How i use AI – Part 2

What was meant to be a single, simple article about my AI use has turned into a series, mostly because i felt the pull to delve into AI apologetics in the vain hope that no one would yell at me in the comments.

In Part 1, i described the experience from the trenches of using AI as a code assistant, and explained why i think the promise of AI completely replacing programmers is Silicon Valley hokum meant to spike startup valuations.

In this part, i’m going to begin to describe the various creative ways in which i’ve been using AI to improve my life and workflow.

People Who Wear Shoes Commit Crimes

i’m writing this the week that MIT released a study demonstrating people who employ LLMs (Large Language Models) in essay writing have decreased cognitive ability and poorer recall. The way i saw this study disseminated on social media was from people trumpeting that “people who use ChatGPT are dumber than people who don’t.” Then, cue me with my little blog series titled How i Use ChatGPT. From their perspective, i might as well title it Reasons Why i Am Dumb.

i return to my knife analogy from Part 1. Imagine the social media pundits told you “People who use knives are more violent!” Use them how? Are they whittling wood to make adorable little folksy sculptures? Are they dicing green onions to put in a lovely home made soup?  Or are they stabbing orphans? 

And then you click through to the study, and the intro paragraph says “People who use knives to stab orphans are more violent than those who don’t.”  Rrrright. Thanks, MIT.

“The questions on the physical were strangely worded, like: ‘Have you ever tried sugar? Or PCP?'” – Mitch Hedberg

Anyway, on to it at last. Here are the myriad ways in which i use AI, or in other words, Here’s How Variably Dumb i Am.

Saving Money

When i returned to my board game YouTube channel Nights Around a Table from a 2-year hiatus, i knew i had to make a better go of it this time out. This meant laser-focussing on the videos that were likely to perform well, and cutting out the chaff.  Fresh off my experience of using AI to rehabilitate my programming skills at the video game job, i knew i needed to make two initial, crucial purchases: an Adobe Creative Suite subscription (the Suite has all the software i need to do the YouTube job) for about 800 bucks for the year, and a $20/mo ChatGPT Pro subscription. Groceries could wait.

Me, prioritizing.

i recalled vaguely that years ago, i had been able to talk Adobe into providing me a 40-50% discount on their software, and of course i wanted to shoot for that discount again. So i consulted with ChatGPT to figure out how to do that. i fed Adobe’s text chat agent’s messages into ChatGPT, and had the robot respond on my behalf. By the end of the conversation, i’d gained the deep discount i was after, and it was exactly enough money to pay for ChatGPT Pro for the year.

Eat My Shorts

When i asked ChatGPT what i needed to do to get to 100k subscribers on YouTube, it had some interesting ideas. It told me i needed to lean in hard to Shorts, YouTube’s truncated vertical format for smartphone-using audiences that they they hoped would rival TikTok. ChatGPT said i needed to lean so hard into shorts, in fact, that i should be releasing two of them a day.

“…in a row?”

i’ve never worked a “proper” social media job with a content calendar and a release plan. i was out of my element.

i came up with a list of “content buckets” (side note: i despise the term “content” – that’s a whole other blog post). i ran my list by the robot, asking which ones it expected would perform best. We curated the list, and i even added one of the robot’s own format suggestions. Here’s where we landed:

• Review Round-up – liberally “borrowed” from a TikTokker i watch, this is just me reporting on how other people reviewed a particular game
• Guess the Game – i provide 5 clues indicating a board game, and encourage people to comment with how many clues it took them to figure out the game i’m thinking of
• One Cool Thing – i zero in on one interesting or unique aspect of a given game
• Component Close-up – like One Cool Thing, i showcase a particularly well-made or unique component (a pawn, a miniature figurine, etc)
• My Favourite Card – this is another focus series that showcases a single card from a given game
• The Drift – excerpts from longer-form videos, intended to entice my Shorts viewer to watch the Longs 
• Bust a Myth – this is the series the robot recommended. Essentially, it’s “board game rules we got wrong/misunderstood,” and there were various ways to title it, but the robot thought this would be the snappiest, most provocative title. i’ve released 2-3 shorts in this series so far, and some commenters have expressed frustration with the title (“These aren’t myths! These are just stories about how you didn’t read the rulebook properly!”) … but the adage “there’s no such thing as bad publicity” in marketing applies similarly in social media, where “there’s no such thing as unwanted engagement.” To these algorithms, a “thumbs up” has identical value to a “thumbs down” — it’s all engagement, and engagement is the ultimate good outcome.

Me, leaning hard into shorts.

Off i went, producing a trove of around 30 shorts that i could drip-release daily through the rest of the month.

Pulling ‘Balls

What i didn’t yet understand was how to determine which of these formats was successful and worth continuing, and which formats i could safely bin. If i were running a social media account for a large brand, that task would be fairly straightforward: if the thing gets a million sets of eyeballs, keep doing it.  If it “only” gets tens or hundreds of thousands of views, switch focus.

Fine. But what happens when you’re looking at single-digit numbers?  This format got 35 views and 2 Likes. This one got 27 views and 15 Likes. Which one was more successful?  i had no idea.

There are dozens of us. Dozens!!

But the robot had access to vast amounts of compiled data about post performance and, presumably, an understanding that absolutely dwarfed my own. i can’t synposize a marketing textbook in seconds. But the robot can.

So i had the robot generate a spreadsheet for me that i could fill in as i released each short. YouTube has an absolutely nauseating array of metrics you can tap into, and as a neophyte, i had no idea which ones to focus on. Were traffic sources important? Did i really care about thumbs up and down? Did i really need to look at performance by region? Which time frames mattered? How could i compare a Short that i released on a Tuesday afternoon to one i released on the Saturday of a long weekend?  And how could i tell if i was getting any closer to my goal of 100k subscribers at a better pace than if i was doing all of this without robot assistance?

For the first week, i released one short a day (down from the robot’s recommendation of 2 per day). i faithfully tracked the stats as the robot prescribed. i made some observations:

The first short i posted (in the predictably popular Guess the Game format) did quite well. Subsequent shorts, even in that same format, did not. i conferred with the robot, who confirmed my suspicions: user-generated platforms like TikTok and YouTube vampirically need us to post videos, and will artificially inflate our views by distributing our first few contributions more widely across the platform. Then, when they’re satisfied that we’ve settled into a routine of posting, they’ll close the floodgates and let us sink or swim on our own merits, as we constantly chase that initial engagement high. The first hit’s free.

God damn the pusher man

Then, some more cracks in ChatGPT’s strategy started to show. As i posted my relentless stream of daily shorts, i began to perceive that, beyond the artificial high the platform provided, people were quickly tiring of me. i asked the robot about it, who responded

So i maintained a 4x/wk schedule for over a month, until my well of banked videos had largely run dry. When i asked the robot when i should post videos (like, what time of day), the robot told me Mondays, Wednesdays, and Sundays at midday. i went off-book and tried other days and times just for kicks. i noticed that shorts were performing better on Tuesdays and Thursdays, when the robot told me NOT to post.

When i asked the robot about that, the conversation went something like this:

Once again, this is the long and short of it: if you’re using the robot to help you in an area where you already have some degree of familiarity and expertise, it can amplify your efforts in a fraction of the time you’d normally spend. If you’re using it to guide you in a field that’s unfamiliar to you, you might be hosed. i’m not ready to confess, like MIT, that employing ChatGPT in certain ways will make you dumber. But i will cop to the fact that if you’re already dumb, like i am with social media marketing, the robot won’t do you any favours. You’ll lack the knowledge to be able to discern when the robot is feeding you — in technical terms — cockamamie bullshit, and your work will suffer for it.

ChatGPT and ChatGPTer

Propagation of the Species

So how did the robot do with helping me reach my goal of 100k subscribers on YouTube? i asked the robot to self-evaluate, by feeding it the filled-in metrics spreadsheet it had structured for me.

It was able to very quickly analyze the stats and make inferences that i feel i can’t make, because whenever i stare at a sheet full of numbers, my eyes glaze over and circus music plays in my head.  But while i may be sutpid with regards to social media marketing, i am — on the whole — not dumb. 

i mean, sometimes i wear glasses.

(Also: this is what Sora, ChatGPT’s image generator, thinks i look like, after feeding it 15 different pictures of my actual face in various poses and lighting conditions)

The robot pointed to this or that format in the spreadsheet, saying “look – this format, overall, had 30% more engagement” or “that format performs 40% better on TikTok than on YouTube.” But in all of it, the robot had lost sight of my goal.

But i didn’t.  i want to reach 100k subscribers on YouTube. So like Darwin’s biological imperative, which essentially boils down to “doesn’t matter — had sex,” i didn’t particularly care which format had more comments or more thumbs up. My goal was, specifically, YouTube subscriptions. If one format resulted in more than zero YouTube subs, it was a success, because it furthered my goal. If a format got a lot of views and engagements, it could go hang for all i care.

A more sophisticated and experienced social media manager might retort “Well, it all helps.” Sure, Julie. But i’m too dumb to be able to trace when heightened engagement on one video on TikTok led to a sub click on an entirely different video on YouTube. So i’m just gonna key into that “sub” stat and cling to it like it’s my biological imperative. 

The Receipts

So far, i’ve been giving you the joking gist of my conversations with the robot. Here’s an actual copy/pasted conversation where it dawned on me that the robot had led me down the garden path with its shorts strategy, and because i wasn’t a subject matter expert, i had foolishly and uncritically listened:

i don’t begrudge the robot for feeding me an ineffective strategy. i think it highlights one of the actual dangers of AI for conscientious users, who aren’t just telling it to write their essays for them: if you are a subject matter expert, it can speed you up. And you’re knowledgeable enough to know when its recommendations are bunk.  But if you’re consulting with an LLM in an area where your bullshit-fu is weak, tread with caution. i don’t think ChatGPT is making me stupid. But i do think it’s been amplifying my stupidity.

ChatGPT’s deadly “Duh Ray” is fully operational

That’s my silver lining: i’m smarter now about social media marketing, having gone through this exercise. And i’m learning to trust my instinct and intuition more. If something seems like a bad idea (like posting 2 shorts a day), it likely is. i didn’t need the robot to tell me which shorts format would be the most successful — i knew just by looking at the list, before ever pressing “record.” And i knew that high engagement on TikTok was a smokescreen: my goal is YouTube subs, and i need to refocus my strategy towards meeting that goal.

What’s Next, Swami?

Cool, Ryan. You paid 20 bucks a month to get duped by ChatGPT. What other invaluable wisdom can we learn from you, o sage?

In the next part, i’ll tell you about the ways in which ChatGPT has crashed the gates and trampled gatekeepers, helping me to accomplish things that have always been one comp sci honours degree and 6 years’ Google API implementation experience out of my reach.