How AI/ML & Deep Learning Investing has Changed in the last 18 months
A View from The Investor Seat
The title of this article sounds boring, immensely boring, slam your head through a wall boring, but the reality is, it's kinda cool; let me show you. When I started investing in ML companies pre-hype cycle (yes, there's a crash coming; have you seen the multiples....), the typical founder was highly experienced in deep-learning architectures & algorithms; they were complete nerds, the best kind, my favorite, the kind of nerd that I wish I were smart enough to be. Typically, all the AI/ML companies were started by these folks, and usually, the tech was high quality from day one; the real question would usually be something like, can this somewhat reclusive savant run and scale a team, manage egos, and do founder-led sales? The tipping point was on the operations of running a company, not the tech. I'd feel out the person and get a sense that they do go out and manage relationships well in real life; most of the time, they were managing them better than I was.
Then ChatGPT hit the world, AI became the talk of the day, and the modern-day gold rush began; unfortunately, that gold rush brought; how do I say this in a politically correct way, A fuck ton of idiots into the market. And when I say idiots I don't mean an ML researcher who has been working hard, getting their doctorate, and living on ramen in SF while they try try and try again (I respect those people immensely); I mean someone who lived on their parent's basement couch at 40, talked in circles for years on Clubhouse about Crypto without understanding blockchain, lived with perennially orange fingers due a high rate of Cheeto consumption (not even the good puffy type) and was generally a non-contributing zero to society, all of the sudden these people created AI companies.
This made my job more entertaining and more challenging. You see, I used to have to diligence the person; the tech was generally never really in question; the only thing I wondered was, is this infra a 7/10 or a 10/10, either way, the company was bound to have an ML product, how that product would fit into to a repeatable sales cycle was the main question. With the influx of Cheeto-covered crypto burnouts flooding the field, I have to spend the first 15 minutes of every call trying to figure out if this person has any idea what ML stands for and if their AI is just an API.
I'm going to go on a short rant here, mainly because it needs to be said but also because this is my substack, and I feel like it. A wrapper is a good business, or a least a good enough business to make a little cash on the side; you create a landing page, call an API, link a stripe account, and enjoy your 500K a year while it lasts, I'm cool with that, shit I might have do it myself soon. That said, it's not a venture backable company, let alone a hard-hitting fence rattling, earth-sucking, baby-making, Deep Learning company.
With the influx of new AI founders, I've found myself playing detective more often than not; I need to see the infra, I need to know if someone is training their own models, and if they know what a multi-modal RAG architecture even looks like. I've got to ask these questions not because I care about how they are fine-tuning or training their models and setting up their embedding pipelines but because I need to judge how much time they spend in the field of Deep Learning. When I ask some of the more technical questions and a founder's face lights up, links to research papers start flying; I, too, start smiling and cracking jokes, and we get to talking about why they made some of their architectural decisions. When this happens, I know at least the technical side of the company is in good hands; the founder cares; even though they may be new to AI, at least they took the time to read the journals, get up to speed, and gain knowledge about the field they are playing in, this, I respect immensely.
All this to say, while everyone is talking AI, it's still hard for me to find that many folks who know and care about actual Deep Learning; when I do find those people, I usually invest in them in the first thirty minutes and if you're one of those people i'm John@meridianstreet.co