EP06 - Quant & AI

How quantitative finance and AI use the same math. Learn about time-series prediction, transformers, DeepSeek, High-Flyer Capital, and why quant skills transfer to AI development.

Tweet coming soon

Kurumi, I'm confused. This AI... DeepSeek. It's amazing. It codes better than me.

But look who owns them! **"High-Flyer Capital."**

That's a Quant Hedge Fund!

I thought Quants were useless degen gamblers! How did they build an AI that rivals Google?

You thought Quants were "Finance Guys."

Quants are **Math Nerds** who happened to apply their skills to money first.

Do you think there is a difference between predicting a stock price and predicting the next word in a sentence?

Of course! One is money, one is... language! Shakespeare! Poetry!

To a GPU, these are the **Same Equation**.

It is **Time-Series Prediction**.

You feed in a sequence of tokens (Prices or Words), pass it through an Attention Mechanism, and minimize the Loss.

The math doesn't care about the *semantic meaning*. It cares about the **Statistical Probability**.

So... High-Flyer didn't have to hire new people?

They used the same people.

The guy who optimized the algorithm to predict Soybean Futures? He probably just tweaked the model to predict Python Code.

But Google has billions! And thousands of researchers!

How did a random Hedge Fund beat them on cost? DeepSeek is so cheap!

Because Silicon Valley runs a different business and capitalized by public markets.

When Google has a problem, they buy 10,000 more GPUs. They throw money at it.

Quants are cheap. We are obsessed with **Efficiency**.

In High-Frequency Trading (HFT), if your code is 1 microsecond too slow, you lose money.

We optimize the hell out of every memory access.

High-Flyer treated the AI Training run like a Trading System.

They optimized the communication between GPUs. They used **Mixture-of-Experts (MoE)** to only activate the necessary neurons.

They didn't out-spend Google. They **Out-Engineered** them.

So... Quants aren't just "Degen Gamblers."

Quants are **Truth Hunters**.

We build engines to find signal in noise.

Look closely. These are all the same business.

They collect data to predict: "Will Shez click this shoe ad?"

If Probability > 50%, show the ad.

That's... prediction.

They collect data to predict: "Will Shez crash her car?"

If Probability is Low, sell policy.

Also prediction...

We collect data to predict: "Will the price go up?"

But here is the difference.

AdTech sells the prediction to a Shoe Company. They need a Sales team.

Insurance sells the prediction to a Driver. They need Agents.

Quants don't deal with customers. We don't have Sales. We **Eat Our Own Cooking**.

If our prediction is right, we bet our own money and keep the profit.

That's why quants are so intense!

If Google's ad algorithm is slightly wrong, the shoe company loses money, not Google.

If your algorithm is wrong, *you* go bankrupt.

**Skin in the Game**.

That pressure creates the sharpest engineers in the world.

And that's why a quant firms like High-Flyer could build an LLM that shook the world.

I used to think Quants were wasting their talent on stocks.

But actually... stocks were just the training ground?

Stocks are the hardest dataset.

If you can find a pattern in the chaotic, adversarial, noise-filled stock market...

Finding a pattern in English grammar is **Easy**.

So the skills transfer?

Probability, Linear Algebra, Python, C++. You learn every practical skill that is useful in the real world.

So if I learn Quant trading... I'm actually learning how to build the future?

You are learning how to **Model Reality**. Whether you apply that to money, language, or biology is up to you.

B-But if they are so talented, they should be building things!

If only you knew the things they have built without the public fanfare. Deepseek isn't the only thing that came from the quant industry.

What are they building?

Follow our twitter and stay subscribed to this comic to find out in future episodes.