Imagine a machine that lives high up in the sky. The machine has preternatural ability. It can see the future — wars, weather, and presidents. Its intelligence is ooms greater than any living being. And its power so awesome it’s sometimes seen as a sort of oracle or magic genie. You might be picturing the GPT-5 cluster, but that’s not it. The machine I’m talking about is the market.
The market is the synthesis of every human action. It prices in the day you take your first and last breath, what you eat in between, and even how often you sh!t. We can’t escape it. Every day, 8.2 billion people wake up and live within it. Some buy diapers, others invent vaccines, and a special few buy and sell cryptocurrencies. But all of us feed the market. It ingests our reality, analyzes it, and reflects the updated state of the world.
The market can be thought of as a form of general intelligence. And although there are key differences, the similarities between markets and AI models like GPT-4 are striking. Both train on vast amounts of human-generated data, both compress incredible sums of knowledge, and both implicitly mirror our world back to us.
But despite their similarities, humans use AI and markets quite differently. With AI, we have sci-fi-like expectations. Many talk breathlessly about the day it will take our jobs and bring forth utopia. But when it comes to markets, our views are more muted. We don’t expect them to do much other than price assets. And we certainly don’t expect them to replace our jobs or institutions. This is weird. If markets are a form of general intelligence, why don’t we ask them to do more?
When are people going to realize that capitalism is already absolutely nothing short of an advanced artificial general intelligence? I think it’s one of those stylized empirical facts that is too large to do anything with, so it becomes like a secret.
— Justin Murphy (@jmrphy) December 13, 2020
Our shallow relationship with markets is even more perplexing when you consider how quickly we’ve adopted SOTA AI models like GPT and Claude. These technologies are only a few years old and already more widely applied than markets, which have existed for millennia. Perhaps markets can learn something by looking at where AI has found initial PMF.
Today’s AI models are primitive yet powerful. They memorized the Internet but struggle to count the r’s in “strawberry.” But despite these limitations, they can do incredible things. And we take full advantage by putting them to work across society. There are many compelling use cases, but at the highest level, we can group them into two big buckets:
- answers
- actions
ChatGPT is popular because it’s an answer machine. You ask the model a question — any question — and you get an answer. The models know all sorts of useful information, so we often look to them for knowledge. Who won the US Open? When is Election Day? How long is this damn report? AI has essentially replaced Search and entire swaths of the Internet. But its impact has been surprisingly limited.
Imagine telling someone a century ago “we have a system that correctly translates things 99% of the time, informal speak and tone and slang and all, usable for free, all the time, and it didn’t really change anything”
— Jason Phang (@zhansheng) June 17, 2024
The problem is that AI is passive. And we need agentic AI before we can start climbing the Kardashev scale. Today’s models can take some action, but only in limited ways. For instance, Claude has proven popular with developers because it’s adept with code. You can ask it to do all kinds of stuff — generate code, build a website, write documentation, etc. This moves AI beyond the friendly chatbot paradigm to something closer to an agent.
Ok mind == blown
This afternoon I managed to build a complete exchange UI in @nextjs with practically zero frontend experience using a combo of @cursor_ai + claude + @v0. It even supports @solana wallets out of the box + fetches rpc data via our own SDK.
It’s so over for devs. pic.twitter.com/2nuc9PW4Kg
— Tristan (@Tristan0x) September 1, 2024
Our relationship with markets is far more primitive. We ask them few questions and give them no ability to let them take action. Instead, we tend to view markets as fickle slot machines and often base our opinion of them on the answers they give us.
The rules of CT are pretty simple
If the token goes up, it gets threaded, thought-pieced, philosophized, twitter-spaced, celebrated, vc-funded, future-of-all-x
If the token goes down, it’s a scam, a rug, dead, inversebrah’d, ratioed, canceled, and everyone must go to jail
— vibhu (@vibhu) May 31, 2024
There are many possible explanations for why we’re dramatically underutilizing markets. But the most obvious is we lack the tools to use markets in a more generalized way. Err, at least we used to lack the tools.
Prediction markets and futarchy are two new tools that leverage the general intelligence of markets. They let us pose questions, extract answers, and give markets agentic power. It took ChatGPT to reveal the power of AI, and it will take these tools to do the same for markets. The entire crypto movement is an experiment in free markets. First, it was money, then the Internet, and now society itself. A golden age of markets lies ahead.
Truth Machines
If ChatGPT is an answer machine, then Polymarket is a truth machine. ChatGPT compresses the Internet to give us answers, while Polymarket does the same with markets. The only difference between the two is trust. AI is an alien life form made of silicon, not flesh. All we can do is trust that the high priests at OpenAI solve alignment and keep our best interests in mind. But markets are different — they are made up of human beings. They are us. There’s no need for alignment when it comes to markets. At any given time, they already reflect consensus across all humans — a living, breathing truth, if you will.
This lofty concept of truth is the biggest reason prediction markets have exploded in popularity. People trust these markets more than other information sources. Don’t get me wrong, markets are imperfect. After all, they are made up of a bunch of flawed humans. So, perhaps the “truth machine” framing oversells things a bit. But it’s kinda catchy, so we’ll rock with it. A more accurate moniker might be information discovery machines. But when you really think about it, “information discovery” is our best attempt at finding truth.
It’s not like markets have a whole lotta competition in the truth department. For decades, public trust in institutions has been downonly. And it’s not just the media; almost every institution you can think of — the church, government, and even the scientific community — is staring down record-low levels of trust.
After a decade of VC threads, Vitalik blog posts, and Reddit-coded obscurity, prediction markets finally burst into the public consciousness this year thanks to the Presidential Election. The attention the election is generating has meant big business for Polymarket. You can literally see when its “Presidential Election Winner” market went live.
2024 will be remembered as the first time prediction markets played a significant role in an election. From the jump, markets were early, contrarian, and right. Polymarket spun up a market on whether Biden would drop out on Sept. 21, 2023 — over one year ago, today. Even back then, traders were pricing in a 22% chance he would drop out — much higher than the conventional wisdom inside the Beltway and roughly 3x the odds that Trump would drop out.
The Biden market wasn’t the only one that was proven right. The premier Presidential Election Winner — the most liquid prediction market in the world — was also prescient. The real action started in May. The race was tight, and most polls gave Trump a slight edge. But in May, Biden’s odds began to roll over. At the time, this puzzled outside observers. Nothing had happened. The polls and the intelligentsia were all saying the race was still close. But the markets were flashing red for President Boden.
As summer dragged on, Biden’s Polymarket odds began to tumble. Traders were straight up ignoring polls that showed an unchanged race. By the time the Trump-Biden debate rolled around on June 27th, markets were pricing Biden’s odds at 33%, while the polls had him at 45%, or about even with Trump.
We all remember how that debate went down. When it started, Biden’s odds of winning the Presidency were chilling around 33%. But by the time everyone woke up the next morning, his odds had nuked to 20%. A few days later, they found a temporary bottom at 9%.
Now that some time has passed, we might be tempted to gloss over these numbers as just another statistic. But it’s worth lingering for a second on their significance. After all, this is Jeo Boden we’re talking about — President of the USA! At the time, he was the unanimous Democratic nominee. Yet Polymarket gave him a measly 9% chance of winning! Talk about speaking truth to power…
In the days after the debate, the Biden campaign played damage control. There was a coordinated effort from the White House and other top Democratic officials to meme “Biden isn’t dropping out” psyops all over the airwaves. And at the time, it was pretty effective. It really felt like Biden was ready to go down with the ship for a hot second. At least, that’s what Congress believed.
This effort ultimately culminated in a tweet from Boden himself, err, a top aide, doubling down on staying in the race.
Let me say this as clearly as I can:
I’m the sitting President of the United States.
I’m the nominee of the Democratic party.
I’m staying in the race.
— Joe Biden (@JoeBiden) July 5, 2024
Yet the markets were unphased. Polymarket stared down the onslaught of pro-Biden headlines and still priced in 66% odds that he would drop out. To state the obvious, this was a big F-ing deal. Here was a magic Internet money website that most normal people had never heard of openly defying the word of a sitting President. Of course, the press ate it up and ran with the story, amplifying the narrative that Polymarket had birthed and heaping further pressure onto President Biden.
By the time Biden stepped down on July 21, 2024, the markets were already pricing Kamala Harris’s chances of winning the presidency at nearly 3x higher than Biden’s. From start to end of the whole Biden saga, Polymarket was early, contrarian, and right. The market gods could not have scripted a better coming-out story for prediction markets.
Market Minutia
Now that we’ve done some proper prediction market bullposting, it’s time to steelman the bear case and discuss flaws in the current model. But first, let’s address the elephant in the room. What will happen after the election? This year, a whopping 78% of Polymarket volume has been election-related.
It doesn’t take a genius to figure out that Polymarket activity will probably fade after November 5. Duh. But that doesn’t mean things won’t get even crazier first. It probably will. In 2020, over 40% of all election-related volume was traded in the last week before the vote. Attention on these markets will go up before it goes down — especially if Polymarket drops a token as rumors suggest it might.
The more fundamental challenges prediction markets face boil down to three thangs:
- liquidity
- oracles
- opportunity cost
Liquidity is by far the biggest issue. When markets are illiquid, they are inefficient
Read the full report
This report is part of Delphi Pro.
- 800+ Pro reports across every major sector
- Talk directly with our analysts
- Private community of funds and builders
Already a Pro member? Log in
0 Comments