A theme that’s weirdly come up in a few unrelated private conversations with various tech foIk is how it will become very difficult to make money in the future. Like the meme that says you have 2 years to get rich because by then anything you can imagine will be solved with compute (ie electricity) and therefore capital. Labor will be worthless.
It’s a comically reductionist view but its directionality occupies a pied-à-terre in my brain. It’s a messy mental apartment. The thoughts are scattered:
- I remember when the threat of mass automation was contained to autonomous driving. “Truck drivers should learn to code”. Now it’s the hair stylists who are safe (at least until literal robots take over) and everyone with an email job looks cooked. During covid, I wrote about how it felt so unjust that high-paid workers who sit behind a screen not only kept their jobs but thrived while anyone on the front-lines of humanity got ravaged. It was a deeply regressive crisis. At least an EMP tragedy would have been progressive. It looks like the universe now wants to atone for its prior path.
- Interest rates, stocks, and home prices are all at their generational highs. But odds are if you read this letter you live in a town or city where the median homeowner affords their 7-figure mortgage with a W2 email job. If automation is deflationary are the most widely held assets entirely mispriced?
- Is the correct game theory response to loot institutions, pump bags, and discount the value of reputation? Are long-term thinking and “compounding” overbid ideas? (I don’t live as if they are because to believe this requires sociopathic nihilism but I weren’t being paranoid about being sucker I’m probably blind.)
I’m harboring all these strange notions. Then, as an example of confirmation bias in real life, Liv’s tweet interrupts my mindless scroll.
My interest in the tweet is the part which gestures at market efficiency — “better than any human at maximizing their own capital”. It echoes the theme I opened with — that acquiring substantial capital in the future will become nearly impossible against a master robot race.
(The whole part of creating a human-centric economy is really about political choices. Issues like H1B visas, maternity leave, or how the tax code treats various forms of incomes/costs are the rules upon which American capitalism rest. These are democratically determined even if markets themselves are not democratic consensus mechanisms. That’s not the subject of my musing.)
So this skynet stuff is just floating around my consciousness, then I hear something that sparks a total right turn (not a 180° to be clear). I’m driving back from Placerville last weekend (strongly recommend Marshall Gold Discovery State Historic Park!) listening to Dwarkesh’s recent interview with “two of the most important technologists ever, in any field”, Jeff Dean and Noam Shazeer.
[Some background on them:
Jeff Dean is Google’s Chief Scientist, and through 25 years at the company, has worked on basically the most transformative systems in modern computing: from MapReduce, BigTable, Tensorflow, AlphaChip, to Gemini.
Noam Shazeer invented or co-invented all the main architectures and techniques that are used for modern LLMs: from the Transformer itself, to Mixture of Experts, to Mesh Tensorflow, to Gemini and many other things.
We talk about their 25 years at Google, going from PageRank to MapReduce to the Transformer to MoEs to AlphaChip – and maybe soon to ASI.]
Most of this interview was above my head and I admit my knuckle-dragging self only got halfway through before switching to Marc Maron interview Wolfgang Van Halen but this one section caught my ear before giving up:
One of the big areas of improvement in the near future is inference time compute, applying more compute at inference time. I guess the way I like to describe it is that even a giant language model, even if you’re doing a trillion operations per token, which is more than most people are doing these days, operations cost something like 10 to the negative $18. And so you’re getting a million tokens to the dollar.
I mean compare that to a relatively cheap pastime: you go out and buy a paper book and read it, you’re paying 10,000 tokens to the dollar. Talking to a language model is like 100 times cheaper than reading a paperback. So there is a huge amount of headroom there to say, okay, if we can make this thing more expensive but smarter, because we’re 100x cheaper than reading a paperback, we’re 10,000 times cheaper than talking to a customer support agent, or a million times or more cheaper than hiring a software engineer or talking to your doctor or lawyer. Can we add computation and make it smarter?
This entire way of reducing signal to tokens and computation made me think of investment alpha. It made me think of how financial innovation*, the cycle of hunting for alpha, alpha decay, and efficiency works.
Professional investment managers with teams of salepeople pitch their funds “alpha”. They used to just say “look, how much money we made” but now they have to adjust for beta or some other benchmark. Allocators get smarter. The process is uneven but the waterline of basic aptitude does inch higher. If the market is mostly free of captive frictions then the need for return and the quest to provide it finds a clearing price.
But what does that equilibrium look like in the future? Well, probably something like the way it looks now regardless of how efficient it gets. The paradox of provable alpha is a stable attractor whereby the average person in pursuit of excess returns gets a random number distributed around a fair return so long as they avoid outright stupidity (like investing in levered ETFs for the long haul).
An efficient market feels random. Not to throw another paradox out there but this is exactly what Mauboussin refers to in “paradox of skill”. When the competition is evenly matched, luck determines the outcome. The funds and their marketers who survive all tout information ratios and charts which suggest their alpha, net of fees, is persistent. The best we can probably say about them is they might be good enough such that the claim is a coin flip. If they were deterministically better than a coin flip, they’d be inaccessible or would charge enough to skim the surplus.
But it’s getting into the realm of a coin flip that is the answer to job security. If you’re good enough to have your results look random it comes down to story-telling and dinner. Your value here might not be in delivering true alpha, but if you find a buyer you have created value — by definition you have given someone what they want.
[There’s a piercing conversation to be had here about commerce in general. If I sell snake oil or astrology is my income compensation for creating value? GDP and capitalism say yes. Is profiting from misinformation value-creation? How far does this go? The conversation gets quite “nudgy” if you focus too much on “what’s efficient” vs what tradeoffs are we willing to tolerate and for what ends. There’s a continuum.
Disagreement about object-level policy is downstream of differences in where people lie on the philosophical continuum. Nobody says “you should be allowed to pollute for a profit” but should you be allowed if that’s what people want? You’d want a polluter, the one with the upside to bear the risk, but I doubt even that anodyne statement is universally acknowledged. I could just as easily imagine an uber-industrial view that the people asking for the product should be forced to live with the pollution. Feel free to use pollution as a metaphor for FUD, artificially constrained university admission, or whatever profitable hobby horse irks you.]
If capital-compute-fueled efficiency makes it harder to create alpha or more generally “value”, it still won’t take away the types of value that are not provable. Maybe the only jobs remaining will be selling figurative placebos.
Now are they gonna pay enough to cover junior’s riding lessons?
Fck if I know. Ask a robot.
I know I’ll still be listening to Van Halen like the peasant I am. There’s probably something to that.
[Expanding on the asterisk from above]
*Financialization is an innovation because slicing up risks to sell to their highest bidder increases liquidity. An liquidity is a “good”. Liquidity reduces the cost of capital for industry. You will finance more businesses if you believe you can sell your shares. Insurance and mortgages are concrete products of the abstraction known as “liquidity”. All of this liquidity is an instance of abstractions that sit even higher in the stack — specialization, comparative advantage, and the surplus we enjoy from trading, I mean “trading” in the positive sum cooperative sense, with each other.
This harkens back to the liquidity premium discussion in Why Investing Feels Like Astrology:
Liquidity Premium
@Jesse_Livermore’s work refers to an idea he calls “transactional value”. It is the value that permits you to pay more than intrinsic value for an asset because you know you could sell it back into a liquid market.
Here’s Jesse parsing intrinsic value from transactional value:
The intrinsic value of equities would be the cash flow stream of the equities themselves, which you can collect and they belong to you and you can spend them and do whatever you want with them.
The transactional value would be the value that comes from the fact that there’s this “network of confidence” in the market, that people have been doing this for hundreds of years and we know that when you wake up tomorrow, the S&P is not going to be at 500. It’s going to be near where it was yesterday and people are kind of anchored to where its price is…You can basically take all your money, 100% of it, and put it into the stock market and know that you’ll be able to get a lot of that out anytime you need to. That’s the transactional value, which is the premium.
The idea that liquidity commands a premium is not new. If you have any money in a savings account today, you are paying a liquidity premium in the form of negative real interest rates. The treasury market discounts off-the-run securities because they are thinly traded even though they mature to the same value as their on-the-run counterparts. But I don’t want to dismiss Jesse’s notion of transactional value because it’s not novel. His expression of it is illuminating. For example, currency is made entirely of transactional value. The fact that we can rely on it to trade warrants a premium entirely out of proportion to the value of paper that represents it.