Connect with us

Accounting

Microsoft researchers teach LLMs to use spreadsheets well

Published

on

Large language models like ChatGPT have traditionally had trouble reading and interacting with spreadsheets, limiting their application in this realm, but recent research from Microsoft claims to have found an answer. 

The paper, SPREADSHEETLLM: Encoding Spreadsheets for Large Language Models, described the problems LLMs typically face with spreadsheets and proposed what it called the “SheetCompressor” framework to address them. 

The issue LLMs have with spreadsheets has to do with tokenage requirements. LLMs, generally, run on “tokens,” which are the basic units of data the model processes. Tokens are words, character sets, or combinations of words and punctuation that are used by large language models to decompose text into. LLMs operate by converting input text into a series of tokens, which the model then uses to understand and generate responses. 

The number of tokens determines the computational cost and capacity needed to handle the input, making token management crucial, especially for complex data like spreadsheets. For example, the phrase, “I heard a dog bark loudly at a cat” would be represented by eight tokens, one for each unique word. In order to preserve system resources, many LLMs have token limits, but even in a limitless environment, complex jobs are resource intensive, with significant computational effort that affects both performance and efficiency. 

Typically, each part of a spreadsheet — even blank cells or repeating cells or those with irrelevant information — costs tokens, meaning even a simple spreadsheet has a much higher token requirement than traditional text. Furthermore, LLMs often struggle with spreadsheet-specific features such as cell addresses and formats, complicating their ability to effectively parse and utilize spreadsheet data. These challenges have limited just how much generative AI models can be applied to reading and interacting with spreadsheets. Considering how many spreadsheets the profession tends to use, this consequently limits their application towards deep accounting work. 

What Microsoft researchers discovered, in short, is that the LLM does not need to burn tokens reading and processing the entire spreadsheet. Instead people can create a compressed version of the document to function as something like an index, with markers or “anchors” indicating especially important information like totals. Additional compression comes from grouping together similar types of data like date columns. So, in a sense, the LLM does not work through the spreadsheet itself but instead references it via a much more efficient index. 

Complex spreadsheets are further supported through a concept called “chain of spreadsheet,” which is similar to “chain of thought” prompting. The method unfolds in two stages. First, the model identifies the table that is relevant to the query and determines the precise boundaries of the relevant content. This step ensures that only pertinent data is considered in the subsequent analysis. Then, the query and the identified table section are re-input into the LLM. The model then processes this information to generate an accurate response to the query.

“Through the CoS, SPREADSHEETLLM effectively handles complex spreadsheets by breaking down the process into manageable parts, thus enabling precise and context-aware responses,” said the paper. 

Experiments with this method found that it significantly increased performance on larger spreadsheets where token limits are a particular challenge. The F1 score (which is used to measure the accuracy of an AI model) for massive spreadsheets was 75% higher than GPT-4 and 19% higher than TableSense-CNN, another spreadsheet methodology for AI; for large spreadsheets, the difference was 45% and 17% respectively; for medium spreadsheets it was 13% and 5%; and for small spreadsheets it was 8%. Overall, the results show that while the method gets more effective the larger the spreadsheet, it can still improve the efficiency of even small spreadsheets. 

“Through a novel encoding method, SHEETCOMPRESSOR, this framework effectively addresses the challenges posed by the size, diversity, and complexity inherent in spreadsheets,” the paper concluded. “It achieves a substantial reduction in token usage and computational costs, enabling practical applications on large datasets. The fine-tuning of various cutting-edge LLMs further enhances the performance of spreadsheet understanding. Moreover, Chain of Spreadsheet, the framework’s extension to spreadsheet downstream tasks illustrates its broad applicability and potential to transform spreadsheet data management and analysis, paving the way for more intelligent and efficient user interactions.”

Implications

Donny Shimamoto, founder and managing director of accounting tech-focused accounting firm IntrapriseTechKnowlogies said, by enabling LLMs to “understand” tabular spreadsheets, accountants will have increased ability to either summarize or analyze a set of data. More than that, however, he said this will likely allow even non-accountants to do the same, removing the accountant as the middle person. However while some accountants may see this as a threat, he said what this would mainly do is clear the majority of simple inquiries from their plates, letting them save their energy for more complex questions and deeper analysis.

“Implementing something like this will require good testing to ensure that the risk of hallucinations is minimized, especially if it is going to help provide non-accountants with information to support decision-making,” said Shimamoto.

David Wood, a Bringham Young University accounting professor who specializes in AI within the profession, raised a similar point, as it would allow those without significant technical knowledge to do the same kinds of tasks that, previously, could only be done by seasoned accounting experts. He raised the example of novices being able to use generative AI to make spreadsheets that only expert professionals could put together. However, while he thinks this could be possible soon, he said that, despite the Microsoft research, it hasn’t arrived just yet.

“However, there are at least three challenges holding back using GenAI with spreadsheets: the size and complexity of the spreadsheets, and the required accuracy for most uses of spreadsheets. This paper takes a large step in the right direction, but it doesn’t solve all the challenges and more work will still be needed in each of these three areas. It would be a mistake to assume that after reading this paper, we have fully figured out how to use spreadsheets and GenAI together. More work is still needed. … I think the path these researchers are taking is significant, but the research “hasn’t arrived yet” meaning that more work is needed. The accuracy rates are just not high enough…yet. Hopefully this paves the way for the next researcher to move it forward further.” he said in an email.

Continue Reading

Accounting

FASB proposes guidance on accounting for government grants

Published

on

The Financial Accounting Standards Board issued a proposed accounting standards update Tuesday to establish authoritative guidance on the accounting for government grants received by business entities. 

U.S. GAAP currently doesn’t provide specific authoritative guidance about the recognition, measurement, and presentation of a grant received by a business entity from a government. Instead, many businesses currently apply the International Financial Reporting Standards Foundation’s International Accounting Standard 20, Accounting for Government Grants and Disclosure of Government Assistance, by analogy, at least in part, to account for government grants.

In 2022 FASB issued an Invitation to Comment, Accounting for Government Grants by Business Entities—Potential Incorporation of IAS 20, Accounting for Government Grants and Disclosure of Government Assistance, into GAAP. In response, most of FASB’s stakeholders supported leveraging the guidance in IAS 20 to develop accounting guidance for government grants in GAAP, believing it would reduce diversity in practice because entities would apply the guidance instead of analogizing to it or other guidance, thus narrowing the variability in accounting for government grants.

Financial Accounting Standards Board offices with new FASB logo sign.jpg
FASB offices

Patrick Dorsman/Financial Accounting Foundation

The proposed ASU would leverage the guidance in IAS 20 with targeted improvements to establish guidance on how to recognize, measure, and present a government grant including (1) a grant related to an asset and (2) a grant related to income. It also would require, consistent with current disclosure requirements, disclosure about the nature of the government grant received, the accounting policies used to account for the grant, and significant terms and conditions of the grant, among others.

FASB is asking for comments on the proposed ASU by March 31, 2025.

“It will not be a cut and paste of IAS 20,” said FASB technical director Jackson Day during a session at Financial Executives International’s Current Financial Reporting Insights conference last week. “First of all, the scope is going to be a little bit different, probably a little bit more narrow. Second of all, the threshold of recognizing a government grant will be based on ‘probable,’ and ‘probable’ as we think of it in U.S. GAAP terms. We’re also going to do some work to make clarifications, etc. There is a little bit different thinking around the government grants for assets. There will be a deferred income approach or a cost accumulation approach that you can pick. And finally, there will be different disclosures because the disclosures will be based on what the board had previously issued, but it does leverage IAS 20. A few other things it does as far as reducing diversity. Most people analogized IAS 20. That was our anecdotal findings. But what does that mean? How exactly do they do that? This will set forth the specifics. It will also eliminate from the population those that were analogizing to ASC 450 or 958, because there were a few of those too. So it will go a long way in reducing diversity. It will also head down a model that will be generally internationally converged, which we still think about. We still collaborate with the staff [of the International Accounting Standards Board]. We don’t have any joint projects, but we still do our best when it makes sense to align on projects.”

Continue Reading

Accounting

In the blogs: Questions for the moment

Published

on

Fighting scope creep; QCDs as the year ends; advising ministers; and other highlights from our favorite tax bloggers.

Questions for the moment

  • CLA (https://www.claconnect.com/en/resources?pageNum=0): One major question of the moment: What can nonprofits expect from future federal tax policies?
  • Mauled Again (http://mauledagain.blogspot.com/): Not long ago, about a dozen states would seize property for failure to pay property taxes and, instead of simply taking their share of unpaid taxes, interest, and penalties and returning the excess to the property owner, they would pocket the entire proceeds of the sales. Did high court intervention stem this practice? Not so much.
  • TaxConnex (https://www.taxconnex.com/blog-): What are the best questions to pin down sales tax risk and exposure?
  • Current Federal Tax Developments (https://www.currentfederaltaxdevelopments.com/): In Surk LLC v. Commissioner, the Tax Court was presented with the question of basis computations related to an interest in a partnership. The taxpayer mistakenly deducted losses that exceeded the limitation in IRC Sec. 704(d), raising the question: Should the taxpayer reduce its basis in subsequent years by the amount of those disallowed losses or compute the basis by treating those losses as if they were never deducted?

Creeping

On the table

  • Don’t Mess with Taxes (http://dontmesswithtaxes.typepad.com/): What to remind them, as end-of-year planning looms, about this year’s QCD numbers.
  • Parametric (https://www.parametricportfolio.com/blog): If your clients are using more traditional commingled products for their passive exposures, they may not know how much tax money they’re leaving on the table. A look at possible advantages of a separately managed account. 
  • Turbotax (https://blog.turbotax.intuit.com): Whether they’re talking diversification, gainful hobby or income stream, what to remind them about the tax benefits of investing in real estate.
  • The National Association of Tax Professionals (https://blog.natptax.com/): Q&A from a recent webinar on day cares’ unique income and expense categories.
  • Boyum & Barenscheer (https://www.myboyum.com/blog/): For larger manufacturers, compliance under IRC 263A is essential. And for all manufacturers, effective inventory management goes beyond balancing stock levels. Key factors affecting inventory accounting for large and small manufacturing businesses.
  • U of I Tax School (https://taxschool.illinois.edu/blog/): What to remind them — and yourself — about the taxation of clients who are ministers.
  • Withum (https://www.withum.com/resources/): A look at the recent IRS Memorandum 2024-36010 that denied the application of IRC Sec. 245A to dividends received by a controlled foreign corporation.

Continue Reading

Accounting

PwC funds AI in Accounting Fellowship at Bryant University

Published

on

PwC made a $1.5 million investment to Bryant University, in Smithfield, Rhode Island, to fund the launch of the PwC AI in Accounting Fellowship.

The experiential learning program allows undergraduate students to explore AI’s impact in accounting by way of engaging in research with faculty, corporate-sponsored projects and professional development that blends traditional accounting principles with AI-driven tools and platforms. 

The first cohort of PwC AI in Accounting Fellows will be awarded to members of the Bryant Honors Program planning to study accounting. The fellowship funds can be applied to various educational resources, including conference fees, specialized data sheets, software and travel.

PwC sign, branding

Krisztian Bocsi/Bloomberg

“Aligned with our Vision 2030 strategic plan and our commitment to experiential learning and academic excellence, the fellowship also builds upon PwC’s longstanding relationship with Bryant University,” Bryant University president Ross Gittell said in a statement. “This strong partnership supports institutional objectives and includes the annual PwC Accounting Careers Leadership Institute for rising high school seniors, the PwC Endowed Scholarship Fund, the PwC Book Fund, and the PwC Center for Diversity and Inclusion.”

Bob Calabro, a PwC US partner and 1988 Bryant University alumnus and trustee, helped lead the development of the program.

“We are excited to introduce students to the many opportunities available to them in the accounting field and to prepare them to make the most of those opportunities, This program further illustrates the strong relationship between PwC and Bryant University, where so many of our partners and staff began their career journey in accounting” Calabro said in a statement.

“Bryant’s Accounting faculty are excited to work with our PwC AI in Accounting Fellows to help them develop impactful research projects and create important experiential learning opportunities,” professor Daniel Ames, chair of Bryant’s accounting department, said in a statement. “This program provides an invaluable opportunity for students to apply AI concepts to real-world accounting, shaping their educational journey in significant ways.”

Continue Reading

Trending