• AI Street
  • Posts
  • Permutable AI Teaches AI to Trade

Permutable AI Teaches AI to Trade

Hey, it's Matt. Here’s what’s up in AI + Wall Street.
Was this email forwarded to you? Sign up below:

TRADING

The Startup Training AI with Expert Traders

Humans taught AI the difference between cats and dogs. Can a trader teach AI how to trade like a human?

That's the bet Permutable AI is making. The London-based startup uses reinforcement learning, where humans judge a machine's actions and the AI learns from feedback, to create profitable trading strategies.

Founded by former Merrill Lynch and Citi fixed income trader Wilson Chan, the company began experimenting well before ChatGPT became a household name.

"Three or four years ago people thought we were crazy," Chan told me.

Unlike conventional systematic trading that relies on deterministic rules, Permutable's LLM-based system can “reason” through market conditions and “explain” its decisions.

Made with ChatGPT

The system uses reinforcement learning, where models improve themselves over time through iterative training.

Their edge, according to Chan, comes from having expert traders evaluate the AI's decisions, effectively “teaching” the machine human trading intuition.

"Put an AI in a room, ask it how to deal with certain situations and put an expert trader in the same room to assess the AI," Chan told me. "We can extract a lot of intuition and trading instinct even within a couple of hours."

This human-in-the-loop methodology mirrors techniques OpenAI used to improve its models, leveraging domain experts to refine the AI's understanding and decision-making capabilities.

Since live trading began in October, Permutable has outperformed the S&P Global Macro Commodity Index, the benchmark for commodity trading advisors. The platform earned a 26% annual return with sub-10% volatility and a negative 20% correlation to the S&P 500 during the Trump Tariff period.

Chan says they publicized their results in part because they’re looking to raise capital and to attract partners, given the significant computing costs involved. Permutable aims to expand beyond commodities trading (their initial focus due to an early oil and gas client) and scale into equities.

While Chan declined to name specific firms, he says he’s had conversations with some of the world’s largest hedge funds, who are exploring his firm’s approach.

Chan also has an open invitation: If you’re an expert trader, Permutable wants to borrow your brain—for a few hours, at least.

NEWS

Robinhood to Offer AI Stock Summaries, Trading Ideas

Robinhood is offering AI-assisted investing with the launch of several new products, following last year's acquisition of AI research platform Pluto Capital.

  • Cortex is a tool that acts like an in-app analyst, surfacing headlines, technicals, and strategy suggestions without executing trades.

  • Robinhood Strategies is an actively managed portfolio product with a flat $250 fee cap.

  • Robinhood Banking is a forthcoming private banking platform with perks like 4% APY and estate planning. (Press Release, Business Insider)

Whether this constitutes formal “financial advice” remains an open regulatory question, but many retail investors are increasingly comfortable receiving investment ideas from AI. Robinhood's approach creates an interesting hybrid: offering low-cost guidance (traditionally associated with passive investing strategies) while encouraging more active account management.

Financial Firm Gives Investment Advice in Gen Z Slang

The Wall Street Journal has a story about Arta Finance, which I’ve covered before (see below*):

Artificial intelligence is coming to the world of investment advice, and it can speak in Gen Z slang.

That is the pitch from Arta Finance, a wealth-management startup led by an ex-Google executive and backed by the former chief executive of Swiss-banking stalwart UBS. Arta is rolling out an AI assistant that can dispense financial advice in spoken conversations—and in any preferred tone and argot. Even for the 20-something millionaire set.  

“Low-key gonna break down ur investment plan rn,” the Arta assistant says, responding to a client’s query on his investment portfolio. “No cap, ur portfolio is fire!”

WSJ

The idea that important financial decisions have to be conveyed in formal business language is frankly an old media way of looking at things. (Obviously, I would say that given that I run a newsletter and also because there’s an Adam Sandler meme right below this : )

I think a lot of new tech gets treated this way, like, oh, it's new, novel, weird. It reminds me of Blockbuster’s overconfidence in physical stores and how they were convinced viewers wanted to browse in person. I mean, people buy mattresses online now!

*I spoke with one of Arta’s cofounders, Chirag Yagnik, for a Q&A back in October, about how they’re targeting folks with $500,000 to $25 million in assets.

Lloyds Bankers Head Back to School to Learn AI

Lloyds partnered with Cambridge University to train 300 senior leaders—part of a broader tech plan that includes migrating to the cloud, developing an AI agent, and hiring for tech roles. (Bloomberg)

UniCredit Uses AI to Chase Smaller M&A Deals

AI helps scale tasks cheaply, which will create new tasks and in some cases, open new business lines.

The Italian lender’s platform, DealSync, flags sub-€50M deals from across its corporate and wealth units, surfacing leads for M&A bankers that were previously missed or handled ad hoc.

So far, the tool has generated 2,000 leads and 500 mandates. It’s part of a broader three-year plan to grow client solutions revenue by €1.4B by 2027. UniCredit earns fees if DealSync successfully matches buyers and sellers, with additional upside from financing.

So yes, AI will likely curb certain work, but you still need humans to do deals.

Also: The M&A Boom Wall Street Wanted Is Here, if You Know Where to Look (WSJ)

REGULATION

Finance in 'Early Innings' of AI Adoption: SEC Forum

I had planned to write up something on the SEC’s AI investment roundtable, but there wasn’t too much actionable, which is often the case with these roundtables.

I think the industry is in the early stages of AI adoption, and the SEC is even further behind, especially given all of the recent DOGE news.

PYMNTS has a good roundup here.

WHAT ELSE I’M READING

Quantum Computer Truly Random Numbers: JPM

JPMorgan Chase & Co. has generated and certified so-called truly random numbers using a quantum computer, in a world-first that the bank hopes will have applications for security and trading. (Bloomberg)

Random numbers typically aren’t truly random. They’re created by following a set of rules known to the creator. There have been lots of cases of mathematicians hacking state lotteries because of just this issue.

AI Use Jumps in Corporate Legal Departments: Survey

Generative artificial intelligence adoption in corporate legal departments increased significantly in 2024, a new survey found. (Bloomberg Law)

Hard to see what slows this down since the law is pretty static relative to markets, so there’s less need to update AI models.

Point72 CTO Plans Tech Expansion and AI Integration

In his first interview since joining the hedge fund last September, Ilya Gaysinskiy told BI about his big plans to ramp up Point72's tech organization to meet and continue to drive the firm's expansion. Gaysinskiy formerly held a number of tech leadership positions at Goldman Sachs in core engineering and its consumer business.

Even though Gaysinskiy considers himself a strong coder, he had a begrudging realization in the last few weeks that "AI is starting to get better at this than me," he said, laughing. (Business Insider)

How did you like today's newsletter?

Login or Subscribe to participate in polls.

Reply

or to participate.