<?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 Yuval Incze on Medium]]></title>
        <description><![CDATA[Stories by Yuval Incze on Medium]]></description>
        <link>https://medium.com/@yuvalincze?source=rss-71c7d65882e------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*8p-ygNtBa8Vs6-FE3YvlHA.jpeg</url>
            <title>Stories by Yuval Incze on Medium</title>
            <link>https://medium.com/@yuvalincze?source=rss-71c7d65882e------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 10:02:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@yuvalincze/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[I’m a Product Manager, and Cursor is My 10x Shortcut]]></title>
            <link>https://medium.com/@yuvalincze/im-a-product-manager-and-cursor-is-my-10x-shortcut-df6d3f16e9ee?source=rss-71c7d65882e------2</link>
            <guid isPermaLink="false">https://medium.com/p/df6d3f16e9ee</guid>
            <category><![CDATA[efficiency]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[cursor]]></category>
            <dc:creator><![CDATA[Yuval Incze]]></dc:creator>
            <pubDate>Tue, 09 Dec 2025 13:02:09 GMT</pubDate>
            <atom:updated>2025-12-09T13:02:09.834Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QMe0MXM7GRAO1z8N8E2bLQ.png" /></figure><p>You know how people always tell product managers, “Don’t get lost in the weeds,” and that we should always stick to the ‘what’ and not the ‘how’? I used to believe that, too. But lately, I’ve been using <strong>Cursor</strong> 2–3 times a week, and I’ve completely changed my mind. It’s not about becoming a developer; it’s about being 10x faster and more effective.</p><p>Give me a few minutes to tell you 3 beautiful stories that tell the story behind that feeling, and I believe they could be your stories as well.</p><h3>The Day We Saved the Demo</h3><p>A couple of months ago, we were gearing up to present the <strong>Brand Visibility Tool</strong> to our key partners. It was a huge deal, and we only had a few hours left before the meeting. I was reviewing the staging environment, and the UI just felt… off. A few labels were awkward, and the final data table alignment was janky. Traditionally, I’d have had to ping a dev, interrupt their workflow, write a quick ticket, and hope they got to it before the presentation.</p><p>Instead, I opened up Cursor. I used it to navigate the frontend repository, quickly figure out where the problematic component lived, and make the small CSS and text tweaks myself. It was magic. Within 30 minutes, I had polished the UI and even pushed a clean little PR to GitHub. The devs just had to hit ‘merge’ after a quick review. When we presented a few hours later, the partners were thrilled — the demo was seamless and professional. Cursor lets me control the <strong>brand experience</strong> instantly, without becoming a bottleneck.</p><p><strong>Bottem line:</strong> <br>Short path to a small tweak with a great impact on partners’ impression</p><h3>The Time We Avoided a Data Bottleneck</h3><p>Another time, one of our engineers was staring down a full day of painful, boring work. He needed a <strong>bulk of data</strong> from a specific API endpoint for a new feature build. Getting this data required complex authentication and figuring out how to stitch together multiple pages of results, a classic half-day data engineering sinkhole.</p><p>I walked over, opened my terminal, and used the Cursor. I described the data I needed, and the tool immediately helped me generate the complex script to hit the API, handle the bulk pagination, and even throw in a few quick checks on the data (like counting rows and getting some basic stats) to make sure it was clean. Within ten minutes, I had the exact, verified dataset he needed, sitting in a file ready to use. That’s a task that would have cost the team half a day of focused effort just a couple of years ago. I simply <strong>removed the dependency</strong> and kept the development velocity high.</p><p><strong>Bottom line: <br></strong>Save<strong> </strong>0.5 Dev day net, more than a week (in the best case), bruto.</p><h3>The Internal Platform Mockup That Was Perfect</h3><p>My favorite use, though, was for our new internal platform, a tool designed to help our teams generate <strong>10x more image assets</strong> for all our campaigns. This project touched Design, PPC, and Development, meaning requirements had to be ironclad.</p><p>I used the Cursor not to write code, but to <strong>rapidly build a full-fidelity mockup</strong>. It helped me structure the platform’s flow and interface incredibly fast. I took that mock, iterated instantly with the Design team on the look and feel, and then walked through the entire user journey with the PPC team to ensure every step matched their workflow. By the time I presented the mock to the developers, the whole thing was perfectly defined. There was no ambiguity. We skipped weeks of design-dev back-and-forth and went straight into building because the requirements, thanks to the quick iterations supported by Cursor, were bulletproof.</p><p>So yeah, the traditional advice is wrong. Using the Cursor isn’t about being in the weeds; it’s about having a tool that gives you <strong>superpowers</strong> to fix demos, unblock engineers, and define products perfectly, all while still keeping your eye on the bigger strategic picture.</p><p><strong>Bottom line:<br></strong>Save weeks of iterations preceding tool launch in 1.5 months (or more).</p><p>If being ‘in the weeds’ means getting a 10x velocity boost for your entire team, isn’t it time we redefined the job? What’s holding you back from finding your own PM shortcut?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=df6d3f16e9ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Stop Guessing: How to Pick the Best AI Visibility Tool for Your Brand]]></title>
            <link>https://medium.com/@yuvalincze/stop-guessing-how-to-pick-the-best-ai-visibility-tool-for-your-brand-9c902e555050?source=rss-71c7d65882e------2</link>
            <guid isPermaLink="false">https://medium.com/p/9c902e555050</guid>
            <category><![CDATA[brand-strategy]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[visibility]]></category>
            <category><![CDATA[marketing]]></category>
            <dc:creator><![CDATA[Yuval Incze]]></dc:creator>
            <pubDate>Tue, 26 Aug 2025 11:26:22 GMT</pubDate>
            <atom:updated>2025-08-26T21:35:25.222Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xc1MzEmCSF9f-EWi5XnyLw.png" /></figure><p>Let’s start with the obvious.</p><h4>Why does every brand need a tool?</h4><p>I’ll make it short since it’s pretty clear at this point. AI engines such as ChatGPT, Gemini, Perplexity, and so on are growing rapidly.<br>The old competition in Google search user attention starts to transform to those engines.</p><p>Good brand, wants to be where its audience is.</p><p>Therefore, the first step for your brand is awareness: start tracking your <em>visibility</em> across all AI engines.</p><h4>How do those visibility tools work?</h4><p>First, quick but important note on <em>how AI engines work, </em>crucial for you to understand before continuing, before jumping to the visibility tools mechanism:</p><blockquote>AI engines operate somewhere between a ‘computer memory’ and a ‘lottery’: on one hand, they can recall patterns, facts, and associations with incredible speed (like memory); on the other hand, every answer has an element of randomness, meaning you don’t always get the same response twice (like a lottery).</blockquote><p>And back to reality - no tool truly knows your audience or actual user activity within these AI engines. However, they can give you a useful reflection by <em>simulating the prompts</em> your audience is likely to use when searching for a solution in ‘context areas’, and your brand offer should be relevant.</p><p>For example, if my brand is HubSpot, I offer CRM, Marketing, and Sales management software (and many more). In this case, as a brand, I want my audience to encounter me in prompts like “Best CRM software for my business?”, “I’m looking for a tool to manage my sales operation” and so on.</p><p>To summarize, it’s not bulletproof, and it doesn’t reflect reality since AI engines are way too complex and personalized and not a simple memory with a one-to-one answer (aka lottery metaphor), but it does give a ballpark, and most importantly, it’s the best shot currently.</p><h4>What tool should I pick then?</h4><p>The key factors when you make a decision:</p><ol><li><strong>Need and Budget:</strong><br>a.<em> Enterprise</em>: <br>You probably need a tool that covers multiple geos, offers cross-market benchmarks, integrates with your BI stack, and has strong compliance/security layers. You’re not only buying data, you’re buying <em>trust</em> at scale. Budget: 15–30K a year.<br>b.<em> SMB</em>: <br>You want something simple, affordable, and fast to set up. You don’t need ten dashboards or a complex API feed. A clean UI, a few key visibility reports, and maybe some lightweight optimization insights will do the job. Budget: 2–7K a year</li><li><strong>Trust: <br></strong>Visibility data is still new, fuzzy, and far from standardized. That means you should ask:<br>- Are they a new player or a player with a legacy and authority in the market?<br>- Do they show <em>how</em> they simulate prompts?<br>- Do they give transparency on updates, coverage, and data sources?<br>If you don’t trust the reflection they give you, the rest doesn’t matter.</li><li><strong>Data:<br></strong>a.<em> Relay your brand data only</em>: <br>Inject pixel and adapt your prompts and their deliveries to your brand data only.<br>b.<em> Include cross-market data</em>: <br>Tools that show you rely on their cross and unique data of user behavior, such as volumes, intent, and quality, might be pricier, but usually way more powerful if you want to know where you win/lose vs. others.</li><li><strong>Optimization</strong><br>a.<em> Mapping focus areas</em>: <br>Which prompts will influence the drive impact, and which prompts don’t really matter?<br>b. <em>Content generation</em>: <br>Some tools already suggest or auto-generate content to improve your visibility.<br>c.<em> Technical analysis</em>: <br>The geek layer: analyzing model responses, tracking drift, and providing structured recommendations for dev, SEO, or marketing teams.</li></ol><h4>The Winner is…</h4><p>The competition just started.</p><p><strong>The winning formula (in order of priority) is: </strong><br><em>Quick GTM </em>X<em> Intent Data </em>X<em> Effective GEO </em>X<em> Dev Velocity</em></p><ul><li><em>Quick GTM</em>: <br>Sometimes it’s just important to be the first (bitcoin), this is one of those cases. In cases like the Visibility tool, where the core product doesn’t have much difference between all competitors, and one brand needs only to be the first might be enough.</li><li><em>Intent Data</em>: <br>When everyone is shooting in the dark, trying to guess users’ prompts, tools with user intent data, who knows where the volumes, intent, and quality will help their brands to be more focused on the right prompts, which translates to impact faster.</li><li><em>Dev Velocity</em>: <br>Always right, in this specific case, as I said earlier, no one got a significant competition advantage (except the data), therefore, the one who will deliver the right product and react to the market the fastest will have an advantage.</li><li><em>Effective GEO</em>: <br>Finally, why last? Because GEO knowledge is still weak across companies, why does it last? Since GEO knowledge currently is pretty poor across companies, it’s a new area, no one really knows what to do, and if so, at the moment those tools will share their secrets on their platforms, it wouldn’t be a secret anymore.</li></ul><p>Like that? comment or give a like for fun :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9c902e555050" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why It’s Still Hard to Find Good Solutions? Even With AI]]></title>
            <link>https://medium.com/@yuvalincze/why-its-still-hard-to-find-good-solutions-even-with-ai-ea1a58cb1ca4?source=rss-71c7d65882e------2</link>
            <guid isPermaLink="false">https://medium.com/p/ea1a58cb1ca4</guid>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[disruption]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[predictions]]></category>
            <dc:creator><![CDATA[Yuval Incze]]></dc:creator>
            <pubDate>Mon, 11 Aug 2025 06:47:05 GMT</pubDate>
            <atom:updated>2025-08-12T19:00:01.856Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*sIUZCjYp4qcLJbTc" /><figcaption>Photo by <a href="https://unsplash.com/@miquel_parera_mila?utm_source=medium&amp;utm_medium=referral">Miquel Parera</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>We live in a golden age of tools.</p><p>AI can brainstorm, write, code, design, summarize, and analyze instantly.<br>And yet, despite that power as our co-pilot, we’re still struggling with <br><strong>finding good solutions</strong>.</p><h3>So… What Is a Good Solution?</h3><p>Changes that, once in place, make the system better aligned with your goals.</p><p>Allegedly, it’s a sentence that doesn’t help much… you&#39;re right.</p><p>But I think it makes the point. Why? It might not say anything practical, but underneath, it’s really about <strong>prediction</strong>.</p><p>Meaning, we don’t know what a good solution will be; we can only guess, which means the only thing we <strong>can do</strong> is to have a really good guess.</p><p>Therefore, while we lack generic, practical ‘how-to’ advice, we can be confident that <strong>our approach is sound</strong></p><p>So if it’s just about having the right approach, why is it still so hard? Because getting there means going through <strong>discomfort</strong>.</p><p>You have to ask harder questions. You have to guess better than your competitors. You have to tackle some good arguments that claim your good solutions aren’t good guesses.</p><p>You have to give a sense of certainty when you deal with discomfort.<br>Keep your and everyone involved&#39;s reputation when risking it at the same time.</p><h3>Hi AI, Problem Solved?</h3><p>The first question should be, does it improve our predictions for having a good solution? Not quite.</p><p>AI is brilliant. <br>It helps us move faster, write clearly, and see broadly.<br>But it doesn’t fix our thinking. In fact, it might make it harder to go deeper.</p><p>Because AI is excellent at producing things that sound good.<br>And when something <em>sounds</em> good, we often stop asking if it’s right.</p><p>Because it gives us answers, not questions.<br>Because it builds on what already exists, not on what hasn’t been thought yet.</p><p>Because it reflects our mindset, but rarely challenges it.</p><h3>So What Can We Do?</h3><p>Improves our accuracy in guessing.</p><p>How to do it? I wish I knew.</p><p>The only way to get good at anything? Start.</p><p>Here are <strong><em>5 guesses</em></strong> I made while trying to embrace the <strong><em>discomfort</em></strong>:</p><ol><li><strong>My agent was born</strong> - I’m seeing the emergence of something strange and new — a <strong>digital character</strong> we didn’t fully intend to create. The AI we talk to isn’t just a tool anymore. It has a tone. A rhythm. A personality. It shapes us as we shape it. It’s becoming a “someone.”</li><li><strong>Remember Heroes?</strong> I’m noticing how daily life starts to feel like a game, not metaphorically, but structurally. We’re making moves. Earning tokens. Getting rewarded by algorithms. Gemini gives you credits. GPT gives you usage quotas. We’re all playing, whether we realize it or not.</li><li><strong>Why so serious?</strong> I’m wondering why almost no one is talking about <strong>gaming</strong> as a native domain for AI. Not just to build better games, but to reimagine systems of learning, behavior, even identity. Games have always been simulation spaces. And now the simulations are starting to talk back.</li><li><strong>Daddy, what was a website?</strong> I’m questioning what happens to the browser itself. Do we still “visit websites” in the future? Or do we let AI interact on our behalf, talk to services, compare products, and make bookings, while we stay inside the conversation loop? What happens to the open web when the AI becomes the front-end?</li><li><strong>Modeling replaces following?</strong> I’m revisiting the idea of <strong>social proof</strong>. For years, we trusted likes, followers, reviews, and testimonials. But what happens when AI consistently recommends better things than our friends do? What happens when it becomes a better judge of quality, not just faster — but more <em>right</em>?</li></ol><h3>Keep guessing. Risk your reputation.</h3><h4>Follow for more flawed (but honest) attempts.</h4><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ea1a58cb1ca4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Where I can use GPT-4 as a Product Manager on Day-to-day work?]]></title>
            <link>https://medium.com/@yuvalincze/where-i-can-use-gpt-4-as-a-product-manager-on-day-to-day-work-3cb2b3fe8763?source=rss-71c7d65882e------2</link>
            <guid isPermaLink="false">https://medium.com/p/3cb2b3fe8763</guid>
            <category><![CDATA[genai]]></category>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Yuval Incze]]></dc:creator>
            <pubDate>Sun, 10 Mar 2024 10:22:13 GMT</pubDate>
            <atom:updated>2024-03-12T13:17:58.384Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LmBa89pTLZFoJQBU" /></figure><p>There is a statement in many industries which says “everyone is nobody” and “everything is nothing” , It’s right when marketing teams try to segment the right audience and suggest “anyone” fit, and it’s also right when people who suggest you to use AI such as GPT-4 for everything!</p><p>It’s a wrong answer if you try to describe the practical uses, which people most curious about. but it’s not a wrong answer when talking about the potential. So GPT-4 can help you for EVERYTHING! now let’s try together to unlock the potential.</p><p>Before i’ll give a few examples, I want to give an abstract guidelines for you to have a better intuition to the question “Is GPT-4 can assist here?”</p><ol><li><strong>Is there any data to rely on?</strong> we see the most exciting update in GPT-4, you basically can upload whatever files you want: PDF, XLS, Images etc.</li><li><strong>Is it structured enough? </strong>and i don’t mean to a high definition tables since the magic of the AI is to ‘understand’ ideas which we as a human recongize as unstructured, but, structure means also common language. Common language means, i can explain the AI what i want and there is high chance it would understand the instruction.</li><li><strong>What would an X expert do?</strong> GPT-4 use an existing tools, modules, capabilities, which means it can produce anything you ever see data analyst does, designer produced once or code a developer can create. Try to asks for deliveries you saw (but wasn’t dare to ask)</li></ol><p>Enough with the guidelines, here is some examples:</p><h3>1/3 Project management</h3><h4><strong>Export</strong></h4><p>From Monday</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Nveva52SAVQYTh6w4dnGoQ.png" /></figure><p>OR Jira</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m23psdBTmVaEd3QeHipPKQ.png" /></figure><p>And basically any other tool, here I chose CSV since it’s the most common format, but feel free to choose other formats if you feel the different structure have any adventage for you</p><h4><strong>Import</strong></h4><p><em>CSV </em>to GPT-4</p><h4><strong>Results</strong></h4><p>I’ve asked for a summary of number of tasks + est. effort in () group by task owner and split to the task status:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DfwTeayyGh2MgdfNzwVPlQ.png" /><figcaption>Blurred the left column with my colleges names (with GPT-4)</figcaption></figure><p>Then, I’ve basically asked for a Pie Chart of the design resource portion any department had, and after a few aesthetic requests, such as different color per pie, make sure pie titles aren’t override each other (which was challenging as you can see..) I had it:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3G46Jaa2C5zIBK9biTeP8g.png" /><figcaption>Have to blurred that, but the GPT-4 makes it crystal clear.</figcaption></figure><p>So I know people might think now, wait those stats are pretty basic, so yes they are basic, but, we are talking about a day-to-day stuff and the magic here is — (1) It takes way less time than usual, I will say 1/10 times actually; (2) No expertise needed; (3) The ability to instantly customize every aspect from chart type, theme, metricas and calculations and so on and so on it’s amazing.</p><h3>2/3 Design (or UI/UX)</h3><h4><strong>Export</strong></h4><p>From Figma</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/602/0*Kmwns2g87iRew_9d" /><figcaption>B<a href="https://www.businessheroes.com/tools">usinessHeroes.com/tools</a></figcaption></figure><h4><strong>Import</strong></h4><p><em>An Image</em><strong> </strong>to GPT-4</p><h4><strong>Results</strong></h4><p>First, I’ve asked for to describe the average user journey:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wmwK_LMeh83nKSacur8Hiw.png" /></figure><p>Then, I described the website’s premise and asked for its evaluation, specifically on aspects we executed well that should be emphasized, and on elements that might be missing and should be added:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o2TTSXWp4EHioCjgq-__Pg.png" /><figcaption>Yes, we are professionals.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EJ5CNaOuG7vq8HNJ4sNzSQ.png" /><figcaption>Oh, wait nobody is perfect.</figcaption></figure><h3>3/3 Data Analysis</h3><h4><strong>Export</strong></h4><p>From Tableau (with a screenshot)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*qq_XuM6tXYmaEnIA6EQ7Qg.png" /></figure><p>OR XLS / CSV files from Excel or Spreadsheet</p><h4><strong>Import</strong></h4><p><em>Screenshot</em> and <em>CSV</em><strong> </strong>to GPT-4</p><h4><strong>Results</strong></h4><p>I’ve asked for insights suggestions for the a/b test results view shared above. After a few words to describe what KPIs I cared about, the GPT-4 perfectly ‘import’ the data from the screenshot and understand results meanings:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F4D6EXnDYwR_DwAkhQqLSg.png" /></figure><p>In another analysis, I shared an AI Chatbot data (what the bot write to the user) with the GPT-4, and I’ve asked for a Word-cloud to have a sense of this bot generates to the users:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/828/1*_TITakbEWmuqx1DL_Mhwyg.png" /></figure><p>That’s it for now. I appreciate your time! I’m looking forward to your ideas for follow-up discussions and would love your likes :)</p><p>Liked the content? read now my previous content regarding <a href="https://medium.com/@yuvalincze/its-always-not-just-the-product-it-s-always-not-just-the-product-c53302077101">a recent challenges I had with integrate an AI Chat</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3cb2b3fe8763" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[It’s Always Not Just the Product]]></title>
            <link>https://medium.com/@yuvalincze/its-always-not-just-the-product-it-s-always-not-just-the-product-c53302077101?source=rss-71c7d65882e------2</link>
            <guid isPermaLink="false">https://medium.com/p/c53302077101</guid>
            <category><![CDATA[product]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[chat]]></category>
            <dc:creator><![CDATA[Yuval Incze]]></dc:creator>
            <pubDate>Sun, 17 Sep 2023 20:30:22 GMT</pubDate>
            <atom:updated>2023-09-18T17:11:29.218Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*aIP2w4WcSSejh7kz" /><figcaption>Photo by <a href="https://unsplash.com/@tamarushphotos?utm_source=medium&amp;utm_medium=referral">Tamara Malaniy</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>It’s Always Not Just the Product — More Than Ever When Talking About Gen-AI.</h3><h3>It’s always not just the product</h3><p>Pisa it’s not a building, when it comes to a product, whether it’s a UI, backend , small or large feature, the important aspects of the product — the ones that can significantly impact how users perceive the product — are often not the core product itself. We can use three common examples almost everyone would be familiar with in order to illustrate this point: Netflix, PlayStation, and the NBA.</p><p>What is Netflix’s product? Streaming TV shows, movies, and more. <br>What is the NBA’s product? A basketball league. <br>What is PlayStation’s product? A video game console.</p><p>However, these descriptions don’t capture the essence of these products fully. Netflix is more than just a streaming service; it’s a platform that turns users into content addicts, predicting and delivering exactly what they want to see, and effectively competing with sleep. It rarely (00.001%) disappoints users when they seek out content. Does this sound like the core product of a streaming service?</p><p>NBA, gives you the best, transparent, colorful, stylish, organized, beyond human skills show off . is it the basketball? is it a reality show?</p><p>And lastly, as a loyal PlayStation user, let me tell you something: their console sometimes feel like a the computer before someone thought to adapt it to humans.<br>Everything there is HARD. You want to play? Oh, wait, there’s a 30-minute update. Oh you didn’t know? the controller is empty please wait. Why there is speaker on the controller if the volume is so low? Never mind. Let’s play together online against someone else… Not again, we both have to sign in? Amazing let’s type it down with a Nokia keyboard and thanks god if we remember the password. And yet, being first and having exclusivity on the most popular games, lead them to ±90% console market share (Statista).</p><h3>So why is it relevant to Gen-AI as well?</h3><p>There is one rule in Gen-AI as well, It’s not about the Gen-AI.<br>As time passes, we are exposed to more and more magical moment the Generative AI and when it comes the practical usage, only a few solutions are solid enough to provide a real value for the user.</p><p>In my perspective here, the new products 2,5,10 years from now that would using based on Gen-AI algorithm, won’t be successful as there algorithms for a few reasons:</p><ul><li>Everyone can use it, except of regulation which might jump in some point, Gen-AI capabilities would be open to everyone and it doesn’t matter if we are talking about are real open source models which we already see (And seems like it’s even Facebook agenda for now) or we as a companies would pay for that, since the business model that based on pay-per-tokens is here to stay.</li><li>Let’s be honest, we as individuals or even as a big companies are not genius (or have any other resources such as money, human power and so on) enough to improve those models performance, meaning, the core product.</li></ul><p>At that moment I would assume we largely agree, and if so, the right thing to do is to focus on all other product aspects except the product core.</p><p>Next, I will talk about what do i have learn during ad-hoc AI Chat for a comparison site called top10.com, hint, it’s won’t include algorithm diagrams.</p><h3>What’s about a Gen-AI Chat product without core?</h3><p>OR why ChatGPT did what they did.</p><p><strong>It’s the UX/UI</strong> <br>ChatGPT didn’t tell anyone it’s an AI platform. we just got it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*7yBgAhG71LJazINMz7VT-w.gif" /></figure><ul><li>The on-spot <strong>streaming typing </strong>— yes, this is how we are going to call this ongoing writing/text creation you see at the GIF above. If you ask me, it was a solution to mitigate the ridiculous period of time the user is waiting for ChatGPT answer. but hey, now it’s a thing.</li><li>The <strong>Text Cursor indicator</strong> we managed to forget from the very first computers, Word editor, and so on.</li><li><strong>One message at a time</strong> (or AI is teaching humans how to be polite), I would guess this is their way to give the algorithm a coherent context, and more than that, keep the response time and request load reasonable.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4M3eV5eB-budv27YatXFuQ.png" /></figure><p>Switch to:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BxM7C0lV4abl4F-gZ_xmUA.png" /></figure><ul><li><strong>Re-generate;</strong> none of the chat bots or the customer support ones we’re familiar with can produce the same question but different, at scale.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Pc-jl4uRve6HTZNKo3-qGA.png" /></figure><p>P.S. I have also wondered why they didn’t give us response with a generated <strong>quick replies </strong>and honestly, I didn’t have a conclusive call here. Although I would guess it’s related to some technical considerations (response time, request load, etc.) or they’re not confident enough to show some pre-made connections the model made without human interaction in between, to avoid blame in cases of non-PC content.</p><p><strong>It’s the code, I mean, the prompt<br></strong>Prompt engineering is real.</p><ul><li>Prompt creation feels more like coding than a writing assignment. It would change his response based on the mission, the system’s character, and the expected audience that would consume his response.</li><li>More than that, (1) CAPS LOCK does matter. (2) “You must do…” is not equal to “You should do…”, and (3) even the location where you place the instruction within the prompt is important.</li><li>It can read (understand?) formats that people are not capable of reading, such as tables, proper code, and anything basically produced by letters, signs, and symbols.</li></ul><p><strong>Last but not least, It’s the QA<br></strong>here is the AI revolution is reflected the most.</p><ul><li>There is absolutely no way, I repeat, no way, to conduct a pre-QA test for that. Okay! I’ll be honest, I couldn’t find any good way to perform QA for the chat, but here is a way.</li><li>QA stands for quality assurance, which, until now, we have been accustomed to thinking that as product builders, we would be capable of catching the bugs before the user. This way of thinking relies on the idea that our product is predictable and follows a deterministic finite path.</li><li>Apparently, Gen-AI does not work like that; it generates its response based on some likelihood that we would never know in advance.</li></ul><p>So yeah, I didn’t talk about it, but I’m on my way to launch an AI chatbot. I’m pretty excited and plan to share my lessons and thoughts from time to time. From your POV, it can be even better, as you would benefit from a stranger but still useful person mistakes :)</p><p>Thanks for your time, more to come. See you next time 👌</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c53302077101" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>