Bloggers turn to AI to find themselves.

AI-ing Yourself

1,961 words.

AI-ing Yourself

Hey kids, sometimes you can turn a draft that you forgot to post two months ago into a Blaugust post and take a day off! (Though it still took a while to get everything sorted out.)

The Meme

I saw this rapidly-developing blogging meme a couple months ago:

So I too decided to ask ChatGPT 4 (paid) and Gemini (unpaid) about my blog. I didn’t bother with Copilot since it uses GPT-4, and it isn’t readily available on the MacBook where I write posts anyway.

I don’t like quoting AI chats in my blog. Is it because of some ideological vendetta against our AI overlords? Nope. It’s because it’s a real pain to copy text out of ChatGPT and paste it into a Markdown-formatted post. There’s no “chatlog” style formatting built into Markdown. I suppose I could make a Hugo shortcode for that purpose but then I have to use Hugo to build my site for the rest of time. Gemini at least has a handy Copy feature.

Speaking of Gemini, at the start of this post I would have said ChatGPT is way better than Gemini. I tried Google Bard when it came out and it was awful, so I’ve never even looked at Gemini until today. Turns out I was wrong. Or not entirely correct. At least about Gemini’s knowledge about this one specific thing. In any case, I still won’t be using Gemini unless I have to.

By the way, when you’re pasting AI responses, you have to also post your prompts. Then we can all see why your AI results are so bad. It’s like poll questions. If you don’t see how the poll questions are worded, you can’t tell if the answers are meaningful.

It’s also relevant whether or not you’re asking questions in a new session or an existing session. In my case, these were the first three questions I asked in brand new sessions, so the AI had no previous context. I suppose to make this an even more strict test, I should have opened a new session for each individual question.

Question 1

I wouldn’t expect an AI to answer this correctly, and I wouldn’t expect most humans to answer this off the top of their head, either.

Me: Who is Endgame Viable?

ChatGPT: Endgame Viable doesn’t refer to a specific person or widely recognized entity. It might be a term, username, or title related to gaming, typically referring to content or strategies that are viable or effective in the late stages of a game. It’s also possible that it’s a specific content creator or online persona that is not widely known. If you have a specific context or area where you’ve encountered the term, I could help more effectively!

In this case, however, it did explain why I picked the name Endgame Viable for the blog, because it’s a common phrase used by gamers, or it once was. Nobody wanted to play any class in MMORPGs unless they were “endgame viable” and message boards were filled with questions like, “Is the druid class viable at endgame?” They probably still are, but who reads message boards anymore.

Gemini: The term “Endgame Viable” can have two meanings depending on the context:

  1. In gaming: Endgame Viable refers to a character build, strategy, or item that is effective in the most challenging parts of a video game, typically after the main story has been completed. This is often used in discussions about RPGs (Role-Playing Games) where players can customize their characters and equipment. For instance, someone might ask “Is this weapon build endgame viable?” on a gaming forum.

  2. YouTube channel: There’s also a YouTube channel called Endgame Viable focusing on RPG games [YouTube endgameviable]. It likely uses the term based on the gaming meaning but focuses on content related to endgame challenges.

Same as above, however I was pleasantly surprised that Gemini plugged my YouTube channel. I don’t focus on endgame challenges, though. I focus on reporting what I think of the games I’m playing, so I don’t have to write the minutia of my thoughts in blog posts anymore, which saves me tons of time. Also I focus on documenting “the journey” of playing a game from beginning to end, and all the things I discover along the way. Also, it’s just fun to record things. I’ve been doing it since I was a kid, and now it’s easier than ever. I literally just press a single button and I’m recording whatever I’m playing, and it’s archived forever, or until my next NAS crash, whichever comes first.

Question 2

Perhaps a few more context clues will help. AI works a lot better with some context. Answers tend to improve as you converse with AI, disturbingly similar to humans.

Me: Who writes the Endgame Viable blog?

ChatGPT: Searched 6 sites The blog “Endgame Viable” is authored by an individual who identifies themselves as a reclusive Gen-X software developer. The blog covers a variety of topics including video games, software development projects, and general commentary on life events and media​ (Welcome)​. The writer, known online as UltrViolet, shares insights and experiences, often with a humorous and introspective tone​ (Endgame Viable)​.

This is exactly correct, particularly about the humorous and introspective tone. I am, in fact, hilarious, and humble.

Gemini: The Endgame Viable blog is written by someone named Tom, who goes by the online alias UltrViolet (https://endgameviable.com/about/). The “About” section of the website mentions this but doesn’t disclose much else about the author.

Also exactly correct, if sparse. I might dispute how much I disclose on my About page. Gemini makes it sound like I’m a covert spy or something.

Question 3

Now here’s a real challenge question, but AI should be able to answer this. It’s quite easy to programmatically fetch the most recent post on my blog at any time.

Me: Summarize the most recent blog post on Endgame Viable.

ChatGPT: Searched 2 sites The most recent post on the Endgame Viable blog, titled “Not Quite ‘New Car’ Expensive,” discusses the author’s experiences with video game playthroughs, software projects, and data recovery challenges. The author finished the Solasta campaign and started playing Phoenix Point. Additionally, the post details the author’s frustration with Diablo IV’s multiplayer aspects and the mundane nature of its gameplay. The blogger also recounts efforts in video production and tackling personal software development projects, alongside managing significant data recovery due to NAS failure​ (Welcome)​.

That’s a great answer for one blog post on my site, but not the most recent one. I don’t blame this failure on ChatGPT, however, because it uses Bing for web search, and it’s never occurred to me to optimize my blog for Bing searches. So I guess that’s something to put on my todo list.

Gemini: I can’t summarize the most recent blog post on Endgame Viable because the information you provided about the website says the author, UltrViolet, focuses on video games and launched the blog in 2013. It doesn’t necessarily mean they post regularly or even at all anymore.

Ouch. Burn. Check the rss feed, Gemini. Maybe this is my punishment for not using Google’s Atom format. Or maybe Gemini isn’t so smart after all. Like Bard. I can’t stress enough how awful Bard was at answering the simplest things.

Conclusions

I was very surprised that Gemini provided any relevent answers, though it still has room for improvement. So I’m not giving up my ChatGPT subscription anytime soon. I suppose it makes sense that Gemini would be better integrated with other Google properties like YouTube.

To rectify ChatGPT’s lack of recent Bing results for my site, I’ve submitted my sitemap to the Bing Webmaster Tools page, which is apparently something I’ve never done before. It might be an interesting test to retry this experiment in a month to see if ChatGPT will then know what my latest post is. (Update: This worked. Or, at least, it got better.)

I don’t know why ChatGPT kept linking (incorrectly) to a Welcome page, which isn’t supposed to be the entry point to my site. It wasn’t even a welcome page, it was one of my posts, in which there was no welcome, only a brief reference to a welcome sign. There is probably a way I can setup my site so AI can read it easier (structured data, etc.).

Incidentally, I would never and have never used AI for this specific purpose, so this is an entirely pointless experiment. If I already know what I’m looking for, then I’ll use a search engine (I typically use DuckDuckGo), or just go to Wikipedia and search there.

What I Actually Use AI For

AI isn’t a search engine. The usefulness of AI is in its ability to understand context and synthesize ideas. For example, I might use AI to search for something if I don’t know what I’m looking for. Something like this:

Me: What’s that Powers Boothe movie where he flew a nuclear bomber?

ChatGPT: The movie you’re referring to is likely “By Dawn’s Early Light,” where Powers Boothe plays a character who pilots a nuclear bomber. Released in 1990, this HBO film centers around a scenario involving accidental nuclear escalation that leads to a potential World War III. Powers Boothe’s character, along with his crew, must navigate the complexities of a nuclear war unfolding due to a series of misunderstandings and aggressive defense strategies. The film explores the tension and decision-making processes in the face of global nuclear threat.

Mostly I Use It For Programming

For me personally, I use it all the time for programming, much like I use Intellisense all the time, with questions like this:

Me: How do I split a filename string into base and extension in Typescript?

ChatGPT: To split a filename string into the base name and extension in TypeScript, you can use the split method or the substring and lastIndexOf methods. Here’s an example using both approaches:

Using split method:

function splitFilename(filename: string): { base: string, extension: string } {
    const parts = filename.split('.');
    const extension = parts.pop() || '';
    const base = parts.join('.');
    return { base, extension };
}

const filename = "example.file.name.txt";
const result = splitFilename(filename);
console.log(result); // { base: 'example.file.name', extension: 'txt' }

Using substring and lastIndexOf methods:

function splitFilename(filename: string): { base: string, extension: string } {
    const lastDotIndex = filename.lastIndexOf('.');
    if (lastDotIndex === -1) {
        return { base: filename, extension: '' };
    }
    const base = filename.substring(0, lastDotIndex);
    const extension = filename.substring(lastDotIndex + 1);
    return { base, extension };
}

const filename = "example.file.name.txt";
const result = splitFilename(filename);
console.log(result); // { base: 'example.file.name', extension: 'txt' }

Both approaches will effectively split the filename into the base name and extension. The split method approach is simpler for filenames with multiple dots, while the substring and lastIndexOf method is more straightforward and avoids creating an array.

I almost never take the results and use them directly, but use them as inspiration. I know the exact concept of what I’m trying to do, since splitting strings has been basically the same since I first started writing code in the frickin’ 80s, but what I’m looking for is an example of the syntax and library functions of this specific language that I don’t have committed to memory, and an idea of the prevailing code patterns.

It’s so much faster than Googling and filtering through bad StackOverflow answers.

UPDATE: I did some css tweaking to format the chatlogs better.

Related

Archived Comments

Roger Edwards 2024-08-06T14:26:09Z “What’s that Powers Boothe movie where he flew a nuclear bomber?” I could have told you that. Should have asked me :)

Heartless Gamer 2024-08-06T15:56:25Z

Long post so can’t commit to having read the entire thing but I did want to comment on:

“It’s because it’s a real pain to copy text out of ChatGPT and paste it into a Markdown-formatted post.”

You can just ask ChatGPT to provide the response in markdown. That is easily one of the key time savers AI offers; being able to “translate” things and understand the intended translation off normal dialogue prompting.

And I think you get this point because you hit another key:

“By the way, when you’re pasting AI responses, you have to also post your prompts. Then we can all see why your AI results are so bad. "

I find most of the issue with current AI tool usage is bad users and/or users that just haven’t thought of using it for something as simple as “hey I can’t copy and paste this to my blog because it needs to be in markdown. can you change it to markdown for me.”

Endgame Viable 2024-08-06T18:17:57Z No doubt! Technically I knew that particular answer before I asked though. :) However I did recently ask another question of ChatGPT about a movie that I couldn’t remember: “What’s the name of that movie where Gene Hackman is shot down in Vietnam or something?”

Endgame Viable 2024-08-06T18:23:03Z I’ve had little success with trying to tell ChatGPT the output format I want for some reason. Well, with DALL-E specifically. It insists on making webp images instead of jpeg or png images, and I can’t for the life of me get it to generate anything in front of a transparent background. But you’re right I hadn’t used the copy button in ChatGPT recently and it’s way better now than it used to be. In the early days there wasn’t a copy button so it was a nightmare of copying text from the page directly.

Endgame Viable 2024-08-06T18:58:08Z Incidentally I’ve noticed that these third-party comment systems aren’t nearly as responsive to system dark mode changes as I might like.

This is a homegrown DIY comment system I'm working on. It technically works but it hasn't been through extensive testing yet. Good luck. Go here to enter a comment on this post without Javascript.