I built a thing (with AI), part 1: What I built
As the post’s title says, I built a thing. With AI.
What did I build?
Well, the basics are that I built a substack called One Idea Per Day; you can go sign up for it here. Every day, One Idea Per Day scours the web for investing ideas and then does a brief write up of whatever idea it likes the most. Some days the idea is an event driven idea, some days it’s a value idea, etc. The Substack also has an iterative function; reader feedback on ideas (likes on posts, feedback or suggestions on its feedback form, etc.) can change what ideas get published in the future.
I will be honest with you: the posts are nothing particularly special. They remind me of the types of posts and ideas I would come up with when I was getting serious about investing as a junior / senior in college: really eager, relying heavily on other investors’ ideas and thesis, and without a lot of my own thinking.
If the posts are nothing special, then why did I start the Substack?
Because I wanted to experiment and play around further with AI. And here’s the thing about that Substack: I don’t write a single thing on it. I used AI to build the whole thing and I set up an automated AI that now does all of the writing / research / publishing (it even automatically generates all of the images in the post!). In fact, since I launched it, I haven’t looked at the Substack or touched it unless it surfaces a name I want to brush up on.
So I think the whole thing is fascinating. I built it over the course of a few days with maybe 3-4 hours of total work on my end. Again, I’m not claiming it’s the most impressive thing in the world, but I wanted to share it (and some of the process that went into it) with you for three reasons. One of the reasons is positive; the other two are negative.
Let’s start with the positive reason for sharing: I talk to a lot of investors, and the disparity in AI usage is striking. Some investors are so deep into AI that they’re talking about using agents to go out and build their own replacement for software that they’re currently spending tens of thousands of dollars on (Bloomberg, Capital IQ, etc), while others are still wondering if they can even use AI as a google replacement for fear of hallucinations. My hope is that sharing a silly little project like building and automating an idea generation substack will help encourage some added creativity on what’s possible with AI and get a few more investors to dip their toes further into the water (and maybe give me some notes and tips if they’re seeing anything that’s really helping them improve as an investor!).
Alright, let’s turn to the two negatives.
First, one of the real issues of our time is going to be the flood of AI generated content. This chart with the proliferation of e-books post-ChatGPT has lived a little bit in my mind for the past few months:
The rise of AI content is going to apply to a ton of places, but for investors it’s fundamentally shifting how you can read investor letters, substacks, and other sources of investment ideas. I find that investors who haven’t experimented much with AI are much more prone to fall for all of the tricks of AI (“it’s not X; it’s Y”) than investors who have been using it, so a small hope of the One Idea Per Day substack is that by openly admitting it is 100% AI generated it will help open investors’ eyes to what AI generated content looks like.
The other negative reason I launched the substack is that I remain very concerned that all investors are “cooked” (to use language the youth will understand). As an investor, your job is to gather and process a bunch of information (10-Ks, podcasts, industry journals, management interviews, etc.) and use that to identify pockets of the markets that are mispriced. AI doesn’t need to sleep, and it can identify and process information faster than any human could ever hope to. If AI can consume more information faster and more completely than a human, what is the missing ingredient that will keep AI from becoming better than humans at investing?
I don’t know the answer to that question1…. but I wanted to play around with the One Idea Per Day substack to experiment with how an (admittedly unsophisticated) AI might approach idea generation.
/Fin
I’ll let you check One Idea Per Day out and decide how useful or interesting it is for you… but, in a “teach a man to fish” way (and similar to what I did with my Sweetgreen review tool), I also want to show some of the things I did to build out the substack (or, more accurately, the prompts I used to have AI build it out!). I’ll include that as a separate post later this week just because that post is running long / has a ton of images in it. I’ll see you then (and, if you’re not subscribed, be sure to to make sure you don’t miss that post!).
I kind of hope Buffett’s famous “sell your IQ points if you’re >130” quote is right, and the rise of AI with 200+ IQs creates more opportunities for dumb dumb humans.