Building Intuition // BrXnd Dispatch vol. 009
Thoughts on how I'm going about trying to understand what's happening in AI.
Hi everyone. Welcome to the BrXnd Dispatch: your bi-weekly dose of ideas at the intersection of brands and AI. I hope you enjoy it and, as always, send good stuff my way. NYC spring 2023 Brand X AI conference planning is in full effect (targeting mid-May🤞🤞). If you are interested in speaking or sponsoring, please be in touch.
A few weeks I was on a call with a friend talking about … AI. One of the things he said that really struck me was that he had been talking to a bunch of folks he knew who built during the original web explosion and asking them how they built intuition about the then-emerging technology. To his eyes, this explosion looked a lot like what happened in the 90s, and he was curious if there were any tips he could glean from people who had successfully navigated Web 1.0.
In the world of computer science, there’s a concept called explore/exploit. It’s essentially a question about diminishing returns: how long should you let a system search for the optimal solution before making a decision? As Brian Christensen puts it in his excellent book Algorithms to Live By, “exploration is gathering information, and exploitation is using the information you have to get a known good result.” Here’s a deeper explanation from the book:
In computer science, the tension between exploration and exploitation takes its most concrete form in a scenario called the “multi-armed bandit problem.” The odd name comes from the colloquial term for a casino slot machine, the “one-armed bandit.” Imagine walking into a casino full of different slot machines, each one with its own odds of a payoff. The rub, of course, is that you aren’t told those odds in advance: until you start playing, you won’t have any idea which machines are the most lucrative (“loose,” as slot-machine aficionados call it) and which ones are just money sinks. Naturally, you’re interested in maximizing your total winnings. And it’s clear that this is going to involve some combination of pulling the arms on different machines to test them out (exploring), and favoring the most promising machines you’ve found (exploiting).
One solution to this problem is called “Win-Stay, Lose-Shift,” in which you stick with the machines that pay out and jump away from the ones that don’t. The specifics don’t really matter for our purposes. The point is that when you face a scenario like a room full of slot machines or a new technology ripe to change the way work happens, you’ve got to decide how to spend your time and how much to devote to going wide before you go deep. (As an interesting aside, there’s actually a specific answer to this question, which is 37%: if you have a time dimension, like a month to move, you should spend 37% of it exploring and then make a decision.)
And so, in this moment of exploration, how do you build intuition (or what my friend called “finger feel”) for this new technology? While I certainly can’t answer it for you, I attempted to answer it for myself last weekend as part of a seminar I taught at the University of Montana. My talk was about what I’m doing to wrap my head around what’s going on and what I think is likely to happen in the future.
Support From
A big thank you to LinkedIn and Redscout for sponsoring the upcoming BrXnd conference and my work. If you would like to support us (and see your logo here), I have various sponsor levels available for the newsletter and event. Be in touch.
Again, this might not work for everyone, but I think my general approach looks something like the pyramid below:
Foundational Beliefs: recognize and understand my own beliefs and biases (priors in Bayesian terms) and consider how strongly I hold them.
Approaches to Understanding: work through a few specific techniques (none overly complicated) to build an intuitive understanding of this new technology.
Execution: This is the exploit step. For now, I’m not too worried about that.
I’m going to skip over foundational beliefs for now, other than to say they’ve fundamentally wrapped up in some thoughts I have around technology cycles, media, and complexity. So let’s focus on approaches to understanding. Again, these aren’t rocket science, but faced with some very new ideas, I did my best to articulate to the class how I am going about trying to build what my friend called “finger feel” for AI.
Tinker
Tinker takes a few forms. I’m not a mathematician or a computer scientist, I’m a self-taught coder who likes to read and explore. My fundamental understanding of the math and science behind computers and machine learning is pretty shallow. But that doesn’t mean I am not trying to build that ground-level understanding of the tech. For example, a few years ago, excited after reading Andrej Karpathy’s blog post “The Unreasonable Effectiveness of Recurrent Neural Networks” I tried to train my own model. Or, more recently, watching Karpathy’s excellent Youtube video “Let's build GPT: from scratch, in code, spelled out.” Having some understanding, even a basic one, of what’s going on behind the scenes will only help in building up the kind of intuition that’s helpful in finding ways to use this technology to solve problems.
On the other side of tinkering is lots of the experiments I’ve been building. It’s a lot of following my nose, trying new things, and writing lots of code. I created my CollXbs experiment, which led me to ask what these large language models know about brands. That question is now leading to another experiment I hope to release soon that attempts to answer that question (if anyone wants a sneak peek, reply to this email or leave a comment). Beyond that, I’m constantly integrating LLMs (most often GPT3 via its API) into my work. That’s how I discovered my killer prompt and have generally built up my current understanding of what’s real and not in the world of AI. (I need to write more about this, but my headline is that the focus on creative output is a bit of a head fake for the current transformational use cases in data extraction and summarization. Put simply, I think Microsoft cares a lot more about LLMs in Excel than Bing.)
Nassim Taleb has an interesting take on tinkering in Black Swan, arguing that it is central to discovering fundamentally new ideas. “The strategy for the discoverers and entrepreneurs is to rely less on top-down planning,” he explains, “and focus on maximum tinkering and recognizing opportunities when they present themselves.” That’s certainly my experience as both a tinkerer and entrepreneur.
Follow Doers, Not Talkers
One of the finest compliments someone paid to this newsletter is that they appreciated that it was about building things, not just talking about things. That has been my objective, as there is no shortage of people talking about this stuff. It’s too cheap to get your hands dirty with this technology to pay attention to those who aren’t in the weeds, and I find myself most interested in what the researchers and hackers are doing rather than the talking heads.
To that end, I just marked my AI Twitter List public. It’s full of people who are pretty deep in the space. Of course, it’s far from perfect, and I’m sure I’m missing many good people (feel free to leave comments or reply with suggestions), but I still find Twitter is the best way to get a direct brain feed of people doing interesting work in a space.
Beyond that, I’m reading lots of papers and watching videos, like this great talk by Melanie Mitchell, with people who have depth in the space I know I’m missing.
Wrapping Up
As I said, this is my approach to answering the problem of building intuition. It’s not complicated, it might not work for you, but I hope it’s at least a valuable set of guideposts on the path to understanding.
New BrXndscape Companies
New companies listed on BrXndscape, a landscape of marketing AI companies (writeup in case you missed it). If I missed anything, feel free to reply or add a company. (The companies are hand-picked, but the descriptions are AI-generated—part of an automated pipeline that grabs pricing, features, and use cases from each company’s website and one of many experiments I’ve got running at the moment.)
[Email Generation] SellScale: SellScale AI is a powerful AI-powered outreach platform that helps you generate, send, and tune outbound that converts. It integrates with common sales tools to automatically populate personalizations, and its AI learning capabilities help you optimize personalizations to the persona you're reaching out to. It also adapts outbound to your brand, industry, and personas, so you can reach the right people with the right message.
[Image Generation] Accomplice: Accomplice's AI-powered platform helps your team generate 100% royalty-free logos, photos, and graphics while saving time, cutting costs, and simplifying your workflow. It offers image generation, editing, training, upscaling, and organizing.
[Colors] ColorMagic: ColorMagic is a color palette generator with AI. Generate colors from keywords.
[Interface Design] Galileo AI: Galileo AI is an AI-driven interface design platform that enables users to generate high-fidelity UI designs from text prompts quickly. It also offers features such as AI-generated illustrations and images, full product copy, and more time for a bigger impact.
Thanks for reading. If you want to continue the conversation, feel free to reply, comment, or join us on Discord. Also, please share this email with others you think would find it interesting.
— Noah