets and comprise billions and billions of features. How you acquire, process, and distill down that much information — using the current thinking around data infrastructure and privacy — is if all of that data is centralized, collocated, and owned by the same entity/company.</p><p>Under this current infrastructure, the only way to utilize AI at its highest level dramatically shrinks the aperture of who can contribute to its usage and development.</p><p>This is where I see the main issue of centralized AI. Centralization (and its dollar cost) is choking off innovation. And as a result, AI is far from where it should be.</p><p>Here’s where I think most investors get stumped when they talk about Web3. As AI has largely thrived in centralization, their concern becomes how it will perform in a decentralized network.</p><p>This brings us back to the main question on the floor.</p><p><strong>Is Web3 the End of AI Investing?</strong></p><p>My response is…not likely. AI doesn’t have to be centralized in order to be effective. In fact, Web3 offers a unique opportunity for datasets.</p><p>One in particular is that the development of public blockchains will generate a preponderance of information that’s available to those who have access to that chain. The information will be public, verified by consensus, and complete going back to the start of the chain.</p><p>It goes without saying, but the more perfect the information, the better AI can perform. Therefore, I see AI playing a big role in Web3, simply for the fact that, if all goes to plan, we’ll have perfect information available in a public security domain, as well as in a format that is easier for computers to consume.</p><p>I also believe that some of the same data problems that exist in Web2 will continue in Web3, and therefore, there are new opportunities to bring the best data product learnings forward to the decentralized web — with minimal evolution.</p><p>Think about how decentralization can help companies that are currently leveraging data and transactional tracking. Heights Labs, for instance, monitors cash flows and transactions to help businesses identify money laundering and other financial crimes across the globe. Chainanalytics offers forecasting tools to aid companies in supply chain planning. Similarly, brands like Dune Analytics leverage on-chain crypto data, making it accessible and consumable to users via their digital dashboards. All three companies offer baseline business analytics. The information they need in order to provide these services is available, but it’s also hard to get to.</p><p>Investing in data products that could ultimately analyze and relay the information that’s available in Web3 may present a unique advantage for these companies and improve their capabilities. In turn, the need for this new technology provides an opportunity for data investors to intersect with Web3, as well.</p><p><strong>Understanding How AI Will Intersect with Web3</strong></p><p>I’ve given you a small taste of how AI and Web3 could intersect. In part 2 of this 3-part series, we’ll discuss a few other ways AI can thrive in a decentralized network.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b6ab324a8ff8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Purpose of Platforms in Data Science]]></title>
            <link>https://medium.com/@novakkm/the-purpose-of-platforms-in-data-science-965e2124edf8?source=rss-5cd09200323e------2</link>
            <guid isPermaLink="false">https://medium.com/p/965e2124edf8</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[uber]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[kevin novak]]></dc:creator>
            <pubDate>Tue, 12 Apr 2016 18:03:42 GMT</pubDate>
            <atom:updated>2016-04-12T18:15:21.152Z</atom:updated>
            <content:encoded><![CDATA[<p>Much has been written and discussed about the field of data science and its role in industry; what it is specifically, how to become a practitioner, and the relative attractiveness of the field. However, comparatively little has been written about scaling an organization of data scientists and the organizational support necessary to do data science well, at scale.</p><p>I’ve spent the past year at Uber working on exactly this question, and almost 5 years working on data science at Uber generally, and I wanted to share some of my learnings on the power of platforms when it comes to data science.</p><h3>Uber Data’s Early Days</h3><p>First, some history. Data has always been fundamental to Uber’s story as a business — two of our first ten technical employees were data scientists — and to our product, where we’re building a system that provides effective transportation around the globe that’s so reliable people forget about the complexity of the underlying optimization problem.</p><p>Uber’s always been proud of hiring data scientists who’re comfortable straddling both sides of the divide between an engineer and a researcher. In fact, many data scientists regularly write and ship their own features. This, coupled with a highly entrepreneurial culture, gave us the ability to create an embedded organization that scaled quickly; Uber Data Science grew from a team of 2 to a team of 35 in less than 9 months. We were able to create platoons of scientists going deep as individuals on problems ranging from <a href="https://newsroom.uber.com/us-new-jersey/uber-improves-public-transportation-in-new-jersey/">public policy</a> to dynamic pricing to <a href="https://newsroom.uber.com/us-california/its-a-beautiful-pool-day-in-the-neighborhood/">uberPOOL</a>, yet who still benefitted from having a distinct organizational identity and leveraging the collective wisdom of the whole organization.</p><blockquote>“Scaling a startup quickly as a leader is essentially an exercise in identifying and iterating on the aspects of your product and company that must evolve to survive.”</blockquote><p>Scaling a startup quickly as a leader is an exercise in identifying and iterating on the aspects of your product and company that must evolve to survive while simultaneously holding the line on other aspects that cannot change. Essentially, a constrained optimization problem. Growing a data science org quickly meant growing from recruiting scrappy entrepreneurial data scientists comfortable with ambiguity and eager to install structure, to finding domain experts with the experience to iterate quickly and zero in on the non-obvious areas for improvement. At the same time, we couldn’t compromise on the level of technical rigor we expected (as we felt it was necessary to keep us nimble) nor our requirement that we find people with a visceral passion for our mission and the fearlessness and empathy necessary to engage in hard debate as a team.</p><p>In many ways, the decision to form data science platforms is a natural extension of this process. Platforms allow us to appoint owners over critical data science problem domains, give them non-linear leverage, and give us a space to tackle strategic, broad-impact projects.</p><h3>Introducing the Data Science Platform</h3><p>When we made the decision to create distinct platform teams a year ago, Uber Data Science was in need of evolution. Our pattern of sharing best ideas via communally owned libraries, wikis, and <a href="http://jupyter.org/">Jupyter notebooks</a> weren’t scaling and each platoon started possessing “tribal knowledge”. The engineering infrastructure necessary to query data at scale, process experimental data, or build and maintain machine learning models wasn’t built with data scientists in mind; we became our database admin’s worst nightmares. Duplicative work was happening, or worse, we missed opportunities to leverage data because of friction.</p><p>The genesis of our platforms came from a simple idea: find the fundamental problems that every Uber data scientist faces, and form cross-functional teams dedicated to solving that issue awesomely. Collective ownership is great for agility and quickly identifying needs within the organization, but there’s real power in creating an organizational role where somebody comes to work every day with the mission of making machine learning at Uber incredible.</p><h3>The Power of the Platform</h3><p>Where does this power come from? Multiplicative impact. In a high-intensity, mission-driven workplace, it’s a fair assumption to assume that the team is operating at maximum individual intensity. Ergo, the only way to get more work done next week than last week, short of hiring, is to invest in <em>efficiency</em>. Leadership can achieve wins in efficiency through refining organizational process or mentoring and employee growth, but a platform is an investment in technological efficiency.</p><blockquote>“The only way to get more work done next week than last week, short of hiring, is to invest in <em>efficiency … </em>a platform is an investment in technological efficiency.”</blockquote><p>Uber operates under the seductive technology principle — no team can be coerced into adopting a technology — and technological efficiency is what makes our platforms seductive. Our machine learning platform has developed tools to make Uber’s features reusable and shareable across teams and is investing in push-button deployment of models to production. Our experimentation platform makes it simple to deploy experiments “over the wire” and collect analytics automatically. Our <a href="https://eng.uber.com/argos/">anomaly detection group</a> makes outage detection accessible and root cause analysis faster. All of these represent investments in the company’s <em>acceleration </em>through platforms, not just its speed.</p><p>Yet efficiency, while critical to the success of the company’s platforms and a way of providing immediate value, is only half of the story. Uber’s customer facing programs operate in a mission-driven style, focused on high speed of execution against a mission and delighting their proximate customer. It also naturally follows that their execution style becomes focused on rapid iteration and agility, with an emphasis on what we can do in the next few weeks to move the needle. However, not all data science efforts fit into this paradigm. As most of us know, research is an iterative process involving indeterminate amounts of discovery time, blind alleys, and tangents. Platforms provide a great opportunity for Uber to incubate some of its more strategic projects that have a scope or development cycle that simply wouldn’t be possible in a more mission driven environment.</p><h3>Where We’re Headed</h3><p>One year in, Uber’s made great strides in developing three foundational data science platforms — we’ve built a machine learning platform from the ground up, pushed our experimentation platform’s functionality and customer base into new business units, and turned anomaly detection for a passion project into a general service platform open to all of Uber’s engineers.</p><p>However, our story is far from complete. We’re working on a metrics platform that’ll scale our business concepts across all our streaming and batch data infrastructure, for example, and the existing platforms continue to grow. On that note, we’re always on the lookout for new additions to the group. If you’re interested in working with us, check out the job openings below; we’d love to have you join us.</p><p><em>Data Science Platform roles:</em></p><ul><li><a href="https://www.uber.com/careers/list/13258/"><em>Data Science Manager — Data Infrastructure</em></a></li><li><a href="https://www.uber.com/careers/list/13260/"><em>Data Science Manager — Experimentation Platform</em></a></li><li><a href="https://www.uber.com/careers/list/12789/"><em>Data Scientist — Monitoring &amp; Anomaly Detection</em></a></li><li><a href="https://www.uber.com/careers/list/12801/"><em>Data Scientist — Experimentation Platform</em></a></li><li><a href="https://www.uber.com/careers/list/12802/"><em>Data Scientist — Data Exploration</em></a></li><li><a href="https://www.uber.com/careers/list/12790/"><em>Data Scientist — Machine Learning</em></a></li></ul><p><em>Other Data Science roles:</em></p><ul><li><em>t.uber.com/DSjobs</em></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=965e2124edf8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>