November 2022 BenGoldhaber(.com) Newsletter
Playing with Fire
In college a friend would make fun of me for how frequently I remarked ‘wow I can’t believe summer is nearly here’. I would chronically fret over the fact that time felt like it was moving fast, that we only had a certain number of days left until the end of the quarter, until graduation, etc.
It’s been on my mind because November was absolutely a month where time was very Jeremey Bearimy shaped - it seemed to ebb and flow in unpredictable ways, somewhat correlated with the degree that I was doomscrolling twitter for election and finance news.
I’m grateful that relative to college age Ben, I’m more at peace with the fact that time is not quite the constant that I expect it to be. Sometimes you need to constantly be refreshing twitter to see what absurd story is going to drop next; if that causes you to wake up, Rip Van Winkle style, and realize all of a sudden its’s December, well so be it.
However, I do still think it’s important to be aware of and mark the passage of time. That’s one of the best parts of the holiday season - they serve as checkpoints for reflection on the year.
It’s in that spirit that I cordially invite you to virtually attend my seventh annual “Burn all your Regrets in a Fire” festival. As per the ancient tradition, laid down in the long forgotten past of 2016, you are encouraged to write down anything you regret and they will be cast, with great force and most likely inside a piñata, into a bonfire. This year I’m doing things a bit differently, where we’re also going to include hopes or aspirations for the new year. Those will be… burned in a separate fire? Turned into origami birds and thrown into the ocean, maybe on fire? TBD, but my intention is that we all welcome the new year free of regrets and full of hope, and I know of no better way to make that happen, than through the magic of a large open-air fire.
A lot of my newsletter links have been related to developments in AI; the past two years have been exciting ones for that field, to say the least. Of course, if you’re researching a particular area of tech or business, it’s important to occasionally ask the question - ‘if you know so much, why aren’t you rich’? AKA, do you know things that the market doesn’t know, and if so, idk, maybe try to make money from it? I like this as a general forcing function to refine beliefs into testable models, even if there’s a lot of types of knowledge that can’t be immediately turned into cold hard cash.
When it comes to AI, I’d make several claims:
There have been significant advancements in the field.
In the next three years these will have real world economic impact.
Over the next decade it will have significant real world impact.
It’s unlikely there will be another AI winter.
This doesn’t seem priced in, probably for similar reasons why COVID wasn’t priced in (they’re events that were outside the distribution used to train the market, to borrow an ML term).
If you think this, what kind of portfolio would you create to operationalize those views? There have been a few theses that I’ve considered, split into three categories:
General Market Changes: AI as a transformative technology would have broad effects on the market, but it might be hard to predict individual winners and losers. Instead, pursue a strategy that bets on overall trends like software sectors over performing, or a general increase in volatility as capital moves to new higher productivity sectors.
Investing directly in FAANG: DeepMind continues to be at the forefront of AI, and is owned by Alphabet. OpenAI is not a public company, but given the significant share of it owned by Microsoft exposure to Microsoft ~= OA. Apple and Meta have their own AI plays that are good. It’s not clear who is going to ‘win’, but I’d be very surprised if, in particular, Alphabet/Microsoft were not represented in some fashion.
Emerging Tech Companies: AI will enable a new class of startups that will upend existing tech companies, and even if they’re already public companies they’re not yet ‘enterprise companies’ - i.e. Tesla has room to grow. This is what I round off the ARK Invest thesis to.
While I have more specific thoughts and considerations on all three of these, at a high level I’d say I most lean towards 2, as I expect strong winner take all dynamics in AI.
The literature on how to construct a portfolio is, uh, vast, and in this case there are other considerations outside of strict economics that might motivate you when considering AI investments, but if you're considering this or similar ideas I'd love to hear how you're approaching it (especially because I know that compared to some of you sophisticated, beautiful people, I am a humble LARPer).
Of course, I mostly intend to keep my money in the tried and true store of value that is my 90s Beanie Baby collection.
Part of this reflection was spurred because I recently read the Burry Case Studies, a collection of posts written by Michael Burry in the early 2000s that are extremely ‘lucid’, and are motivating for my own efforts to think clearly about finance stuff. Thanks to Connor F. for the link.
Cicero: An AI playing at a human-level in the game of Diplomacy. Zvi, a professional game designer and player, has a good analysis of it, though I think he undersells the achievement a bit. In particular I think it’s less important that the AI would likely lose if the other human players knew it was AI; what’s important is the speed at which an arbitrary domain can be mapped to techniques that AI is very good at, and the set of these domains is expanding fast. Noam Brown’s team, after making a superhuman poker bot, went on in ~three years time to making a human level bot that can play and win a game relying on coordination and deception, all using natural language.
Self Driving cars took a step forward with the public launch of Waymo in SF. I’ll be visiting SF in a few days, I’m hoping to try one out! Related: George Hotz, noted hacker, joined Twitter, and it reminded me of his other venture, creating a DIY self-driving car company. A comment on a Diff post had some great details on the business model and the extreme ‘pared down’ nature of the company:
Comma AI is an incredible example of capital discipline. With ~15 engineers and $8.1 million raised capital, it's managed to win first-place in Consumer Reports' rankings of autonomous driving systems. This is partly due to their superior end-to-end learned architecture, yet also due to their superior business model… by restricting himself to not raising vast amounts of money, George Hotz forced himself to focus on figuring out the "sine qua non" of autonomy: an intelligent agent that (like the human brain) doesn't require perfect maps and perfect vision to drive safely.
Potluck: Ink and Switch is a cool industrial R&D lab that experiments with novel Human-Computer Interface design. Potluck is an experiment in dynamic documents, where the free form document is treated as the primary artifact of the users experience, and it’s gradually enhanced with dynamic annotations that bridge the gap between document and app.
I really like this idea, and they have a beautiful, operational prototype you can play with. It jives with an idea I’ve been toying with of ‘Discordbots’ as the best form factor for interacting with AI models. I've been experimenting with a personal discord server where I create bots and channels and using them for 'augmented' work. For example, a channel where I can make external calls to a GPT-3 bot to rewrite a sentence, a channel for creating Stable Diffusion calls, or a channel where a bot monitors price information on a stock and sends alerts.
I think the key similarity is the importance of creating an 'environment for thought' where you can interact with application logic in a playful manner. When I’m writing a doc or chatting with someone, I’m not worried that I’m going to ‘do something wrong’ that will break everything or make the app mad at me. It’s quite freeing!
Enovid: A nitric oxide spray that has been used to prevent viral replication of COVID, and seems to also be effective against the Common Cold, from some reports cutting it down to 1-2 days. It would be truly *amazing* if we found a cure for the cold.
The scale and revenue potential of streamers and YouTube channels is pretty shocking, to me at least. If the numbers are accurate - and other numbers floated in the below twitter thread include 10m+ for a streamer in the top 200 on Twitch, my intuitive estimates of the profitability of video content are wayyy off.
Related, I have noticed that the boom time of VC funded lifestyles has ended, and maybe the crunch will be coming for many parts of the Millenial lifestyle.
Mutual Contempt: Another book about LBJ, this time on his and Bobby Kennedy’s contentious relationship. Perhaps this is in part an artifact of the type of the writers drawn to LBJ, but they all seem to be of, or adding credence to, the ‘Great Man Theory’. Quite a lot of US policy and politics in the 1960s seemed contingent on the personalities and personal grievances of a handful of people. The Caro books are superior, but this was still a good read with a lot of color on the Kennedys that I didn’t know.
Midnights: Powerful. Taylor does it again. My current favorites are Midnight Rain, Vigilante Sh*t, and Bejeweled.
Andor: This was, somewhat unexpectedly, very good. A ‘gritty’ take on the early days of the Star Wars Rebellion. The first episode was slow, but it picks up. Also, Disney is insanely good at making you develop emotional attachments to robot minor characters.
Disclaimer from the BenGoldhaber(.com) Risk Management team: If you see anyone selling these, perform a citizen’s arrest.
There can be any number of barriers preventing a good idea from being profitable. There was an anecdote from some biography I can’t find now about an economist working in the White House who learned about a great regulatory arbitrage involving importing steel cheap, and he excitedly told a friend, who pointed out it would require him to more or less start a smelting company, which might pose other challenges.
Related: Gwern on the general difficulty and uncertainty of turning a good idea into a profitable venture.
The right moment cannot be known exactly in advance, so attempts to forecast will typically be off by years or worse. For many claims, there is no way to invest in an idea except by going all in and launching a company, resulting in extreme variance in outcomes, even when the idea is good and the forecasts correct about the (eventual) outcome.
There might be a downturn in investment in AI, but I don’t think we’ll see a flight of human capital like what happened in the 50s and 80s.
AI seems likely to be a disruptive technology, potentially leading to job losses in certain sectors, or worse. On the margin I don’t think an individual public investor will move enough capital to affect things, but depending on your level of investment I think it’s reasonable to have this concern, in particular for investments in startups and emerging tech.
I specifically think Discord is good for this. Maybe it's the more 'personal', low-key vibe of it, where it feels like I can spin up a channel at no cost and play with it, and then get rid of it, that makes it feel 'safer' to play with compared to Slack's cold polished salesforce professionalism.