Hi, I'm Matt. Welcome to AI Street, where AI meets Wall Street. Every Thursday, I share curated news, analysis, and expert interviews.

The Rundown 

• AI’s Rapid Uptake 🚀 

• Five Minutes w/ Snowflake’s Jonathan Regenstein

• Aiera’s AI Financial Leaderboard

• SEC on AI’s Systemic Risk

• Scalapay, Desia, Mako Fundraising

• AI Bests Currency Traders

• Roundup

• 🇺🇸 Matt in Milan 🇮🇹 Kitchen Sink Edition

ADOPTION
AI’s Rapid Uptake

I’ve previously touched on how quickly AI adoption is happening, but it bears repeating. From this Fed Study:

1. Generative AI: 39.4% adoption after two years

2. Internet: About 20% adoption after two years

3. PCs: About 20% adoption after three years

More than half of workers in finance, information, and real estate are already using generative AI at work - the highest adoption rate across all industries, according to economists from the Federal Reserve Bank of St. Louis.

And unlike your augmented reality headsets, which are probably in some closet somewhere (like mine), 22% of finance professionals are using it every day plus another 23% using it weekly.

Widespread Adoption Across Industries and Job Functions

Usage extends beyond tech-centric roles, with 28% of employed respondents using it at work. Notably, 22% of workers in "blue collar" jobs report using generative AI.

Potential for Significant Labor Productivity Gains

The study suggests AI could potentially boost overall labor productivity by up to 0.875%, which represents a significant impact on the U.S. economy given current levels of AI adoption in the workplace

Demographics Mirror Previous Tech Waves

Similar to previous technological shifts, generative AI adoption is higher among younger, more educated, and higher-income workers.

• College graduates are twice as likely to use generative AI at work (40%) compared to those without a degree (20%).

Big Picture
Unlike other much-hyped technologies — looking at you crypto 👀 — hard to see what slows Gen AI adoption.

Also:
Goldman interns embrace AI—confident it won’t steal their jobs (Yahoo/ Fortune)

INTERVIEW
Five Minutes with Snowflake's Jonathan Regenstein

A lot of business data is organized a bit like sticky notes — all over the place. But large language models (LLMs) can help turn this disparate data into organized index cards. As AI reshapes the financial landscape, firms are grappling with how to leverage the technology effectively. 

To learn about the current state of AI adoption, I spoke with Jonathan Regenstein, head of financial services AI at Snowflake. The AI data cloud company has become central in helping financial institutions manage and analyze their data. 

Regenstein works with financial institutions across banking, asset management, insurance, and payments to develop and implement AI strategies. In this interview, Regenstein shares insights on:

• The shift from structured to unstructured data in financial services

• How AI is changing decision-making processes in the industry

• The importance of robust data strategies in successful AI implementation

• The potential for AI to unlock value in previously untapped data sources

Regenstein is a former attorney and self-taught coder who has co-authored books on working with macroeconomic data using R and Python. His latest book is “A Practical Guide to Macroeconomic Data with Python" coauthored with Georgia Tech Prof. Sudheer Chava and Texas State Prof. Emmanuel Alanis. 

Hope you enjoy our chat!

This interview has been edited for clarity and length.

How are quants using LLMs?

Quants are finding big value in using LLMs to analyze and create numeric factors from textual data at scale. Once you've created those numeric features, they become an input to your quant model. Now you have something that's back-testable.

It's not a question of whether the model hallucinated, but whether it added anything to your model. If it did, then it's probably valuable. This ability to create numeric factors from unstructured data is really where quants are seeing the most value right now.

Read the full interview here

BENCHMARKING
AI Leaderboard Shows Top Models for Financial Analysis

Determining the “best” LLM for financial tasks is hard. Take earnings summaries: there are multiple ways to convey the same information, and LLMs must make a lot of judgment calls.

Imagine this basketball game:

• “The Wildcats, who once had a 15-point lead, now find themselves in a deadlock as the surging Tigers add more pressure.”

• “After building a commanding 15-point advantage, the Wildcats have lost their edge, allowing the resurgent Tigers to claw back and tie the game.”

Both convey roughly the same outcome but with different emphasis.

General LLM benchmarks are based on a small set of standard tasks – not particularly telling for narrow financial tasks.

To help analysts determine which model is best for specific tasks, Aiera, a company using LLMs to derive financial insights, has publicly released its internal leaderboard. Aiera tested the models’ ability to handle real-world financial scenarios using sample problems and answers.

“Specialized systems require task-based evaluation,” Jacqueline Garrahan, lead author and Senior Machine Learning Engineer at Aiera, tells me. Here are the results:

Aiera Score

• Claude-3.5-sonnet (Anthropic) had the best overall performance, demonstrating strong capabilities across various financial tasks.

Sentiment Analysis

• gpt-4o-mini (OpenAI) was the most effective at determining the sentiment of financial statements.

Summary Generation

• Claude-3.5-sonnet excelled at summarizing earnings call discussions, capturing key points and details effectively.

Financial Calculations (Q&A)

• Claude-3.5-sonnet was the most accurate in solving numerical problems based on financial documents.

Speaker Identification

• Claude-3.5-sonnet was also the top model for identifying speakers in business meetings.

Learn more about how Aiera tested the models here.

REGULATION
AI Concentration Poses Systemic Risk: SEC's Gensler

SEC Chair Gary Gensler warned of potential systemic risks arising from widespread reliance on a small number of AI models and data sources in finance in a YouTube video last week.

Gensler suggests that the concentration of AI capabilities among a few tech giants, combined with the financial sector's growing dependence on these technologies, could lead to cascading problems across thousands of institutions if a central AI system fails or provides incorrect information.

AI Startup Scammed Millions from Investors: Feds

The SEC has charged the former CEO of AI startup SKAEL with fraud for allegedly deceiving investors to raise over $30 million through false revenue claims and forged documents, according to the regulator.

This kind of case is pretty low-hanging fruit for the agency. More sophisticated cases are likely in the works, as Skadden’s Dan Michael told me last week.

Bank of England to Form AI consortium

The Bank of England is seeking financial firms to join its AI oversight consortium for the sector. (Bank of England)

FUNDRAISING
Scalapay Co-founder Raises $3.3M for AI Platform

Desia, an AI-powered platform for investment professionals, raised $3.3 million in pre-seed funding led by Dig Ventures, with participation from 2100 Ventures, Exor Ventures, and Octopus Ventures. Co-founded by Raffaele Terrone, the startup aims to automate data analysis, saving significant time and resources in investment decision-making.

Terrone is known for co-founding Italian unicorn Scalapay. The new capital will support Desia’s growth in the UK, US, and Europe, as it seeks to transform the financial industry by harnessing AI to streamline investment workflows. (Tech Funding News)

French AI Startup Dotfile Raises €6M for Compliance

Summary: Dotfile, a French RegTech company, has secured €6 million in funding to expand its AI-powered compliance automation platform for financial institutions, aiming to streamline customer onboarding and regulatory compliance processes. (Tech Funding News)

AI Startup Raises $1.55M for AI PE Associate

Mako AI, a fintech startup cofounded by a former Bain consultant, raised $1.55 million in seed funding from Khosla Ventures for a platform designed to automate data analysis and reporting tasks typically performed by junior private equity associates. (Business Insider)

TRADING
AI Model Is Better at Pricing Currencies Than Humans, ING Says

ING Groep NV is starting to use AI to price currencies, replacing a job traditionally performed manually by the bank’s traders.

The Dutch lender’s new AI model employs “reinforcement learning” — a technique that mimics the trial-and-error process humans use — to make pricing decisions to keep up with market volatility, said Simon Bevan, its global head of electronic trading, in an interview. That was previously a crucial yet time-consuming task for its trading team in London.

ROUNDUP
AI Market Will Surge to Near $1 Trillion by 2027, Bain Says

The global market for AI-related products is ballooning and will hit as much as $990 billion in 2027, as the technology’s quick adoption disrupts companies and economies, Bain & Co. said. (Bloomberg)

AI Stock-Picking Chatbot Approved by Israeli Regulator  

Bridgewise, an Israeli startup, has received approval from the Israel Securities Authority to launch “Bridget,” a chatbot offering stock-buying and selling advice in partnership with Israel Discount Bank. (Bloomberg)

Axyon AI makes UK-based hire to drive global growth

Axyon AI, a fintech provider of artificial intelligence to the European asset management industry, has appointed Nicholas Greenland as senior vice president, business development. (Funds Europe)

🇺🇸 Matt in Milan 🇮🇹 
BYOK

For those new here, I moved to Milan in July after 15+ years in NYC. 

I always thought that the phrase “everything but the kitchen sink” was just a saying. But after moving to Italy, I learned that you can rent an apartment that has no kitchen — nevermind the sink. You get a room where you have to install a kitchen. In fact, you can actually move with your kitchen. So, in short, I’m shopping for a kitchen. 😆 

See you next week! 🚰