New York, NY – May 8, 2024 – In a shocking development that has the finance world buzzing, researchers at JPMorgan Chase have demonstrated their advanced AI system is capable of passing the notoriously difficult Chartered Financial Analyst (CFA) Level II examination – a grueling test that stumps scores of gifted human candidates each year.
But in a twist that has upended conventional wisdom, the AI didn’t just squeak by – it achieved a superior score to the average human test-taker by leveraging an unexpected capability: coding.
The groundbreaking study was led by Dr. Leonidas Tam, a pioneering AI researcher and Stanford-trained computational physicist at Amicus AI. According to Tam, integrating basic coding skills with a powerful language model like GPT-4 allowed the AI to crack fiendishly complex case studies that had previously been its Achilles’ heel.
“We were dumbfounded at the results,” admits Tam. “By giving the AI access to a Python coding environment through Google’s ReAct framework, its performance on Level II case studies spiked by 3 percentage points – taking it from a borderline passing score to truly elite territory.”
The JPMorgan team had previously established that GPT-4 could likely pass both the Level I and Level II CFA exams when prompted with advanced techniques like Chain-of-Thought reasoning and Few-Shot learning from sample solutions. But its struggles on Level II case studies, which require analyzing realistic scenarios involving sophisticated financial concepts, seemed impossible to overcome without laborious manual coding of analytical models.
“We thought there was no way an AI could master cases without explicit programming of things like Monte Carlo simulations and discounted cash flow modeling,” says Tam. “The fact that introducing basic coding ability through a simple REPL was enough to vault the latest models over that barrier is frankly astounding. It calls into question what fundamental boundaries might be left for sufficiently advanced language models.”
The accomplishment has sent shockwaves through the $3 trillion USD wealth management industry, where top-scoring CFA charterholders have long been revered for their ability to synthesize complex financial data into cogent recommendations. With Tam’s AI now demonstrating superior skills, executives are left pondering an existential question: have language models officially emblematized the singularity – rendering entire classes of white-collar analysts obsolete?
The accomplishment sends shockwaves through the $3 trillion wealth management industry. As Julia Bonafede, CFA, Corey Cook, CFA, and Glenn Doggett, CFA noted, “One of AI’s many benefits is that it can identify patterns that are not apparent to humans. Cheap computing power and the availability of vast datasets make this possible. In some ways, AI can help reveal the ‘unknown unknowns,’ which Donald Rumsfeld, former US Secretary of Defense, referenced in his now famous quote.”
Others are taking a more circumspect view. “We find that Delphia and Global Predictions marketed to their clients and prospective clients that they were using AI in certain ways when, in fact, they were not,” said SEC Chair Gary Gensler. “We’ve seen time and again that when new technologies come along, they can create buzz from investors as well as false claims by those purporting to use those new technologies. Investment advisers should not mislead the public by saying they are using an AI model when they are not.”
Wherever the debate ultimately settles, Tam’s work has forced a reckoning over the frontiers of machine intelligence – and postulated an incendiary proposition: that lingua franca programming ability may prove the Rosetta Stone for solving tasks historically reserved for humans alone.
For more information and the full study, visit:
https://leotam.github.io/general/2024/04/17/jpmcfa.html
https://arxiv.org/abs/2310.08678
https://www.sec.gov/news/press-release/2024-36
Media Contact
Company Name: Amicus AI
Contact Person: Barrett Williams
Email: Send Email
Address:3561 Homestead Rd, Suite number. #292
City: Santa Clara
State: CA 95051
Country: United States
Website: www.amicus.ai