Hah! Fixed!

By Bay Posted Tuesday Mar 21, 2023

Filed under: Epilogue 26 comments

Alright, so I have a confession to make; I have been doing the big procrastinations on this website. I could argue that it’s the grief, which is partly true, and a pretty easy ‘get out of jail free’ card. But, if I’m honest with myself, it’s not enough of the truth to just write it off. Like, yes, working on the site makes me miss my dad, and more than that makes me want to shake both him and the universe by the shoulders, because working on a dead man’s code is a fool’s errand Why does it work like that?! Why?! Tell me, god damn it!.

But, really, even if he were alive today, he likely wouldn’t be able to tell me. I can imagine the conversation very vividly as it’s one I’ve had with him before.

I’d bring him the issue, explaining that the image I’m uploading isn’t working, or the color I’m inputting isn’t being picked up by the server despite the fact that according to the inspect element, it’s exactly where it should be. He says I’ve done it wrong, since it clearly says that’s where it’s pulled from, and then, he goes down the exact same rabbit hole I just did, and confirms exactly what I just told him.

Eventually he finds that it’s because he told the server or image host to reroute through an identical piece of code in a different place, making the one we’ve been editing redundant. He scratches his head, laughs at himself, and tells me to put it in the reroute location because obviously he did it for a reason.  We never find out why, and three years later when it’s time to edit it again we have to figure it out all over.

I’d fault him for not believing me the first time, but honestly I’ve never met a coder that did otherwise.

So, yes, grief makes me not want to stumble through the coding wilderness without him, but it’s plain and simple ‘don’t wanna’ that made me take this long to finally do it.

Now, I should acknowledge that obviously when I say I ‘fixed it’ I obviously don’t mean I’ve finally moved the site over to a new location with updated plugins and security. If that were true this post would be in, or contain a link to, a new location, which it isn’t, and doesn’t. My job is to work here, with what I have. I don’t have access to the server hosting and Patreon stuff, that’s all Mom. No, my job is to write posts, do bodge style coding, and do basic management and delegation.  When I say I ‘fixed it’ I mean I finally got rid of the transparency issues and figured out how to set colors for different users. But, more than that, I mean I finally dug into the tangled wires of the code enough to figure out how all this works and how to fix little fires.

What made me finally buckle down? A Tom Scott video, and ChatGPT, obviously.

See, I’m not a coder, but I come from coding. Both Mom and Dad programmed when I was growing up. Mom did website design, Dad did game development, two really useful and very different things that I always felt like I ‘should’ pick up. But, coding changed so rapidly that by the time I managed to learn enough of anything to make anything at all, it was entirely obsolete. Dad taught me how to program a version of Snake just like he had when he was my age, and Mom taught me how to move buttons around my WordPress site, but neither skill remained useful for longer than a few months. And, meaning to or not, having parents that can just swoop in and ‘here let me do it’ when you’re slow on the uptake doesn’t really facilitate a learning environment.

So, the Tom Scott video. To summarize, (although it’s a great watch if you’ve got the time), he explains that he used ChatGPT, an AI chat program, to create some code to automate a program to label files for his email. He mentioned that he was out of the coding loop, and mostly got existential about a possible technological boom. I saw this and thought, hey, maybe I can use ChatGPT to automate updating the website.

Of course it wasn’t that simple, and I didn’t really expect it to be, but something I didn’t expect was for it to be a great teaching tool. AI doesn’t get irritated if you don’t get it on the first try. I was able to put in every line of code from the back-end of this website, and it told me line by line what it did, why it was likely there, and if I asked, how to edit it. This would have, of course, been possible with a human being, but it would have taken hours, and I would have been filtering what code to send through a filter of ‘hey, does this bit actually seem important or relevant?’. Sending it through a piece of tech meant I could take my time, wallow for a minute if I needed to, and waste its time with all the emotional garbage of just wanting to understand four hundred lines of metadata back-end code, because I can’t ask my dad anymore.

 

Footnotes:

[1] Why does it work like that?! Why?! Tell me, god damn it!



From The Archives:
 

26 thoughts on “Hah! Fixed!

  1. Michael says:

    For what it’s worth, a post showing what you put into ChatGPT, what you got out, and how it helped you think about the problem is extremely on brand for Twenty Sided. Actually, a couple dozen posts about exactly that would also be extremely on brand.
    Which is to say, if you’re up to writing it, I’ll be here reading it.

    1. Z'Greel says:

      Yup. It’s very much on brand, it sounds interesting, and someone (me) might find it useful as a learning tool too!

      1. Doran says:

        Agreed would be interested to read it!

    2. Octal says:

      I absolutely agree! I think it would be cool to see it working more as intended. The stuff about it that I usually see is more about how it just makes up plausible-sounding stuff because it doesn’t “know” the truth, doing things like telling people to make belladonna tea (poisonous).

      1. Storm says:

        I’d also be happy to see some posts about this. I’ve been largely ignoring/scoffing at the various AI tools out there, but I fully admit that plenty of that is me starting to reach the age of automatically distrusting strange new tools with all their unknowns. And much the same, I’ve mostly been seeing it in the context of large language models producing plausible sounding but wildly incorrect information, plus the kerfuffles surrounding it used in art.

        It would be interesting to see it being used in relation to coding, because that’s where I’ve heard the most consistently positive things about it.

    3. DrCapsaicin says:

      I will 2nd (3rd? 4th? nth?) this comment.

      What brought me here initially was Pixel City (still one of my favorite things Shamus posted) and the coding dives was one thing I’ve been missing. Obviously Bay, you do you, but I will agree that I would find a post on how that process went very interesting if it was something you were passionate about sharing.

    4. Dev Null says:

      My first thought as well.

    5. Trevor Hill-Hand says:

      This would be great, because I’ve been using GPT in the same way for the same reason- it’s great to be able to basically talk to “The Internet” without actually wasting a humans’ time until I’m confident I actually know what I’m doing and asking. Having a way to learn about things in a private, shameless environment is a really interesting aspect of having access to ChatGPT.

    6. evileeyore says:

      As long as it doesn’t break you, you’ve got a string of people interested in your longform on this issue and solution.

    7. Skyy-High says:

      100% agreed, I’d love to see this!

  2. Kincajou says:

    Boss! Whole article on front page :)

  3. Lino says:

    I don’t know if this is the right place to ask this, but wasn’t there a plan to publish all of Shamus’ drafts?

    1. Bay says:

      Most of the setbacks to that have been boring time/money/time/obligations sort of things that just sort of happen. Mom is the one with access to all that stuff, and she’s in the middle of moving over the mortgage and her usual working full time. So, yes, we very much still plan to do that, the ‘when’ just isn’t set.

  4. Obnoxious Bug Finder says:

    If you have the ability to change post colors now, can you change Shamus’s old posts back to gray? They turned beige recently.

    1. Lino says:

      Now they’re pink :D

  5. King Marth says:

    This is a big reason of why tutorial AI is helpful. Magic the Gathering encountered this with their video games, having an unfeeling alien partner with infinite patience helps break through all the social restrictions on seeking the help you need. The computer can’t think you’re frustrating, or stupid, or unattractive. It finds nothing unusual or irrelevant, at worst unrecognizable. It won’t change tactics just as you’re starting to figure something out, or give up and try something else. It just is, and will continue to be, doing exactly what it was told and nothing else. You have nothing to prove and everything to gain.

    There was also an interesting study on therapy and how participants, even when aware that they might have been talking to a human and not a chat bot (which would still be analyzed for the study), were still more willing to open up.

  6. hewhosaysfish says:

    One of the criticisms I’ve read of ChatGPT is that it can answer questions with plausible-sounding untruths, just because they sound plausible (because it is designed strictly as a language model and generates combinations of words that *seem* like they should go together but does not -and cannot- have any idea of meaning beyond that).

    I’m wondering how that would work with programming. If there is only one correct answer to a question, one solution to a problem, and that answer appears repeatedly in ChatGPT’s training corpus then perhaps the AI will reliably give that answer.
    But if there is a more complex question, or something that comes up rarely in the training data, is there a risk that it will just create a “credible” jumble of technical terms, like a Hollywood scriptwriter writing a hacking scene?

    1. DrCapsaicin says:

      I can’t speak to the coding aspect of this, but I’m a physics faculty member at a State University in the US and before this semester we tested ChatGPT on solving freshman-level physics problems.

      We gave it three prompts from old (published) exams and got three very different but interesting results: 1) the correct answer! With *hot garbage* reasoning on how that answer was derived. Like, could seriously confuse people bad (overly simplistic and inaccurate) reasoning. 2) A close-to-right answer with a solid explanation, but was missing a nuance of the problem that frankly lots of first-year Engineering majors would miss as well. 3) Complete nonsense. Tried to solve the problem in a ridiculously incorrect approach and got an answer way off base.

      It was an interesting look at where a language-model AI can get to. My Liberal Arts colleagues are (justifiably) worried about students using ChatGPT to write rough drafts of essays for them. Hell, if I were still in school I would 100% let ChatGpt build my rough draft and I would just edit from there. But for more specific things, it is still fairly obvious it is synthesizing from a training dataset. It’s one of my pet peeves with talking to some people about it: they don’t realize that ChatGPT can’t come up with *new* solutions to things, it can just do a really REALLY good job of synthesizing information from things that already exist.

  7. Philadelphus says:

    I code for a living, but mostly in Python; once in a blue moon I need to do some simple text manipulation for which a Python script would be overkill, but for which various UNIX tools (like awk) work great…except that I’ve forgotten 90% of what I knew since the last time I used them. Today, I needed to compare columns of package versions in two sets of logs, except one had an extraneous column which would make diff report every line as different. Inspired by this post, I asked ChatGPT for an awk command to strip the extra column from one file, which it quickly spat out, along with an explanation of what it did. That messed up the formatting (understandably, I just didn’t foresee it), so I asked a more specific question about removing the first 10 characters from each line in the one file, for which it suggested the program “cut”, which I didn’t know about before (and worked). It was getting towards the end of the day so I haven’t completely managed to compare the two logs (diff is still reporting every line different and I need to figure out why), but I was mildly impressed what it was able to give me, and how it might be useful for help with those sorts of problems that come up infrequently enough to keep fresh in my mind.

  8. rainbow121 says:

    Great work! So happy to hear that you found a tool that works for you :D

  9. Mersadeon says:

    I am very harsh on “AI” (even down to the term itself), but I think this is one of those fields where it can really help.

    Computer Sciences still have a gigantic didactics problem, and although I recognize the irony of coders basically coding their way out of that, as long as it helps people learn, I’m all for it.

  10. PhoenixUltima says:

    I noticed some of the posts in the “Good Robot” category are still transparent-colored and near-impossible to read (the posts by Arvind and Ross, specifically).

  11. Liam says:

    I’ve been using chatGPT pretty extensively lately, and I’m fine it very useful to not necessarily tell me the right answer, but at least give me an idea of what to even look for.

    I was thinking at the time that it would be been good to get Shamus’ take on the whole thing. So I’d definitely be interested in reading your experiences with it

    1. Liam says:

      Wow my autocorrect got me good in that last post :(

Thanks for joining the discussion. Be nice, don't post angry, and enjoy yourself. This is supposed to be fun. Your email address will not be published. Required fields are marked*

You can enclose spoilers in <strike> tags like so:
<strike>Darth Vader is Luke's father!</strike>

You can make things italics like this:
Can you imagine having Darth Vader as your <i>father</i>?

You can make things bold like this:
I'm <b>very</b> glad Darth Vader isn't my father.

You can make links like this:
I'm reading about <a href="http://en.wikipedia.org/wiki/Darth_Vader">Darth Vader</a> on Wikipedia!

You can quote someone like this:
Darth Vader said <blockquote>Luke, I am your father.</blockquote>

Leave a Reply to Z'Greel Cancel reply

Your email address will not be published.