The debate continues on whether generative A.I. large language models (LLMs) like ChatGPT, designed to understand and generate human-like text, will impact the future of work, including finance.

Researchers at OpenAI, the company that launched ChatGPT, and the University of Pennsylvania released a working paper last week titled, “GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models.” 

It predicts that about 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of LLMs. And “approximately 19% of jobs have at least 50% of their tasks exposed when considering both current model capabilities and anticipated tools built upon them,” according to the report. 

They measured the “exposure” of detailed work activities, which is like a job description, and tasks to LLMs’ capabilities. No exposure means using an LLM via GPTs didn’t reduce the amount of time to complete an activity. Direct exposure means it reduces the time at least by half. And a third category indicated that access to an LLM alone wouldn’t reduce the time by half—but additional software developed on top of the LLM could achieve that goal.

The areas that have exposure to LLMs include jobs working in securities commodities contracts and other financial investments, monetary authorities, insurance carriers, data processing hosting and related services, and publishing, to name a few. And some occupations with no exposure include truck drivers, agricultural equipment operators, and glass installers. However, the researchers didn’t assess if or whose jobs would be lost with the increased exposure to LLMs.

The findings are based on a Department of Labor database of 1,016 occupations, in addition to 2020 and 2021 Bureau of Labor Statistics employment and wage data. The researchers used both A.I. models and people to assign exposure levels.

Also noted in the report—predicted effects of LLMs span all wage levels, with higher-income jobs potentially facing greater exposure to LLM capabilities and LLM-powered software.

Daron Acemoglu is a professor and labor economist at MIT. Acemoglu is not an author of the report but is the coauthor of research on automation and wage inequality. I asked his opinion of the research.

“The predictions of the new working paper are reasonable, though, of course, nobody can see the future,” Acremoglu says. He noted the optimism of the researchers, but 50% is a “huge” prediction. “Automation of certain tasks in the past has not always boosted productivity as much as its proponents had hoped. It remains to be seen whether LLMs will do better here.” He also noted a risk for demographic groups who previously specialized in automated tasks to have depressed wages.

When it comes to generative A.I. and the finance function, Michael Schrage, a research fellow at the MIT Sloan School Initiative on the Digital Economy, used the LLMs ChatGPT and DallE2 to demonstrate financial planning and analysis scenario generation, he says.

“Frankly, what we did was demonstrably superior to the ‘humans-only’ scenarios,” Schrage says. “In my mind, there’s no question that—just as macros automated huge swathes of mission-critical Excel spreadsheets—we’ll soon see ‘prompt engineering’ and engineered prompts automate even greater portions of financial process workflows. To me, the wild card is to what extent ‘fine tuning’ models (for example, using your own models to complement or supplement the LLM) get developed, trained, and deployed.”

Sheryl Estrada
sheryl.estrada@fortune.com

Sign up here to receive CFO Daily weekday mornings in your inbox.

Big deal

A report by Emburse, a spend optimization company, takes a look at the talent strategies of 400 finance leaders in a complicated labor market. Employers are implementing automation to deal with current market challenges, according to the report. Respondents said financial reporting (44%), travel and expense management (33%), and cash management and application (32%) were the top three processes that leveraged automation, and respondents plan to improve them through automation even further this year. Smaller organizations have embraced financial reporting automation faster than larger ones. Nearly 58% of companies with under $50 million in revenue said they had automated financial reporting—12 points higher than midsize companies (41%) and companies with more than $500 million in revenue (46%).

Going deeper

"How Pandemic Accelerated Digital Transformation in Advanced Economies," a report by the International Monetary Fund, found that across advanced economies, digitalization increased by an average of six percentage points. Small firms, which have historically been less digitalized, enjoyed the biggest gains. Similarly, sectors that are least digitalized invested more in digitalization.

Leaderboard

Nadir Mahmood resigned as CFO at Nkarta, Inc. (Nasdaq: NKTX), a clinical-stage biopharmaceutical company, to serve as the CEO of a private biotechnology company. Mahmood will remain with Nkarta until June 30 for a transition period. He has been in his current role since 2020. Mahmood joined the company in 2018 as SVP of corporate development. Nkarta has initiated an external search for its next CFO.

Steven Passey was named CFO at Ra Medical Systems, Inc. (NYSE American: RMED), effective April 1. Passey replaces Brian Conn, who has been serving as acting CFO. Passey was most recently CFO at QSI Holdings, Inc. Before that, he was the corporate controller/treasurer for Alsco, Inc. Passey also served as the CFO at TechniScan, chief accounting officer and treasurer for Mrs. Fields' Famous Brands, and controller at REIT Properties. He began his career at KPMG and later worked for Ernst & Young.

Overheard

"I don't have any firm conclusions on that, but I'm certainly very open to that. Ultimately this is gonna be a decision for Congress because that's set in the law and will be working with them as they revisit it too."

—Consumer Financial Protection Bureau Director Rohit Chopra told Yahoo Finance that on Friday when asked whether the level of deposit insurance should be raised. Chopra also suggested he agrees with FDIC Chair Martin Gruenbergon on exempting community banks from new assessments to build up the deposit insurance fund that took at least a $20 billion hit from backstopping uninsured depositors at Silicon Valley Bank and Signature Bank.