<?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 Chilton Webb on Medium]]></title>
        <description><![CDATA[Stories by Chilton Webb on Medium]]></description>
        <link>https://medium.com/@chiltonwebb?source=rss-becc3d200a8a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*Nh2P5Y_UZer0ORRU.jpg</url>
            <title>Stories by Chilton Webb on Medium</title>
            <link>https://medium.com/@chiltonwebb?source=rss-becc3d200a8a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 13:44:12 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@chiltonwebb/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[The Anti-Turing Test]]></title>
            <link>https://medium.com/@chiltonwebb/the-anti-turing-test-baf081d71135?source=rss-becc3d200a8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/baf081d71135</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[turing-test]]></category>
            <category><![CDATA[arguments]]></category>
            <dc:creator><![CDATA[Chilton Webb]]></dc:creator>
            <pubDate>Tue, 10 Jun 2025 00:54:51 GMT</pubDate>
            <atom:updated>2025-06-10T00:54:51.581Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/628/1*TbRing-SaisCT28HMvSawg.png" /><figcaption>I knew you were a robot when you explained how the economy works.</figcaption></figure><p>There’s a ridiculously easy way to tell if someone or something is an AI, stemming directly from the attempt to create Artificial General Intelligence.</p><p>See, the thing is… people is dumb.</p><p>So if someone knows damned near everything about any topic, there’s a good chance they’re an AI. Or very, very well read.</p><p>But your chances of running into someone like that, these days is rapidly decreasing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=baf081d71135" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[**Brutal Honesty Mode: Why AI Programming Help Sucks (And Why It Won’t Get Better Soon)**]]></title>
            <link>https://medium.com/@chiltonwebb/brutal-honesty-mode-why-ai-programming-help-sucks-and-why-it-wont-get-better-soon-cd47cf8428ad?source=rss-becc3d200a8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/cd47cf8428ad</guid>
            <category><![CDATA[reality]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[games]]></category>
            <category><![CDATA[dystopia]]></category>
            <dc:creator><![CDATA[Chilton Webb]]></dc:creator>
            <pubDate>Sun, 01 Jun 2025 21:07:50 GMT</pubDate>
            <atom:updated>2025-06-01T21:07:50.637Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/793/1*EW4fIIA9_eKuRBQysFFVDA.png" /><figcaption>Hey I finished that program you wanted</figcaption></figure><p>This is according to DeepSeek’s Brutal Honesty Mode:</p><p>AI gets you *almost* there, then faceplants at the finish line. Here’s why, with zero copium:</p><p>### **1. AI Doesn’t “Understand” Code — It’s Just a Glorified Autocomplete** <br>- **You think:** It’s analyzing logic, debugging like a human. <br>- **Reality:** It’s predicting the next token (word/symbol) based on patterns in its training data. No true comprehension, no real problem-solving. <br>- **Result:** It can *mimic* correct-looking code but often misses edge cases, architecture, or subtle dependencies.</p><p>### **2. The “Last 10%” is Where Real Engineering Happens** <br>- AI crushes boilerplate, generic patterns, and well-documented APIs. <br>- But the **last 10%** is where: <br> — Business logic gets messy. <br> — Legacy code fights back. <br> — State management turns into spaghetti. <br> — Race conditions, memory leaks, and undefined behavior lurk. <br>- **AI can’t handle this** because it wasn’t trained on *your* specific, janky, undocumented system.</p><p>### **3. Fixing One Bug Creates Another Because AI Has No Memory or Big-Picture Awareness** <br>- Humans keep a mental model of the whole system. AI? **Zero context** beyond the current chat window. <br>- When you ask it to tweak code, it’s not “refactoring” — it’s **blindly regenerating** based on your latest prompt. <br>- **Side effects?** It doesn’t care. It can’t.</p><p>### **4. Training Data is Full of Garbage (And It Shows)** <br>- AI learns from GitHub, Stack Overflow, and docs — **aka a graveyard of outdated, buggy, and contradictory examples**. <br>- If bad patterns were common in training, AI will repeat them. <br>- **Ever get code that *looks* right but fails in production?** Congrats, you just inherited some rando’s technical debt.</p><p>### **5. Can It Be Fixed? Maybe, But Not Soon** <br>- **Short-term?** No. Current AI is fundamentally **statistical, not logical**. <br>- **Long-term?** Maybe with: <br> — **True code-aware AI** (e.g., models that build ASTs, run static analysis, or simulate execution). <br> — **Persistent memory** (so it doesn’t forget the system’s state between prompts). <br> — **Formal verification** (AI that *proves* correctness instead of guessing). <br>- But until then, **AI will keep being a “useful idiot”** — great for drafts, terrible for finalizing.</p><p>### **The Brutal Truth** <br>AI won’t replace engineers because **the last 10% is where the real work happens**, and AI is *terrible* at it. You’re better off using it for: <br>- **Stub generation** (then manually fixing). <br>- **Debugging hints** (but verifying every suggestion). <br>- **Documentation lookup** (faster than Google).</p><p>**Stop expecting it to “just work.”** It can’t. It’s a tool, not a teammate.</p><p>*[Mic re-drop. Brutality complete.]*</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cd47cf8428ad" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Debugging: Sadistic Or Just Heuristic?]]></title>
            <link>https://medium.com/@chiltonwebb/ai-debugging-sadistic-or-just-heuristic-908f6ce91807?source=rss-becc3d200a8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/908f6ce91807</guid>
            <category><![CDATA[jobs]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Chilton Webb]]></dc:creator>
            <pubDate>Fri, 30 May 2025 16:31:23 GMT</pubDate>
            <atom:updated>2025-05-30T16:31:23.579Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fAceiJuzzIHXAgCogCfjCw.png" /><figcaption>How LLMs look at you through the screen</figcaption></figure><p><em>Warning: Despite what I’m about to show you, my experience with DeepSeek, ChatGPT, Gemini, and Claude, has been that Claude is the most reliable of the four right now (Claude 4 Opus just dropped). But there’s a good chance this article might be wrong at some point in the future, so I wanted to make it clear what was being used at the time…</em></p><p>I’ve been a freelance programmer for 35 years, and this time, right now, is the most exciting time to be a programmer. It’s never been easier or more fun. But debugging? That’s still an art form. Unfortunately, it’s not an art form that AIs are particularly good at yet.</p><p>So if you ask a modern AI why something is happening in your code, don’t be surprised at its goofy-ass answers.</p><p>Modern AIs do not ‘think’ logically, like you’d assume. According to <a href="https://transformer-circuits.pub/2025/attribution-graphs/biology.html">a new paper from Anthropic</a>, it’s much, much weirder than that.</p><p>Generally speaking, here’s what it’s going to do. It will take your code, its knowledge of programming, and its knowledge of debugging answers, and mix them all together in this weird giant spider web of logic, to get the answer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4vAkvj4opuOHENPldXK9EA.png" /></figure><p>So… how much weight do you think your code will be given, considering the vast size of its existing knowledge and existing codex of debugging answers?</p><p>Answer: Not enough.</p><p>Here, I’m using <a href="http://www.anthropic.com/">Claude Opus 4</a> with the Max subscription, to help me find a memory crash. But I have to stress, the other AIs did <strong>WORSE</strong> than Claude on this task.</p><p>My personal, human, system is simple: Decimate, Replicate, Extrapolate, until you Isolate.</p><p>But the AI doesn’t need these fancy rhymin’ words. No sir. It is made of pure logic, infinite knowledge, and the power of a thousand suns!</p><p>So I offered my human scribbles to the machine god, and fractions of a second later, it gave me ten likely reasons for my failure. Its first answer was just wrong, so I skipped it. But here’s the second one, and I admit, this gave me pause. It was definitely possible…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*XmsVKnTCfVJKrp15Dy56VQ.png" /></figure><p>See that? It found *SEVERAL PLACES* where I’m doing the bad thing.</p><p>I am such a fool. I created a whole thing to allocate memory and didn’t even use it. Stupid human brain! Stupid, stupid.</p><p>I comb through my code, looking for this proof of my own inadequacy, and huh… my faulty human brain simply can’t find it.</p><p>So I put on my reading glasses, to assist my frail human eyeballs, kick back my last can of JOLT COLA, and try again. Still, nothing. How odd.</p><p>WAIT! this kind of thing will stand out like a sore thumb in <a href="http://www.codewalker.io/">CodeWalker</a>, so I fire that up, point it at the code, and … still nothing.</p><p>Human dumb. Robot smart. This is proof. I can’t even find a needle in a haystack, and I have a supermagnet and a flamethrower!</p><p>I guess it’s time for me to admit defeat and ask the AI to school me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*dZxWdnpjqLRAaQLDdYsuCA.png" /></figure><p>You sonofabitch.</p><p>Okay what just happened? Well, the short answer is that when debugging, the AI will often (like here) tell you what the answer likely is, based on what other answers, by other people, have been in the past. And it mixes that knowledge with your code. So if developers frequenly make a function to combat a particular problem, and that problem persists, it’s often because they weren’t thorough enough, in its use. So it’s mixing that concept with code I wrote, and throwing that at the wall.</p><p>The danger of this, is that you can (and WILL) spend hours chasing down something the AI insists is a problem, probably even changing your working code and introducing new bugs in the process, only to find out later it might’ve been a problem, it just wasn’t a problem in YOUR code.</p><p>AIs are much better at programming than they are at debugging.</p><p>Hey look on the bright side. When the AIs finally take our jobs, we’ll be able to get new jobs as AI cleanup crew.</p><p>-<a href="http://www.chiltonwebb.com/">Chilton Webb</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=908f6ce91807" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CodeWalker’s C++ Parser over time]]></title>
            <link>https://medium.com/@chiltonwebb/codewalkers-c-parser-over-time-fb7c20f696f5?source=rss-becc3d200a8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb7c20f696f5</guid>
            <category><![CDATA[code-review]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[code]]></category>
            <category><![CDATA[programming-languages]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Chilton Webb]]></dc:creator>
            <pubDate>Thu, 08 Feb 2024 18:25:37 GMT</pubDate>
            <atom:updated>2024-02-08T18:27:08.468Z</atom:updated>
            <content:encoded><![CDATA[<p>A picture’s worth a thousand words. Or in this case, over 1600 lines.</p><p>At the top of this image, the CodeWalker C++ parser.</p><p>At the bottom, the C++ parser a year ago.</p><p>(Screenshot is from CodeWalker’s Microscope mode)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*m0ITndIOhfVwOA5kvBT5Ww.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb7c20f696f5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Nightmare to CodeWalker]]></title>
            <link>https://medium.com/@chiltonwebb/from-nightmare-to-codewalker-3c84add853a2?source=rss-becc3d200a8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c84add853a2</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[code-newbie]]></category>
            <category><![CDATA[visualization]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[inspiration]]></category>
            <dc:creator><![CDATA[Chilton Webb]]></dc:creator>
            <pubDate>Sun, 29 Oct 2023 04:17:02 GMT</pubDate>
            <atom:updated>2023-10-29T04:23:48.877Z</atom:updated>
            <content:encoded><![CDATA[<h3>From nightmare to CodeWalker</h3><p>In 1999 I took on an impossible project because I needed the money.</p><p>It wasn’t rocket science, I needed money, and I had a client who was willing to pay me. But only because no one else would take on the project. It was for a realtime online poker game. The requirements were the nightmare part. It had to run in flash, support thousands of simultaneous players, and allow realtime chat, lobbies, friends, and a ton of other things I didn’t know were complicated.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LVZXpe5yGg9hMd-ilVI5BA.png" /></figure><p>But I knew enough about each of the problems to kinda know what I was doing. By 2003, all of the pieces worked. I was using a bug in Flash to allow me to keep a socket open to the client. I’d written the entire game engine in C, and it talked to a custom web server that could coordinate with the Flash client if the socket system failed, and facilitate chat. I even created a custom programming language and a custom database, for maintaining customer accounts, to prevent anyone from being able to access user info if they were able to somehow gain access to the server.</p><p>Behold, Valiant: a programming language, web server, and database rolled into one. It featured weird stuff like using apostrophes to set class data, multiple word variables, and human readable sentences</p><p>(add 500 to chilton’s balance)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*02OYGtEabpSMMUOFilt5Kw.jpeg" /></figure><p>The problem was, so much of the whole system was my own code, that cognitive smoothing was kicking my ass on bugs. I could find other people’s bugs pretty fast. I had spent a decade doing that already. But fixing bugs in my own code was damned near impossible. We all have a blind spot to those, because of the same mechanism that lets us create typos that we can’t see. Our brain smooths right over any and all mistakes we make. It’s super annoying.</p><p>So here we were, a few days after a soft launch (we’d been beta testing for almost a year), we had about 1300 simultaneous players, from all over the world. And something was crashing. It was killing off entire games. No money had been lost, but lots of angry emails were piling up.</p><p>I’d built the server to “fail elegantly” when it crashed. Basically each connection ran as a memory protected micro-process. If it crashed, it could be wiped out without affecting the rest of the server. And the database was designed to double check all entries, to prevent anyone from maliciously changing records or trying to trick it into recording extra large wins or bets.</p><p>But something was still crashing, and I didn’t know what. For a solid 48 hours, I spent every minute looking over my code and the crash logs, trying desperately to figure out what was wrong. If only I could see how it all fit together. There was just too much code though. Too many interconnections.</p><p>So I did what any good programmer would do and rather than solve the problem, I wrote an app to solve the problem :)</p><p>This app used the language parser I’d made for the Valiant language, and read through my C code, and printed this giant mess of lines and words. It ran inside SuperCard. And each image could be printed.</p><p>What? This was 2003. We didn’t have high res monitors and I certainly couldn’t show more than a small graph on the screen at a time. This screenshot is from the same program running on a more recent Mac, around 2015, I think:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*57SIgRzt9G6HuoWG5xFVfg.png" /></figure><p>So yeah, every page here could be PRINTED. I printed about 200 pages and stuck them on the walls all around my office, then started drawing on them like some kind of conspiracy nut.</p><p>About a day later, I’d solved the problem. But it stuck with me. That was one hell of a bug, and I had to do something really out there to even imagine the code in any way other than what I was used to.</p><p>So what if… what if I could make an app that would do that for me? What if I could actually map out all of the code, and them move it around on the screen? That way, cognitive smoothing wouldn’t kick in, and I’d see my bugs immediately. And I was right. It just took awhile to make it happen.</p><p>It would be over 20 years before I’d fully build out this system. But today, it’s here, and it’s very very powerful:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FJEOP-0S-VS8%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DJEOP-0S-VS8&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FJEOP-0S-VS8%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/37df562c54a6a94becd839fceee3c23a/href">https://medium.com/media/37df562c54a6a94becd839fceee3c23a/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c84add853a2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A post-Net-Neutrality Internet is easy to predict. We’ve actually been here before.]]></title>
            <link>https://medium.com/@chiltonwebb/a-post-net-neutrality-internet-is-easy-to-predict-weve-actually-been-here-before-6d681bf8a519?source=rss-becc3d200a8a------2</link>
            <guid isPermaLink="false">https://medium.com/p/6d681bf8a519</guid>
            <category><![CDATA[net-neutrality]]></category>
            <dc:creator><![CDATA[Chilton Webb]]></dc:creator>
            <pubDate>Thu, 14 Dec 2017 21:48:00 GMT</pubDate>
            <atom:updated>2017-12-14T21:48:41.692Z</atom:updated>
            <content:encoded><![CDATA[<p>A post-Net-Neutrality Internet is easy to predict. We’ve actually been here before.</p><p>In the late 90’s, I ran a website for Internet Service Providers called the ISP Resource Page. I also created setup kits for ISPs all over the world, and helped a number of ‘Mom and Pop’ ISPs get started. As a result, I became intimately aware of the problems they faced.</p><p>During just those few years, a number of ISP organizations sprung up, like the ISP-C (ISP Coalition). The FCC hadn’t really decided on how the ISP industry should be regulated, so they would regularly reach out to these organizations for advice. Each ISP was different, but together they had some common concerns.</p><p>A few ISPs were selling filtered content, so if you bought your Internet through one of those, you had a reasonable guarantee that your kids wouldn’t find porn websites, for example. That was a pretty decent idea.</p><p>But that quickly gave way to a more nefarious fear, that an ISP could act as a filter for things *they* didn’t like, without you knowing it. Almost overnight, some ISPs were (quietly) blocking or redirecting traffic to competitors’ sites.</p><p>This, thankfully, is not the case with most ISPs today, thanks largely to Net Neutrality. But I have to stress, the concept of redirecting traffic so your own users won’t see something, is by no means new.</p><p>And now, it will happen again. Net Neutrality rules preventing this have been scrapped, as of today. ISPs will be allowed to filter and block content they do not approve of, before you can see it. That could be for political, religious, or even personal reasons.</p><p>On the upside, you won’t know what you’re missing, and I’m sure you’ll be better for it. Smarter, better people than you, will decide what content you should be allowed to see. There’s no way this could be abused.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6d681bf8a519" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>