<?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 Robert Neer on Medium]]></title>
        <description><![CDATA[Stories by Robert Neer on Medium]]></description>
        <link>https://medium.com/@robertneer?source=rss-eaf31fc5c581------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*N2hpEUjmpcoAQ4Pg.jpeg</url>
            <title>Stories by Robert Neer on Medium</title>
            <link>https://medium.com/@robertneer?source=rss-eaf31fc5c581------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 01:05:23 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@robertneer/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[Lessons on product management from using Dal-E]]></title>
            <link>https://medium.com/@robertneer/lessons-on-product-management-from-using-dal-e-64e74f92f89f?source=rss-eaf31fc5c581------2</link>
            <guid isPermaLink="false">https://medium.com/p/64e74f92f89f</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[product-thinking]]></category>
            <category><![CDATA[design]]></category>
            <category><![CDATA[creativity]]></category>
            <dc:creator><![CDATA[Robert Neer]]></dc:creator>
            <pubDate>Mon, 19 Feb 2024 23:26:27 GMT</pubDate>
            <atom:updated>2024-02-20T04:03:12.411Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>What I learned about product management from Dal-E</strong></h3><p>I have been using Generative AI quite a lot lately, mostly in the form of various ChatGPT4 agents and therefore mostly for written content — LinkedIn post co-creation, content re-structuring, some light curriculum-building, etc. I find it amazingly useful as editor, thought-starter, polisher. It often jumpstarts my thinking, helps me add or pivot structure, and generally makes me more efficient in creating written work. I like that I often disagree with suggestions made by the AI — it assures me my own voice, brand, perspective and conviction are thriving as I use these tools. You can be sure I have used it for this piece. : )</p><p>I have not spent much time with image generators, but recently I was helping my son with a poster. He is drummer for a band, American Broke, and they are looking for a new guitarist. American Broke is a Seattle-based crew, with a rock/grunge sound. Their original lyrics are broad but speak to the challenges, the highs and lows, of young adults making their way in today’s America — love, work, loss, economic uncertainty. I thought it would fit their vibe to leverage the iconic Uncle Sam Wants You recruiting posters for their guitarist-recruiting efforts — but with a twist. So, I took to Dal-E for some inspiration. My struggles to prompt Dal-E to help create the image I had in my head were insightful — here’s what I tried and failed to do and what I learned along the way about human-centered creativity in a GenAI landscape. I imagine these insights have both enduring and fleeting relevance as technology moves at pace.</p><p><strong>What I wanted</strong></p><p>I wanted a poster that showcased the band performing, but had a foreground image inspired by the Uncle Sam of “Uncle Sam Wants You For U.S. Army” fame –</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0-dqiSXxDQMYjrBvxNwKLw.jpeg" /></figure><p>I wanted the iconic elements with a Pacific NW Grunge-ish attire — Perhaps a hoodie under a Carhartt Jacket, and a wool beanie instead of the top hat. The band’s logo is red white and blue, so I wanted to preserve that as an element. It needed to be recognizable as the iconic Uncle Sam but with a wink and nudge to modern style and vibe…</p><p>I started by setting the expectations and the role I wanted Dal-E to play via a chat prompt. Then I went for it and described the whole poster and gave the band logo and an image of them on stage as a reference.</p><p><strong>What I got</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N-P-ZEFSTIf93FshcqP0Xg.jpeg" /></figure><p>Not bad — very iconic poster-like, with all kinds of weird text — this gives us an insight to how these image generation models work — they are trained on millions of tagged images with no real regard for what might be printed on them and so text elements inevitably are consistent from an aesthetic standpoint and completely untethered to linguistic reality; much like Large Language Models are over fit for conversational cohesion with no regard for factual integrity. Good to know on both…</p><p>I then decided I wanted to decouple the poster elements and just focus Dal-E on creating the Uncle Sam image I needed — I’d do the final layout myself. So, I tried to prompt in that direction:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uE3j8gW6RoG-UMEi_-tOAg.jpeg" /></figure><p>WHOA — easy, there Dal-E! What happened to all the good parts of the first image? Let’s try again: I tried “some tweaks — I like the art/sketch style of the original poster including the red, white and blue theme — can you do that? and revert the face and hair to be more like the original Uncle Sam but keep the Carhart and beanie”</p><p>I got this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p0QjZ14FJjhHozZJSDUTcg.jpeg" /></figure><p><strong>Critical insight</strong></p><p>I continued for several tries to get things right — asking to keep certain things but adjust other things — to no avail. See below the series of attempts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2a29tza98BlPSu9kFBxfKA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O7zzbH4vDWjcw8u5BJJUHg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zVCv8wbupb27TiXl1Vpcrg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jD-6VokmLyP442H2sjothw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q0vvTy4rrEP7c7zfc_p4Wg.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qBqwBXP3_HmOZcB8Y0zt4A.jpeg" /></figure><p>This and most image generation models are trained on millions of images and when asked to generate something they are good at creating a synthesis of all their training data. They can create something that meets most of the instructions as a whole but will get some specifics wrong. They do not assemble images the way a human designer might when they use digital tools; they do not understand the discrete layers or elements of an image — and so asking it to deconstruct and only alter one aspect is nearly impossible — it is just not how they work. What they are amazing at is sort of obvious from the name <em>GENERATIVE</em> AI — they generate from a soup of images and their notional markup, but without structured understanding of all the elements of a given image.</p><p>So — knowing what I know about prompt engineering I started over, trying to be crisp with a net new prompt about what I wanted. This got me the best one that was still not spot on — still no beanie for example — but a new vignette that includes a rough and tumble photographic background — nice touch!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ftAm7D4YG5CL_eH6MrWixg.png" /></figure><p>I tried ONE MORE TIME to keep everything the same and just change the hat…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*spkggJePzcWm98j0wcY06Q.jpeg" /></figure><p>Wahh waaaahhh….</p><p><strong>A human to the rescue?</strong></p><p>I wanted to give a human artist a chance to do this for $ so I found one through an online marketplace and gave them essentially the same prompts and assets and I got back a very precise rendering to spec…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fsu6YVl3BvDAd5fg2R-uYw.png" /></figure><p>But to be honest, it feels a little cartoony….and not terribly inventive . Now, this was a fee for service, so I did not expect a lot of iterations or exploration, but the human was a lot truer to spec — robotic almost — than the robot was. What I got was something that matched what I described but did not challenge my perspective, did not push the boundaries of my aesthetic, and was, I have to say, boring. It kind of reminds of a <em>King of the Hill</em> character…</p><p>We’re still talking about what set of images to use but this is the draft I have assembled so far in case you’re curious (and in case you are a talented guitarist):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R8ut6AR8F7S75xk0b5LKhQ.jpeg" /></figure><p>Also, the above is some proof that real designers should not feel threatened…</p><p><strong>So what?</strong></p><p>I’ve been working on some thoughts about the uniquely human in this current moment. For product managers there is a lesson here about unbiased wandering with your customer. Discovery and customer research is not about asking customers what they want and building that precisely — like the skilled illustrator I contracted with did. In many ways this is a function of the transactional nature of these marketplaces — the artist is rewarded and incentivized to produce to spec and get high marks on a review that asks “did this meet my expectations.” But as a product manager you need to be ready to wander openly with your customers, listening and synthesizing, experimenting and iterating and importantly learning from feedback and the process. This is where current Dal-E falls down — it does not learn or iterate; it cuts each attempt from whole cloth. Dal-E was much more orthogonally generative than the human illustrator. It challenged what I thought I wanted by presenting me some unexpected approaches. But it did not enter into a dialogue and did not synthesize across the full set of insights I offered it. What we need in this moment is product thinkers who can be generative, AND structured in their synthesis of learnings — the hybrid human will excel.</p><p>Many product managers find their energy largely consumed with activities deep in the software development life-cycle: user-story creation, acceptance criteria development, non-functional requirements, grooming, bug triage, and other day to day collaborative efforts in the BUILD phase…but how confident are they that they are leading the team to build the RIGHT thing — that it is anchored in customer insight and meeting a real, meaningful customer need in a way that drives business outcomes?</p><p>As more and more of the product lifecycle can be automated and augmented with AI, Product managers who bring their uniquely human skills to the role will thrive and create the most successful products. These human skills are critical in effective customer discovery, driven by listening, synthesis, creativity, empathy, experimentation and validation.</p><p>What do you think?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=64e74f92f89f" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>