I’m in charge of the games we play at holiday get togethers. This past Christmas I used ChatGPT to write a script that would update a Family Feud game with answers and sound effects in google sheets. We cast it on the tv, broke into teams, and good time was had by all.
It’s janky, but it was fun.
(It’s really janky because the “ding” sound effect had to be hosted in google drive and the script used a drive API key to access the file unlike the buzzer sound which was hosted on some public URL. If you call the API too many times you have to regenerate the key so every game the ding sound would eventually stop ringing and I had to paste a new key into the script. Like I said, janky.)
I recorded this last night just so you can see what I mean:
But this brings me to the subject of coding and AI. You won’t become a software developer using LLMs but it’s worthwhile to see just how much you can do by just prompting. More importantly, you can see just how much the LLMs can do.
On Sunday I boosted Khe’s buy/rent calculator.
The most interesting part of the calculator in my view was its creation.
Both Khe and my friend Paul took Nat Eliason’s self-paced course Build Your Own Apps With AI.
🌙Khe has been documenting his learning journey and projects publicly (see thread)
🌙I asked Paul if he’d give me a quick tutorial and he made me a quick vid on the spot. It’s so sick. He gave me permission to share, check it out:
🎥Paul Millerd just casually making an app with AI in under 8 minutes (Descript)
I’ve been using VS Code with Copilot for most of the last year. I stink at coding but many of the posts I wrote last year wouldn’t have even come together if it weren’t for LLMs making the data wrangling doable in an acceptable timeframe.
I hate to have a message that aligns with grindset merchants but I do think it’s a mistake to sleep on this stuff. At the very least, you should aware of incredible capabilities that are freely available and simply did not exist 2 or 3 years ago.
[If I have a doomer thought about this — we are in a goldilocks period. You can stand out today by using these tools to do more. But it will be table stakes within a year or two. If you have a desk job and aren’t automating or accelerating your work, you will look like an underperformer. My analogy is how traders who weren’t fluent in Excel or model-building needed to wait for someone with technology skills to help them code up their ideas. Eventually, you just ideate and iterate too slowly if you aren’t self-reliant. The ultimate doomer thought is that life itself is in a goldilocks period before we have to justify our existence as a species and doing things “just because we can”. If we’re around long enough I suspect that a chart of human adaptive fitness on the y-axis and curiosity on the x-axis that the function is an upside down parabola. Like Y = 2030-(x-2030)² where 2030 is in fact the Gregorian year.
A similar energy:
”Centaur” refers to that period of time where man + machine is the best performer but it’s a transitional period. The robot became singularly dominant in chess and Go.]
I urge you to read Paul’s recent post. His take articulates my best guess even if the range of possibilities makes it hard to have confidence in one’s best guess.
LLMs are Dissolving & Creating Work (5 min read)