<?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 nobody cares dude on Medium]]></title>
        <description><![CDATA[Stories by nobody cares dude on Medium]]></description>
        <link>https://medium.com/@jed.deaver?source=rss-e8267c311c67------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*WRvPgC28ov0DeBMWqdfvyQ.jpeg</url>
            <title>Stories by nobody cares dude on Medium</title>
            <link>https://medium.com/@jed.deaver?source=rss-e8267c311c67------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 22:17:51 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@jed.deaver/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[doing my part]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@jed.deaver/doing-my-part-f6ce674bbab3?source=rss-e8267c311c67------2"><img src="https://cdn-images-1.medium.com/max/2600/1*9PDaaMJQ_wf0r-DatDd2Ig.jpeg" width="2752"></a></p><p class="medium-feed-snippet">Please read Lucas Hahn&#x2019;s story here: https://outsidein51.substack.com/p/societal-strongholds-on-profound</p><p class="medium-feed-link"><a href="https://medium.com/@jed.deaver/doing-my-part-f6ce674bbab3?source=rss-e8267c311c67------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@jed.deaver/doing-my-part-f6ce674bbab3?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6ce674bbab3</guid>
            <category><![CDATA[autism]]></category>
            <category><![CDATA[neurodiversity]]></category>
            <category><![CDATA[autism-awareness]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 19:19:41 GMT</pubDate>
            <atom:updated>2026-11-09T19:21:53.183Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[I get it, universe]]></title>
            <link>https://medium.com/@jed.deaver/i-get-it-universe-6a0a8f1a7835?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/6a0a8f1a7835</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[self-help]]></category>
            <category><![CDATA[publishing]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Tue, 08 Sep 2026 09:06:17 GMT</pubDate>
            <atom:updated>2026-09-08T09:06:17.904Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve been getting a pretty clear message for some time, I just wasn’t listening.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZaOA4i5SWqWfjq5CM99G2g.png" /></figure><p>There was a time where I contributed openly and freely to the internet conversation. I used to spend countless evenings on IRC chatting with students at SUNY Genesee Community College about hacking different things, computers in general, life, all of it. If you’re unfamiliar, IRC is the grandfather of things like Discord, Slack, and WhatsApp. Except when someone shared a photo on IRC it usually took 2 hours to download all of the parts, then reassemble them into the source image. We’ve got it nice these days.</p><p>I was very prolific in my writing back then. I would just throw a bunch of thoughts together, sometimes even just writing notes in a notebook, and pass that along to others. I wouldn’t worry about attribution or credit for anything: if someone were able to turn my thoughts into something tangible, that’s great.</p><p>I’ve always had a problem slowing my brain down - I just wrote about it in <a href="https://medium.com/@jed.deaver/bringing-others-along-e81a1caef498?sharedUserId=jed.deaver">my last article</a>. I struggled with the essay portion of English all through school. And this is what led to the revelation that I am a terrible storyteller.</p><p>It’s fine, I used the words. Stings a bit knowing how accurate it is, but that is what these journal entries on Medium are intended to be.</p><p>Before I was required to socialize my thoughts professionally, it was a lot easier. Most of the engineers I’ve been fortunate enough to work with over the years were able to understand the same shorthand. It’s an ebb and flow from working together. It’s more improvisational than can be easily documented. It’s going to the balcony and observing myself in the moment, to perceive myself the way others see me.</p><p>Thanks for all of the advice, Alix!</p><p>As I think about it, it has never been about my ability to perform, instead, it’s all about my ability to document my accomplishments. When it comes to typing, my hands feel heavy these days. There is no comfortable position I can find to get my thoughts out. And my mind is restless. It’s impossible to focus. There’s just so much going on all of the time.</p><p>And so much that I want to comment on. So much stuff that a couple of years ago, I started posting videos to YouTube to get some of my thoughts out of the way. It’s sorta turned into a technical essay of sorts, with an AI host that kinda represents my current state understanding of the AI landscape.</p><p>But how do I get these points to land? How can I get them to show up on the right people’s radar? And how do I get past that super unhelpful feedback about what information needs to be conveyed in a certain situation. I’m likely autistic and I need structure, damn it! I think this is where I fell out of love with English, especially when it came to my English teachers. I remember being frustrated by the feedback I always got back then. It was always the same: something along the lines of “needs more” or “not enough”. I remember I wrote 2000 words of a 1000 word assignment just to beat that feedback, but it was still there.</p><p>My career coach said the same thing as those teachers, but in a way where I felt like I have an idea on how to move forward. This may seem natural to a lot of you, but I could never find a way to explain this to AI. My ADD wouldn’t allow it. I couldn’t concentrate long enough to complete my thoughts.</p><p>Not all feedback was as unhelpful. I do have a challenge with passive voice. I need the reminders. And I know I have an issue repeating myself at times while trying to make a point, but I don’t always catch it. I need a proofreader. And I need someone who will be critical but helps me get things out of the way. Help me get my point across effectively.</p><p>And I need a fact checker. There is nothing worse than tripping up on my own words because of an inaccuracy in some of the supporting data. Hey, it’s happened before. And it would be nice to have a professor level opinion of what I have written, with brutally honest feedback, so I can improve.</p><p>So I wrote an agent</p><p>In season 15, episode 8 of King Of The Hill, Peggy Hill outlines a number of issues with a paper submitted by a former student. I used those rules, like “the conclusion must support the thesis” and some other rules I picked up here and there, then asked the agent to grade what was submitted as if they were a college professor. So I set a bar for myself. This blog started with a grade of C-. The final draft got an A, though I feel it’s a solid B effort myself.</p><p>(As a sidebar, when proofreading the first draft of this essay, AI corrected me that King Of The Hill only ran for 13 seasons, however, when I reminded AI of the Hulu revival, it was quick to correct. It also changed my grade unnecessarily.)</p><p>Without a doubt, one of the best use cases for AI is tasks like this. Is that benefit worth the cost of sending this information to the internet? Yeah, that’s where it’s going to end up eventually, so why not? So I wrote a skill in Gemini to act as this college level professor and give me actionable feedback. My agent would ensure the conclusion supports the thesis, that there are no open threads, that all information is verifiably accurate factually, especially when it supports an opinion. At the end, it would give a grade based on these standard criteria.</p><p>My first few attempts using this proofread skill were disastrous, not because of the feedback, but because that feedback was back again: my words were not enough. AI went one step further, actually suggesting how I could help improve. It helped me form a logical conclusion as I was wrapping up this essay.</p><p>Over the years, I’ve indexed too heavily on the feedback of others. I didn’t always look at that feedback as helpful, in fact, over time it became detrimental. I tried too hard to please everyone. I’m acclimating to the way AI works: it is trying to get the words I say into a state where they can be more broadly consumed. This doesn’t mean burying my voice in the feedback, instead, how do I use the feedback to make that voice stronger and more forceful?</p><p>These journal entries help. I’ll be using the written entries like this for more reflective concepts, while the technical stuff is for YouTube.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6a0a8f1a7835" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[bringing others along]]></title>
            <link>https://medium.com/@jed.deaver/bringing-others-along-e81a1caef498?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/e81a1caef498</guid>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[mental-health]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Sat, 05 Sep 2026 16:37:22 GMT</pubDate>
            <atom:updated>2026-09-05T16:37:22.004Z</atom:updated>
            <content:encoded><![CDATA[<p>I just figured out something about myself that I wanted to capture.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cY5SiDf5DMiryfleVIVmaA.png" /></figure><p>My entire career, I’ve heard the same feedback: I needed to teach others how to do the things I do. I had a career coach several years ago who coached me to “take people with me”, to explain my process to others. And that is a really hard thing for me to do.</p><p>Why is it so hard to explain how I came up with a particular conclusion? Consider this example:</p><p>I remember a particular production support issue from years ago. As much as I detest “being” on support calls, they are kinda my wheelhouse. I get to pull together all of the variables, process them as a unit, and start eliminating possibilities. And I’ve got quite a reputation at getting to a solution fast. In this case, another engineer and Ireached a conclusion 4 hours ahead of the rest of the call. His thought processes and mine were aligned. We developed a shorthand on the call, messaged one another off and on with updates, and when we felt good about our conclusion, we spoke up.</p><p>(BTW — There is nothing worse on a problem call than someone blowing the entire call up to get a status. If you do that, please stop. It slows things down. )</p><p>One day, I’ll do an unsolicited code review of Zoom, but Zoom actually makes things worse. On voice phone lines, there’s no delay when multiple people are talking, but Zoom seems to create this weird timing issue with calls. There’s no room for conversational debate. If latency were my only concern with Zoom the solution is simple: mandate audio-only bridge lines for support incidents. There’s other stuff going on, so I get the reasons not to go that route. Digital voice is still just ever so slightly glitchy , and those glitches disrupt conversational flow. This factors into my problem solving ability.</p><p>Status bombs aside, sometimes there are ambient parts to the conversation that provide huge context clues to the actual problem, so I actively listen as much as I can. Using our shorthand allowed us to talk without saying anything. You see this in a lot of places. Medical professionals have a way of speaking to each other. That comes from experience; you are able to pick up what each other are saying. There is no way to teach that kind of intuition.</p><p>For someone to be taught they must be willing to learn. And a lot of people, sadly, just want to be told what to do. Especially in stressful moments like support calls, people hate the ambiguity. It is difficult for a lot of people to deal with. People like certainty. To develop intuition means working to understand that uncertainty.</p><p>At the end of the support call, I was asked to write a training guide for how we reached our conclusions, in case something like that happened again. We documented our root cause analysis, but I was asked for something more, specifically, how could I teach someone else to do what I’d done?</p><p>To this day, I have no idea how to address this. In that particular situation, we did document we had observed extreme latency in trying to connect to one of the servers in question. This was due to oversaturation of the network we used to connect. That is not an application problem, it’s something else. That was an observation: one variable in the list.</p><p>Analyzing that signal along with the hundreds of other signals coming in is impossible to explain, let alone teach to others. And those signals change from problem to problem.</p><p>Intuition is built from experience — thousands of observations piled together over years of experience. How would you suggest taking someone on such a journey?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e81a1caef498" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[misadventures in AI]]></title>
            <link>https://medium.com/@jed.deaver/misadventures-in-ai-f6270d5a1903?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/f6270d5a1903</guid>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[torch]]></category>
            <category><![CDATA[musetalker]]></category>
            <category><![CDATA[cuda]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Thu, 16 Jul 2026 19:23:38 GMT</pubDate>
            <atom:updated>2026-07-18T11:26:05.507Z</atom:updated>
            <content:encoded><![CDATA[<p>any engineer worried about the ongoing AI onslaught, please consider the following. . . .</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9IRiXYwotkoeI9AYXgylwQ.png" /><figcaption>remarkably accurate depiction what it takes to get AI working</figcaption></figure><p>Depending on who is talking, AI is coming for or has already displaced your chosen profession. As someone who has somewhat immersed themselves in AI for the past few months, I’m not convinced.</p><p>My reason is kinda self-fulfilling: AI is not capable of fixing itself.</p><p>In fact, AI is like having an infant. It knows something is wrong, but because it is used to just working, AI is absolutely useless at telling you where the problem is.</p><p>My experience is it could be damn near ANYTHING wrong. I have an older PC I’ve been modifying to suit my AI experimentation needs. In these few weeks, I’ve run into all manner of issues, including:</p><ul><li>out of memory — this happens A LOT. The old adage that RAM and hard drive are like real estate in that you can never have TOO much of either definitely applies. RAM in this instance covers both system RAM and GPU RAM.</li><li>CUDA driver issues. I circled the drain on this one for a long ass time. I finally found a cheap ($150) older GPU, got the necessary power cable for my video card, which is a new experience, but I struggled finding the right driver versions.</li><li>missing or terse error messages. If you do not understand how logging works on your chosen platform, stop now. You’re only going to frustrate yourself. After scouring application and system logs to little avail, I found the exact cause of my problem on my console login screen:</li></ul><blockquote>NVRM: The NVIDIA Quadro M4000 GPU installed in this system is<br>NVRM: supported through the NVIDIA 580.xx Legacy drivers. Please<br>NVRM: visit <a href="http://www.nvidia.com/object/unix.html">http://www.nvidia.com/object/unix.html</a> for more<br>NVRM: information. The 610.43.02 NVIDIA driver will ignore<br>NVRM: this GPU. Continuing probe…<br>NVRM: No NVIDIA GPU found.</blockquote><ul><li>torch dependency hell. <a href="https://medium.com/u/ca8da3166758">Sayed Amman Akhtar</a>’s <a href="https://medium.com/@ammanakhtar8/how-i-finally-got-musetalk-running-for-real-time-talking-face-after-days-of-dependency-hell-a95d009abf46">piece on MuseTalker </a>should be treated largely as canon in my opinion. I have a couple of tiny nits (he’s running as a privileged user and not leveraging sudo) which I’ll get into someday soon. His order of operations is accurate and it turns out necessary. His instructions allow torch to figure some things out for itself to get the right CUDA versions for your graphics driver. <strong>This was instrumental in getting my local setup operational.</strong></li><li>sort of a general bucket of issues I’m going to call “upgrade your hardware”. This includes (in my case) a CPU that lacked the necessary instruction set and an (original) GPU that was a waste of time and money. Do your research on the hardware you are planning to use. There are deals to be had out there to get a setup functioning.</li></ul><p>These observations are not to discourage anyone: quite the opposite in fact!. If you’re well versed in system and OS-level engineering that is. It may be possible without understanding these things, but it certainly helped me in this situation.</p><p>Let’s go down the list of issues, and how strong human (not AI) engineering skills were needed to identify and correct. Keep in mind, these are the problems that exist <em>before</em> AI is even operational, and other AI (Gemini, ChatGPT, and Claude in this case) had <strong>no answers whatsoever</strong> that helped get AI functioning locally.</p><p><strong>issue 1: out of memory issues</strong><br>OK, beyond normal powers of perception, this one isn’t a heavy lift to identify. The messages may be a bit buried, but they are easy enough to spot.</p><p><strong>issue 2: CUDA driver issues</strong><br>This was a steep curve. In fact, it’s a few different things nested together. At first, I approached this with abandon. I installed the video card, downloaded NVIDIA drivers and by all appearances, had a working CUDA configuration right off the jump! Sweet!</p><p>Let’s not go high-fiving each other just yet.</p><p>When I ran out of RAM initially, I did what you’re supposed to do: I bartered and traded for some old skool 8GB RAM modules that my older skool motherboard would support. I maxed out the motherboard and sat frustrated for 2 days before I saw the issue. It wasn’t spelled out; just an error log that said something like this</p><blockquote>[ERROR] unable to allocate 823M of 621M available ( 621M of 4096M available)</blockquote><p>I had WAY more than 4GB, but my GPU had, you guessed it, 4GB.</p><p>So a quick search of cheap GPUs and I identified a Quattro with 8GB for a reasonable price on eBay. I checked and the card in question <strong>would</strong> cover my GPU needs.</p><p>(As an aside, while installing this video card, I discovered it required separate power cable. This was a new experience, as was discovering my power supply supported a cable specifically for this purpose. No new power supply was a win!).</p><p>Getting the CUDA drivers identified proved trickier than expected. I (incorrectly) assumed that the latest drivers would be the right ones, however, there seem to be trillions* of places where the CUDA and torch support have to align <strong>PERFECTLY</strong> for things to work. And the actual link provided in the error message pointed me to a generic landing page. The URL provided was a shockingly generic URL, not a shortened URL perma-linked to a specific document. Feels like a miss to me.</p><p>Ultimately, it was the error message above, first observed on the console remotely located in my attic, that held the answer. I needed the 580.xx legacy driver, but that driver <strong><em>is not linked</em></strong> on the nVidia site. Instead, I searched Google and found the correct URL that way. For the record, there is a perma-link to the right document, it is just not on the generic drivers page.</p><p>After completely purging all of the nVidia drivers from my system — several times — and installing the 580.xx legacy drivers I’d downloaded, there was progress.</p><p>This command seems to be the key to everything:</p><pre>nvidia-smi</pre><p>You’ll want your output not to contain any of the different error messages I received, but instead look like this:</p><pre>+-----------------------------------------------------------------------------------------+<br>| NVIDIA-SMI 580.173.02             Driver Version: 580.173.02     CUDA Version: 13.0     |<br>+-----------------------------------------+------------------------+----------------------+<br>| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |<br>| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |<br>|                                         |                        |               MIG M. |<br>|=========================================+========================+======================|<br>|   0  Quadro M4000                   Off |   00000000:01:00.0  On |                  N/A |<br>| 46%   30C    P8             12W /  120W |      10MiB /   8192MiB |      0%      Default |<br>|                                         |                        |                  N/A |<br>+-----------------------------------------+------------------------+----------------------+<br><br>+-----------------------------------------------------------------------------------------+<br>| Processes:                                                                              |<br>|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |<br>|        ID   ID                                                               Usage      |<br>|=========================================================================================|<br>|  No running processes found                                                             |<br>+-----------------------------------------------------------------------------------------+</pre><p>This doesn’t mean everything is working, it just means CUDA is OK. You’ll do well to note that the driver version AND the CUDA version. You’ll need to understand both at different points, especially when we get to the torch dependency hell conversation.</p><p>There are a few more gears to this set of issues, including a bit of an unwanted Easter Egg buried in all of the dependency chasing.</p><p>We need to talk a bit about the difference between server-grade and desktop-grade solutions.</p><p>Not in a judgy way at all, but <a href="https://medium.com/u/ca8da3166758">Sayed Amman Akhtar</a> was using the desktop version of Ubuntu. I chose the server version for a number of reasons. I want to talk about the security in all of this at some point, but the biggest reason is I am lazy. My AI workstation is not located conveniently, and there are more security issues in general with “desktop” OSes (compared to server OSes). It is easier to connect using ssh than it would be to open X windows remotely. And I’m lazy and that involves some work.</p><p>A consequence of my laziness is there is no graphical user interface on my system. And this causes a bit of a hidden issue: your system needs to support graphical operation natively OR you really need to know how to configure things.</p><p>The easy answer is to choose graphical operation from the start.</p><p>In summary, I ended up installing all of the headless support needed for the various graphics generation capabilities I’d need.</p><p><strong>issue 3: missing and terse error logging<br></strong>The stack required to support all of the operations involved in an AI operation like turning an image into a video is massive. It was written by people across the globe, speaking many different native languages. While there are “standards” out there for logging, they are really only applied in places where a big company is responsible for the entire solution. If you are cobbling together a solution, sometimes logs are insufficient, inconsistent, missing, or impossible to understand.</p><p>The more engineering experience I got, the more the application of good logging became. And even if the logging were aces, could AI really effectively troubleshoot itself? I mean, could this article and others like it serve to inform future AI on how to fix itself? I’m skeptical.</p><p>The more I delve deeper and deeper into the belly of AI, the more convinced I am engineering will be alive and well for the foreseeable future.</p><p>I think it’s a widely held belief that any time a company says layoffs are the result of AI, it doesn’t pass the sniff test. This could not be more true than for engineers, in my opinion. Even if the stacks for Anthropic (Claude), Google (Gemini), and OpenAI (ChatGPT) are immaculately groomed to execute flawlessly in a pipelined manner, there still need to be engineers.</p><p>That stack <strong>will</strong> hit a limit. If it is a condition that has never been experienced <strong><em>ever </em></strong>by <strong><em>anyone, </em></strong>can AI really determine that limit, given it may be having issues starting in the first place.</p><p>Am I saying these things are impossible? Not exactly, but they are not presently possible.</p><p>Even quantum computing, the real oncoming threat to our digital security if you believe the experts, even quantum does not have free will. It has crazy patterns in the data, but quantum hasn’t lost the plot yet as far as I can find.</p><p><strong>issue 4: torch dependency hell<br></strong>There is no kind way to put this. It is a massive snarl. You end up with modules that require different versions of the same software. You end up with issues with chumpy, or flash_attn, or nvcc, or all of those. The key is letting what torch and CUDA require be your guide. If some weird offshoot library requires support for a library and that conflicts with what torch requires, let torch win.</p><p>This was more of an observed behavior than one that can be documented. During the various installation attempts I observed these differences and began to compensate once the criticality of torch fully landed with me.</p><p><strong>issue 5: general issues<br></strong>Odds are, few will be attempting to run a Nehalem-era CPU (circa 2009) that is missing support for AVX and AVX2 instructions. I’ll save the space on documenting everything attempted here unless there are specific questions.</p><p>To wrap up, check out the fruits of my efforts on <a href="https://youtube.com/@nobodycaresdude">my YouTube channel</a> where I go into this setup and how AI is helping me overcome some issues.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f6270d5a1903" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[allow me to introduce my brain]]></title>
            <link>https://medium.com/@jed.deaver/allow-me-to-introduce-my-brain-51ccd278097d?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/51ccd278097d</guid>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[anxiety]]></category>
            <category><![CDATA[autism]]></category>
            <category><![CDATA[depression]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Mon, 06 Jul 2026 20:22:12 GMT</pubDate>
            <atom:updated>2026-07-06T20:37:42.045Z</atom:updated>
            <content:encoded><![CDATA[<p>a true account of modern mental health struggles</p><p>I write this from a decent place mentally. <strong>I am not</strong> thinking of harming myself or anyone else. I’m frustrated and I feel helpless at times, but I am ok.</p><figure><img alt="a cartoon man lies on the sofa of a therapist" src="https://cdn-images-1.medium.com/max/1024/1*Kg1n_aY5HIZkN1xPJrsiQg.png" /></figure><p>I can’t imagine going through my current situation in any other time in workforce history. Physically, I am fine and capable of performing my work, but mentally, not at all.</p><p>I’m working on things and I’ve made tons of progress in a short time. I am completely burned out — there’s no denying that. Emotions are on a hair trigger. It feels like no one is on my side. There is no one to turn to. There are plenty of people <strong><em>available</em></strong> to help, but no one <strong><em>empowered</em></strong> to do much.</p><p>What do I mean by that?</p><p>In the US, insurance is tied to employment. This starts a pretty aggressive cycle. The insurance company’s role is less about getting people “back to healthy” so much as it is about returning them to being a functional worker. This makes for some interesting outcomes.</p><p>When we’re talking about physical health, the variables are easier to understand. Doctor’s can measure efficacy of treatment against a number of factors, and in many cases, there are scans which are performed to ensure treatment is proceeding as expected.</p><p>There few such scans for those struggling with mental health conditions like depression, anxiety, autism, behavioral disorders, and attention-deficit disorder. These conditions generally require more diagnostic measures to identify and acknowledge.</p><p>Therein lies my dilemma. Physically I am fine, but mentally, something is not right. I have theories and I am working with a handful of mental health experts to sort thru this all, but for now, there are no real answers.</p><p>In my case, time would help the most. I need to recover from some very traumatic stuff. Because it’s mental health though and because this trauma chose to surface out of cycle with the events behind the trauma, it becomes tricky to get others to understand the underlying problem.</p><p>Hi, I have had depression in some form for over 50 years, so I feel qualified to intervene here.</p><p>The absolute worst thing anyone can say to someone struggling with mental health challenges is to “get over it”.</p><p>Gee, thanks, wish I would have thought of that! I wish I had a choice in the way my brain responds to these situations, but I’m a passenger same as you. Don’t get me wrong — I do a lot of backseat driving, offering my brain suggestions on how to manage a particular moment. In those moments I even have specific conversations with “Mr. Brain” as I call him. Mr Brain and I don’t always get along.</p><p>Mr Brain knows we can do better. Mr Brain is that voice in my head that is super snarky and on point when he knows he needs to be.</p><p>Mr Brain is also a giant jerk. He is critical, hell, judgmental, about many things. Mr Brain is a bully. His words are unkind, unprofessional, and outright hurtful. Mr Brain is lazy. He knows he and I are capable of solving many problems, but he doesn’t like every problem that comes along.</p><p>It’s not like Mr Brain has an actual voice in my head. He’s definitely the source of most of the negative thoughts. Mr Brain doesn’t care for me and he won’t say why. And that disassociated voice in my head is tricky to reconcile.</p><p>Why say all of this to the internet? Aren’t you opening yourself to criticism or other forms of negative response?</p><p>Perhaps, but these are things that we need to be more comfortable talking about openly.</p><p>I know in my generation, boys were taught to swallow our emotions, to stuff it all down. That’s what I did what a long list of things: new schools every other year, no stable home life, no friends, effects of being bullied for being the new kid, who was also overweight and ridiculously quiet.</p><p>Throughout that time, I always had my mom there to help make sense of it all. Even when I reluctantly became the adult on a few things, mom’s wisdom was there when I needed it. Until 15 months ago that is, when she passed away. There is not a day that goes by I don’t miss her terribly, and therein lies my current dilemma: she’s not there any more, and I can’t stuff it down any further.</p><p>A few weeks ago, I broke. I wish I could explain it. The best example I have come up with is to image a glass of water. The glass is on a level table. It is impossible to add even a small drop of water to the glass or it would overrun. That’s where my brain was.</p><p>Today I’d put my capacity closer to 80% — it’s not ideal but I’m not close to my breaking point either. I’d like to be around 50%, but progress is progress.</p><p>I’m thankful we live in a time where we can talk about these things together, instead of continuing to stuff them down like our predecessors.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=51ccd278097d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[nobody cares about your code review]]></title>
            <link>https://medium.com/@jed.deaver/nobody-cares-about-your-code-review-7e0f98c9677e?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/7e0f98c9677e</guid>
            <category><![CDATA[mobile-app-development]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[kroger]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Fri, 03 Jul 2026 03:53:35 GMT</pubDate>
            <atom:updated>2026-07-03T03:53:35.545Z</atom:updated>
            <content:encoded><![CDATA[<p>I just wanted to buy some groceries.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CR4w5euv4xMhiNprVB0Bcg.png" /><figcaption>I wish my setup were this clutter-free</figcaption></figure><p>Like every other Sunday, I made an early morning run to the grocery store this past Sunday. As part of my alone time, I like going to the store unfettered with anything else. It’s a bit of therapy for me.</p><p>Typically I go to the same store every week. I’m familiar with this chain’s mobile app and store layout, but this week, my wife asked for a specific Kroger-branded product.</p><p>Normally I wouldn’t pollute the internet with such a trivial task, but this is different. It’s in my wheelhouse.</p><p>Because nobody cares what I think about their app.</p><p>grocery store edition</p><p>Let’s talk about the Kroger mobile and web experience for a minute. Consider this very unsolicited feedback from someone who fancies themselves somewhat of an expert on mobile apps.</p><p>First, there is building the list itself. I had a whole thing written, but in researching my feedback, I found I’d overlooked a step. While that’s bad on me for missing the mechanism that determines if you are shopping in store, picking up, or having your items delivered, I’d ask if it’s bad on Kroger for hiding it in this way, and making it available ONLY when you select the location? All I’m after is a pretty low resolution experience: show me a list, sorted by rough store location, executed with a reasonable amount of competence. This isn’t hard.</p><p>I have a theory: the difference between privately held and publicly held companies is buried in there somewhere.</p><p>Kroger’s app feels like it was heavily informed by opinions focused on delivering business to business opportunities on the back end, versus the private company app experiences from places like Publix, Wegmans, even Chik-fil-A. If you haven’t tried it, Chik-fil-A-s mobile and in car experience is really good. The focus is very customer oriented.</p><p>I’m not saying Kroger needs to invest in an in car app, not yet anyways. 😀</p><p>My general expectation is that I can build and shop items on my list quickly. As far as it exists on my device, that list should only serve to manage things I wish to purchase, not intrinsically track every action I take with that item.</p><p>I’m not knocking any of the work done on this app or its supporting platforms. My point is that the constraints are different in corporate environments, versus private company environments. In a corporate environment, the Kroger app could be seen as a cost center — something that is not directly revenue generating. Investments in making the app more intuitive are slower in coming because there is less business demand. It might even be seen as “wasteful” to invest in this way. The team had bad requirements, that’s all. At least that’s been my experience over the past 30 years in corporate America.</p><p>I know what the constraints put on development teams are and I’ve got ideas on how to clear them.</p><p>Private companies realize the mobile experience is all about adaptability. It’s part of the funnel into the store.</p><p>Let’s get back to the app.</p><p>Arriving at the store, I was greeted by a huge produce department. I bring up my kroger app and at first everything is sorted by “Other”. After messing around with a few settings in the app, I returned to my list, refreshed it, and everything sorted by Aisle. Strange, but seeing the list start on Aisle 6 instead of produce, deli, or bakery, all of which I pass on my way to aisle 6 was stranger still. Instead of sorting on the name: aisle comes before bakery, always, add an index to the way the store is laid out. Here’s a quick model of some pseudo JSON to support this</p><pre>{<br>  &quot;list&quot;:{<br>    &quot;item&quot;:[{<br>      &quot;department&quot;:&quot;produce&quot;,<br>      &quot;aisle&quot;: &quot;produce&quot;, ← - instead of sorting on this<br>      &quot;name&quot;:&quot;cucumber&quot;,<br>      &quot;price&quot;:.95<br>    },{<br>      &quot;department&quot;:&quot;grocery&quot;, ← - factor this in instead<br>      &quot;aisle&quot;:&quot;aisle 6&quot;,<br>      &quot;name&quot;:&quot;bread&quot;,<br>      &quot;price&quot;:3.99<br>    },<br>  ],<br>}} </pre><p>Within the app itself, consider a change to accommodate the general layout of every Kroger store. Every location I’ve been to, you walk into produce, there’s the bakery and the florist and the deli and meat or seafood before you get to aisle 1. I’m sure there are variants to that based on location, so make that layout part of the natural flow of the app. Like this</p><pre>{<br>  &quot;store&quot;:{<br>  &quot;info&quot;:[{<br>    &quot;store-number&quot;:&quot;2100&quot;,<br>    &quot;store-name&quot;: &quot;Main St&quot;,<br>    &quot;layout&quot;: [ &quot;produce&quot;,&quot;deli&quot;,&quot;floral&quot;,&quot;bakery&quot;,&quot;cafe&quot;,&quot;cheese&quot;,&quot;seafood&quot;,&quot;grocery&quot; ],<br>  } , {<br>    &quot;store-number&quot;:&quot;2120&quot;,<br>    &quot;store-name&quot;: &quot;Broad St&quot;,<br>    &quot;layout&quot;: [ &quot;produce&quot;,&quot;floral&quot;,&quot;deli&quot;,&quot;cheese&quot;,&quot;seafood&quot;,&quot;grocery&quot;,&quot;bakery&quot; ],<br>  },<br>],<br>}}</pre><p>See the difference? It’s focused on the customer. If I shop the Main street store, the app shows me produce first, etc, before I get to the general “grocery” section, but shopping the Broad St store the bakery is in the corner past the frozen foods, so bakery makes more sense after for that location.</p><p>Speaking of shopping different stores, let’s talk a little about session management.</p><p>To genericize my content, I selected a Kroger location close to corporate headquarters in Cincinnati instead of something close to my actual house. And now the mobile app seems to have an unbreakable affinity to the Ohio store. I’ll get that sorted, but if you drop a comment with your thoughts that’d be cool too.</p><p>It’d be one thing if all stores were laid out like the app, but that’s not practical.</p><p>In this week’s example, it would have been very helpful to have general sorting instead of alphabetical sorting . Every time I checked an item off my list, the app made API calls thru whatever network I was connected to. My guess is this is to perform some form of metrics. That stuff should be invisible to the customer, but it’s actually worse than that. At least from my perspective. The other impact is that my list would return unsorted, forcing me to swipe dxown to restore the sorted order.</p><p>If I remove that Dinty Moore Beef Stew, do we owe it to Hormel to indicate that preference to optimize pricing with them?</p><p>I didn’t buy any beef stew for what it’s worth.</p><p>it was just an example</p><p>If I had to guess, apps from privately held companies achieve the same functions but they don’t make it such a blatant in your face action to the customer.</p><p>And when I remove an item, I should not have to refresh my app to get the list sorted again. All of these seem like reasonable prompts to a decent AI coding model, with appropriate team reviews before promotion to production.</p><p>I had 5 items in my list in the produce section. After checking each item off my list, the list in the mobile app would completely resort under “other”. It takes a 2nd refresh to get the aisles back.</p><p>Let me know if you’d like to hear more.</p><p>Using the Kroger app, I got the sense I was constantly being advertised to. I’m there to shop. There are ways to make it more interactive.</p><p>I finally figured out a ham-fisted way of getting my list cleared and on thru to checkout, then cleared everything out of the unsorted list. That proved faster. I’m not sure what that category API was doing that ran long. It’s hard to imagine what the hangup here is, nor why an API call is even needed?</p><p>I want to point out I’ve looked at zero lines of the code for the Kroger mobile app, but I’ve worked on mobile apps enough to tell when these mistakes are present. I also could be completely off base. Either way I’d love to start the conversation. These little things make such a huge difference, and they are so easy in terms of effort.</p><p>I get the profit optimization elements but Kroger just feels like they gave up on customers altogether. And this stuff feels doable.</p><p>There is no point other than to maximize this aspect of my grocery shopping experience. It’d be cool to affect some positive change somehow</p><p>Take care</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7e0f98c9677e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[watch me get owned by AI]]></title>
            <link>https://medium.com/@jed.deaver/watch-me-get-owned-by-ai-10bd1b8da954?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/10bd1b8da954</guid>
            <category><![CDATA[long-cat]]></category>
            <category><![CDATA[localai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[musetalker]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Wed, 01 Jul 2026 17:20:14 GMT</pubDate>
            <atom:updated>2026-07-01T17:20:14.740Z</atom:updated>
            <content:encoded><![CDATA[<p>In my first article, I talked about my use case. Now let’s get down to solving for it, and why it’s so challenging.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UnJ75qYrggBwgpWd44DC9w.png" /><figcaption>I wish I had this setup, though 4 soldering irons seems unnecessary</figcaption></figure><p>Some time back, I set about building a “local AI” to help with some content production for YouTube. Drastically oversimplified, I wanted to animate an avatar somehow.</p><p>I was looking at things that did what I <strong><em>*thought*</em></strong> I wanted, only to constantly find myself disappointed.</p><p>I even tried using tools like Adobe AfterEffects, Moho and some online tools. Honestly, it’s a lot of work. I want to focus on more important stuff. Maybe when I can hire a graphic artist and we can collaborate some, if that day ever arrives. I’ve got ideas my brain can’t articulate well. Not yet anyways.</p><p>Finding the right combination of AI capabilities checks a big box for me there. And I’ve identified models that will run locally without producing truly bananas results.</p><p>Let’s talk about a few of the models I’ve looked at, and some feedback based on experience.</p><p>Early on I looked at SadTalker. Developed by Chinese researchers, it simply overlays a moving mouth on a static image. That’s a very bare bones solution to my problem. Feedback on SadTalker indicated challenges around lips trained solely on Chinese language, meaning the mouth motions are more closely aligned with Chinese language content. Think of the old martial arts movies from the 70’s. SadTalker also had a strong preference for an actual photographic image. The model uses landmarks like nose and eye positions for performing replacements, and the AI slop drawings I fed it early on missed that mark.</p><p>Models from Kling, Qwen, LTX, and Flux all came with challenges. Kling makes faces appear somewhat dark and evil looking. LTX is decent when you get the prompting right, but I haven’t got the hardware combination right yet. Qwen made some interesting choices regarding hand movements, and Flux, well, Flux needs work.</p><p>LongCat Avatar is looking promising, once I get my CUDAs and my memory issues ironed out. More to come there when I document my final setup.</p><p>Also looking promising is MuseTalk. I created a new 8 bit avatar, based on the memoji on my phone, which is supposedly based on my face.</p><p>Here’s the thing: this is all stuff specific to my individual use case. Sure, there are commercial AI solutions that are free or close to free, but I’m giving my data to others as part of the bargain. I don’t want the production stuff out there until it’s done, so my preference is to do this locally.</p><p>As I’ve shared these findings with others, reactions have been, well, mixed.</p><p>There are those who feel AI is blatant theft. I don’t disagree, but this application of the word theft isn’t really meaningful to me. I could find a graphic artist to work with, and get something that fits the traditional definition of art. But to what end: I’m explaining AI after all. Aside from feeling kinda besides the point, it’s a lot of work, either for myself or someone else. And if it is someone helping me out, now there are 2 schedules to work around. I like being able to have an idea, throw my thoughts together, record the audio, then get things ready to go. The longer between the audio take, the more time to second guess literally everything.</p><p>Instead of avoiding AI, I’ve been trying to knock the less palatable edges off of it. The slop is truly awful and looks very little like me. I’m less put off by the resemblence issue. It’s not like I’m insanely photogenic in real life. If anything, I want something that grabs attention, even if it isn’t positive attention all the time. It’s part of the conversation.</p><p>Next up, let’s talk about what it takes to run these models. Once my solution is operational, I’ll have a full rundown, with a github repo and everything.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=10bd1b8da954" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[nobody cares you’re using AI]]></title>
            <link>https://medium.com/@jed.deaver/nobody-cares-youre-using-ai-2f577dd4faaf?source=rss-e8267c311c67------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f577dd4faaf</guid>
            <category><![CDATA[local-ai-solutions]]></category>
            <category><![CDATA[musetalker]]></category>
            <category><![CDATA[ubuntu]]></category>
            <dc:creator><![CDATA[nobody cares dude]]></dc:creator>
            <pubDate>Tue, 30 Jun 2026 16:34:01 GMT</pubDate>
            <atom:updated>2026-06-30T16:53:01.752Z</atom:updated>
            <content:encoded><![CDATA[<p>Apologies for the trolly title, I’m trying to build a small brand around the notion that I’m just another voice in the diaspora. I’ve got opinions like everyone else, I’m just trying to find ways to articulate them to broader audiences.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UnJ75qYrggBwgpWd44DC9w.png" /></figure><p>I’m trying to share some of my critical decisioning skills with the rest of the internet, like anyone needs that. I don’t want to influence anyone to do anything except find happiness. Maybe it’s a blog, or vlog, or whatever. Think of this as building a podcast inside out. I <strong><em>do</em></strong> want to build an audience somehow. I do want to have people on to talk about whatever.</p><p>I could talk about food, or cars, or tech. Probably tech more than the other stuff to be fair. I’ve struggled figuring the format out for myself, but I’m very close to an AI avatar I can use to help my story telling. Check it out if you’re interested: <a href="https://youtube.com/@nobodycaresdude">https://youtube.com/@nobodycaresdude</a>.</p><p>I found a series of articles by <a href="https://medium.com/u/ca8da3166758">Sayed Amman Akhtar</a> to be helpful in a number of ways. First, they forced me to re-examine a few things.</p><p>AI comes at you like a big blur: it’s a lot of terms and things that don’t make any sense. The proper context is missing.</p><p>AI runs on computers. I can run different AI model capabilities on everything from the computer I am typing this article on, to a decade old PC I built over 10 years ago, to free services in the cloud, to paid services. And all of those paths crisscross one another unabashedly.</p><p>In my AI journey I’ve run into new terms like tensors and weights and ComfyUI and MLX and Kling and LTX and HuggingFace and Vast.ai and Colab. Yes, that is just a lot of words, and for too long a period they all smushed together inside my head. I had to break things down and find a specific task to begin to understand any of this. I chose a hard task because the easy tasks like getting Ollama to run a generative prompt locally is mostly plug and play, but that doesn’t mean Ollama is well suited for all AI tasks.</p><p>Sayed documents so many of the same struggles I’ve encountered, and that others would likely encounter as well. His take is spot on — I didn’t have the mmcv issues (mine worked ok), I did have issues with another component: something called bazel. bazel is very finicky.</p><p>Through all of this, I’ve questioned every part of my setup: I’m 3 video cards deep into this, I’ve tried Qwen, and Kling, and LongCat, and Flux models. I’ve tried Lip2Wav. I find things whose outputs I like only to find some combination of subscription platform, or API usage key, all with their associated fees. I’ve got a good handful of subscriptions I’ll be cancelling when I found out after subscribing I was dealing with a 3rd party company instead of solving the problem myself.</p><p>For now, the journey continues. I’ll try to break this down to keep it readable, and maybe I’ll package the whole thing up to sell when I’m done 😀</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f577dd4faaf" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>