• Garbage Day
  • Posts
  • Very slow and very ugly but I still love it

Very slow and very ugly but I still love it

Read to the end for the James Bond chord

I Used GPT-4 To Make An App (Sorta)

A few years ago, I came up with an idea for a little Garbage Day app. Essentially, it was just going to be StumbleUpon, but only for the evergreen links that went in the newsletter. I imagined a big button you could press that would load a new link at random. When I was younger I loved stuff like youshouldhaveseenthis.com, where some guy would collect a bunch of cool links for you to explore, and I wanted to recreate that experience. Unfortunately, the last time I truly coded anything was a Tumblr blog back in 2013. So I sat on the idea and, every once in a while, I would ask a couple friends with coding experience basic stuff about it, like, “is it possible?” and “do you think my app idea is cool?” etc. I eventually figured out that, yes, it was possible (idk about cool). The links would have to live on a spreadsheet or database and the button would summon them via JavaScript. But that’s as far as I got.

A few weeks ago, I dusted the idea off, while procrastinating amid a bunch of other projects. I asked a few freelance devs I knew if they’d be up for working on it and they were all busy. So, I figured I’d give it a go myself. I went through a bunch of YouTube walkthroughs and, once again, got nowhere. Because, even though I had a fairly good idea of how this would work in theory, I didn’t know enough about what I didn’t know to look up what I needed to learn to make it work.

Then, on Sunday, I finally gave up and just asked GPT-4 how to do it. And I had a working prototype of the app in about 25 minutes. And in the process, I learned a lot about what something like GPT-4 can do and what it can’t.

I started off first using GPT-3.5 and I found it wasn’t nearly smart enough to understand my prompt, which was:

Can you give me very simple step by step instructions on how to make a web app that has a big button that when you press it it opens an in-app browser that randomly loads a url which it pulls from a connected spreadsheet. The button would display over the in-app browser and allow users to keep pressing it, pulling up different urls from the spreadsheet each time

GPT-3.5 kept giving me useless boilerplate instructions about how to download different app building tools and then would jump to stuff way too complicated for me to understand. But GPT-4 figured it out right away. It quickly generated HMTL, CSS, and JavaScript code, walked me through how to organize them in a folder hierarchy and, most importantly, how to connect Google sheets’ API to the JavaScript. I downloaded a code editor called Sublime Text, popped my API information in, and all of a sudden I had a big button that loaded new websites. Except, there was a problem.

Apparently, in the decade since I last coded anything, most websites have decided that they’re not cool with having their entire site loaded inside the HTML iframe of another website lol. So my big button just loaded a bunch of broken links beneath it. But I didn’t know that yet. The error I was getting was telling me the “X-Frame-Options” were set to “deny”. I had no clue what that meant, so I asked GPT-4. It explained what was wrong and, even more impressive, it offered two solutions: Only use websites that allow you to embed their whole site in your site with an iframe (not an option) or just set the big button to open the random link in a new tab or window. So I went with that.

Here’s where it blew my mind. I then asked it, “Can you give me simple step by step instructions on how I turn this same idea into a chrome extension?” And it did. The idea that it could take the conceptual framework of one idea, which I had iterated on with it for a bit, and then turn it into something else, was insanely cool. Except, once again, there was a problem. The Chrome extension didn’t work. I went through the same process as above, where I told it about the errors I was getting and I eventually figured out that it had coded the extension to use a Chrome API called Manifest V2, which is being phased out in favor of V3. So I had to have it fix all that, but it wasn’t smart enough to know what it had and hadn’t coded for Manifest V2. So it took a bunch of trial and error to get it all in working order.

I then made a few tweaks myself to both the web app and the extension. I put a graphic in the button. I moved the styling around a bit. I changed the extension’s UI so that all you needed to do was hit the extension’s button to load a new link, rather than a weird drop-down menu, and I also asked GPT-4 if it could take this idea and turn it into an app for iOS and Android. It gave me instructions that seemed pretty legit, but it coded the whole thing in something called Cordova and I got scared.

Now, for any programmers reading this and getting uncomfortable, there is simply no way that GPT-4 could be a replacement for a human coder for someone like me with zero coding experience. Especially because GPT-4 is currently limited to 25 messages every three hours and the code it spits out is buggy and often confused. So if you want to make something that works really well and is fine-tuned to your specifications, it’s going to take a long time and might not even be possible tbh.

Also, by coding something this way, you end up with all kinds of stuff that you know is wrong and janky about the code, but can’t really easily fix. For instance, I don’t know why the cursor doesn’t turn into a hand when it overs over the button on the web app. And I can’t figure out how to center the button on the page. And I don’t know why the Chrome extension doesn’t allow you to press the back button to a previously-loaded page. And if anything broke for any reason, I doubt I could tell you why or fix it properly.

But what GPT-4 did do is allow me to turn my idea of an app into a semi-working prototype or sketch. Which, if you’re used to coding, is probably very boring, but for someone at my level, it felt revolutionary. And I learned enough about the parts of my app in the process of debugging it that, if I wanted to, I think I could happily spend the next few months just playing around with it. Or I could give my sketch to someone who does know how to code and have an actual version working in no time. In fact, if you’re reading this and you’re up for the freelance project of making this app better, let me know. Let’s see what we can do.

I wanted to share what I built, so I asked GPT-4 where I could host this. Its top suggestion was GitHub, but that was honestly very intimidating. It also suggested Glitch, which seemed easier. So I went with that. So here’s random-garbage.glitch.me. It currently has 135 random links pulled from the last two months of Garbage Day issues. It’s very slow and very ugly and isn’t really what I wanted, but I still think it’s pretty cool.

A Note About Ads

After a lull, I’ve had some interest again about advertising in Garbage Day. So I thought I’d put some basic information about it here for new readers. I offer one ad space in free issues of Garbage Day. It goes right here under the first essay. I only do a flat-fee ad deals because I don’t like to sell promises on what my readers will click. And I offer a price break if you want to buy more than one ad at once. The ads include a graphic, some text, and a link. If you’re interested, reply to this email or shoot me a message at [email protected].

And, as always, if you want to support the newsletter directly, you can subscribe, which gives you access to the weekend edition and the Discord server. Hit the green button below to find out more.

AI Drake Sounds Better Than AI The Weeknd

The song in the TikTok above is called “Heart On My Sleeve” and it was created by a producer who goes by Ghostwriter. The vocals in the song are AI models based on Drake and The Weeknd. This is, of course, once again an AI being used to copy the voices of black artists. Which seems to be all anyone wants to do with this technology.

But the reason this song is notable, other than the fact it features pretty convincing AI models and is produced in a way that makes it sounds fairly professional, is that Ghostwriter has put it on all streaming platforms. And appears to be threatening to upload more. In their bio they write, “I’m just getting started.”

VICE has questioned if this all just a PR stunt. While, Roberto Nickson, an AI developer, went viral on Twitter for sharing the song, exclaiming, “It goes so damn hard.” I don’t think it does, but whatever. Nickson, though, also called it, “a modern Napster moment.” Which is closer to true, though also not really. What I really think this is actually is a modern Girl Talk moment.

For any zoomers who happened across this newsletter and don’t know, Girl Talk was the first true breakthrough mashup artist. His third album Night Ripper is honestly so good that it basically ruined my ability to listen to the original songs without hearing his version in my head. And, hilariously, people spent years actively wondering why he wasn’t ever sued. But he wasn’t because what he was doing was so complex that there was no way to argue he wasn’t making something original. Also, interestingly enough, a study found that songs he sampled actually sold better after they were sampled.

If this Ghostwriter producer is serious about barreling headfirst into the no man’s land of AI music production, that, to me, is the bar to pass. Is what you’re doing with these AI models so complicated (and good) that it becomes an original work in and of itself. And even if it’s not Ghostwriter that figures out how to do this, I do think it’s inevitable that someone uses this tech to make a song that’s so popular we can’t ignore this any longer.

Please Don’t Use My Videos In Army Recruitment Memes

So I came across this thanks to Twitter user @nerdjpg. It’s an army recruiter named Keith Cradlebaugh. He’s verified on Instagram, has almost 12,000 followers, and his account is a mess. It’s full of memes edited to be about joining the army. And he’s using a lot of people’s videos, like the one above.

If you’ve never seen the original clip, it’s from a streamer named Mews. During a stream he got “friendzoned”. It went viral. I don’t know if it’s real or staged, but I assume it was staged. Not the point. Cradlebaugh edited the clip. He put the words “On this day a soldier was born” and then added a shot at the end of him welcoming a prospective soldier into the recruitment office. You can watch the clip on Instagram here. Oh, also, Cradlebaugh did a whole series of “On this day a soldier was born,” featuring random clips he found of men be supposedly rejected by women. Very cool.

Obviously, this sucks and is a nightmare and, honestly, if the army wants to recruit a bunch of violent incels, they should just go straight to 4chan and start posting there (wouldn’t be surprised if they are). But here’s my question, did they get permission from Mews to use his face in recruitment material? What about the other “memes” that Cradlebaugh is posting? Like this one featuring comedian Tim Robinson? Or this one featuring a random TikTok user.

Cradlebaugh’s account is, by definition, about army recruitment. So this content is being used to actively advertise for the US Army.

TikTok Is Also Banned In China

On Friday, Montana voted to ban TikTok in the state. The bill, which is called SB419, it goes into effect next year if it’s signed, and it would block the ability to download TikTok. If an app store allowed a Montanan to download TikTok, they’d be fined. It’s a first amendment nightmare and a troubling sign of things to come. And, as I wrote over the weekend, it feels like the official start of a new Satanic Panic/Red Scare wave of good ol’ fashioned American hysteria around Chinese technology.

It was also an aggressive enough move that China’s wolf warrior Twitter accounts are publicly grumbling about it, with Chen Weihua, a columnist for China Daily, suggesting China ban Montana beef.

Here’s the thing, though. TikTok is also banned in China. There are a lot of folks who keep making this point. YouTube creator Hank Green went viral arguing this same idea back in March.

I think the TikTok ban discourse in the US is ludicrous and feels like a panic response to the waning influence of the American tech industry, but I also don’t think China has any moral high ground here. They don’t let their citizens access the app either. They want all the soft-power influence of TikTok without any of the society-melting algorithmic decay it causes. If China’s online nationalists want to complain about anti-TikTok saber-rattling in Washington, fine, let us all on Douyin.

Why All The AI Guys Like This?

An AI evangelist that goes by Paul.ai went viral (in the bad way) over the weekend after he posted a big thread about how you can generate “realistic food photos” with AI. The thread was a mess and he ended up deleting it. Everything he was suggesting in it is, thankfully, very illegal.

I guess I just don’t get the mindset here. There is a seemingly endless reservoir of unflappably enthusiastic (white) guys who all bought Twitter checkmarks and spend their time promoting how-to guides for getting rich quick with AI. I suppose it’s just a new form of snake oil for a new kind of technological revolution. We get all these promises about how whatever new thing is in the news will make our lives magically better when in reality it just does this.

The Song Of The Summer

Some Stray Links

P.S. here’s the James Bond chord.

***Any typos in this email are on purpose actually***

Join the conversation

or to participate.