- The Breakdown
- Posts
- 🟪 Thursday Links
 | “We have only to persevere to conquer.” |
Thursday Links
Scott Alexander thinks AI superforecasters will soon be telling us who to marry.
“It will look through whatever texts and emails you give it access to,” he explains, “learn what it can about your relationship, and answer something like, ‘If you marry this person, I think there’s an 85% chance you get divorced within five years.”
Whether 85% makes it a yes or no will remain up to you.
Based on how well AIs are already performing in prediction markets, Alexander estimates that this sci-fi future is less than a year away.
AI superforecasters are LLMs that have been modified with a bespoke “scaffold” — an additional layer of software that guides the LLM through a long research process (without it, LLMs are prone to wander).
They’re already nearly as good as the very best human forecasters, Alexander reports, and should be well beyond them by this time next year.
At a conference for superforecasters (human and otherwise), Alexander met a startup founder whose AI had turned $35 into $2 million on Kalshi over seven months. Another said his AI was beating the stock market by 25%. (AI superforecasters are especially good at finance.)
Alexander expects this will soon make prediction markets unsafe for humans. Polymarket and Kalshi will be where “bots duel other bots for the privilege of collecting money from dumb sports fans.”
With luck, though, that will provide an incentive for AIs to become so good at predicting that they’re also societally useful.
“The dream,” Alexander says, “is that, armed with AI superforecasters, the public and the politicians who they elect will make better decisions about policy.”
Or who to spend your next few decades with.
If an unofficial crypto broker absconds with your money, the FBI will be happy to investigate.
They will also check your socials.
Reda Mazen Rida Sabassi, a resident of San Diego, learned this the hard way after attempting to distribute $387,000 in charitable donations he’d raised, ostensibly for the benefit of orphans in Gaza.
Prosecutors say the charity was a front and that Sabassi subsequently attempted to send the funds to a co-conspirator who the government had designated as a fundraiser for terrorists.
The funds never got there.
Knowing he could not send money directly to a designated terrorist, Sabassi enlisted an intermediary. He sent the $387,000 to unofficial crypto brokers, who promised to convert it to USDT and then send the USDT to crypto addresses provided by Sabassi’s co-conspirator (identified as CC-1 in the complaint).
When the brokers (Individuals-2,-3, and -4) decided to keep the funds instead, Sabassi reported the theft to the San Diego police, who recommended he contact the FBI.
Incredibly, he did — although not before asking CC-1 to fake an invoice for “aid trucks.”
According to prosecutors, Sabassi hoped these would “(i) substantiate his fundraisers in support of his claim that Individual-2 and Individual-3 stole funds and (ii) not disclose that the funds were intended for Hamas.”
That is a tough needle to thread!
Sabassi did not thread it. He told CC-1 he planned to preemptively send the invoices to the FBI so that they wouldn’t “come back to him with questions,” but they did — in part because of what they found on Sabassi’s socials (and then, with a subpoena, what they found on his email, phone, and everywhere else.)
Sabassi’s trial began in lower Manhattan this week. He faces as much as 85 years in prison.
A new paper merges behavioral psychology and quantitative trading to develop a novel metric for stocks: the Degree of Rejoicing and Regret (DRR).
The authors say DRR quantifies how our emotions distort market prices — specifically, the joy of picking winners and the regret of holding losers.
When a stock outperforms its peers, the paper explains, investors anticipate future joy and accept lower expected returns, causing stocks to be overpriced. Conversely, when a stock underperforms its peers, investors anticipate future regret and demand higher expected returns, causing stocks to be underpriced.
Like emotions themselves, this effect reverts to the mean.
DRR is negatively correlated to future stock returns, which should make a tradeable signal: a market-neutral long/short portfolio based solely on the metric would have returned an annualized 16.45% annualized between 1963 and 2023, according to a backtest.
16.45%! Over 60 years!!
The paper cites neurobiological research to prove these eye-popping returns really are based on investors' emotions — including brain scans that quantify the psychic benefit and mental toll of constantly comparing our investing decisions to what might have been.
It then uses behavioral theory and some fancy math to prove that our anticipation of these feelings actively distorts our buying and selling.
The authors further support their thesis by comparing DDR across countries. They find the effect is amplified in countries with either “high individualism” (US, Canada, France) or “high uncertainty avoidance” (Japan, Belgium, South Korea).
“High individualism induces overconfidence and self-attribution,” the paper says, “amplifying investors’ emotional responses to both rejoicing and regret.” Conversely, “high uncertainty avoidance leads investors to overreact to potential losses or foregone gains, hoping to mitigate uncertainty.”
Both characteristics amplify emotional trading, which distorts the valuation of stocks.
They also find an exception that proves their rule: the United Kingdom.
In the UK stock market, the DRR metric behaves “atypically,” exhibiting a strong positive relationship with future returns (ie, the opposite of everywhere else).
British stoicism appears to act as an emotional shock absorber for the stock market, keeping stocks from falling as hard, or rallying as high, as the underlying fundamentals would suggest.
Let’s hope it works for their football fans, too.