Generative AI and Version Control
Version control (and variations on that theme) are an important part of the financial professional’s responsibilities. Whether it is making sure a current period’s documents are consistent with a prior period’s (e.g., “last period” numbers in the current report lining up with “current period” numbers in the prior report), ensuring the most recent version of a document is published and not an interim version, or seeking consistency between documents in different languages or levels of detail, mistakes have been made and people are thrown at a task to limit those mistakes.
I have, coincidentally, engaged in a number of “version control” tasks over the last couple of days, some more important and others more fanciful. I wondered if ChatGPT and friends could help. Here’s two of them:
- Reviewing the latest EDGAR Filer Manual draft
The US Securities and Exchange Commission recently released the draft of version 67 of the EDGAR Filer Manual. As I started to analyze the EFM back with version 10, that has been many, many times I have searched for the redline in Chapters 5 and 6 to see if I can assess, summarize, and report on any significant changes.
I have not yet spent significant time honing my prompts, but I am somewhat confident that the right generative AI tool will be able to help in a big way. Although ChatGPT does not natively read PDF files, GPT-4 with plug-ins offers a wide variety of additions that can read PDF. Bing AI Chat in Microsoft Edge can work with PDF; Claude (Anthropic) supports PDF files.
No, I don’t think the tools are ready to provide Eric-level commentary related to the changes, but identifying and summarizing the changes is a great first step.
- Researching the latest reboot of “Takeshi’s Castle”
Japanese game shows have a long-standing reputation for being “wild and wacky” to the outsider. Those familiar with the franchise known as “Wipeout” (e.g., ”Wipeout Canada”) may not know that the American Broadcasting Network (ABC in the US) had been sued by TBC (Tokyo Broadcasting System, not Turner) for copyright infringement of its venerable 1980s show “Takeshi’s Castle”, and the TBS co-owned US series “MXC” (Most Extreme Elimination Challenge), a US-focused, adult oriented somewhat risqué reworking of Takeshi’s Castle that reworked the video but reused none of the audio for humorous effect.
Lovers of puns and adult humor can find MXC online in many countries. I, fitting into the former category, recently watched MXC and wondered why people put themselves through what appeared to be great pain. Having binged the series, I found out that – after 34 years – the original series, Takeshi’s Castle - had been rebooted, and is available on Amazon Prime Video/Prime Video Canada.
I watched the 46 minutes official preview as a Youtube video – in Japanese, with English subtitles – introducing the new series. I wondered if the English and Japanese were significantly different, and also wanted to ask questions, such as whether key actors from the 1980s original series appeared (the press was clear that Mr. Takeshi himself, now 75, was part of the effort). And – yes, there are plugins to GPT-4 that can search Youtube transcripts.
Was my favorite character (as humorously portrayed in MCX) in the latest series? No, but he is involved in politics now. Who are the new people? Comedians, mostly, and what they are best known for provided. Is the English translation the same as the Japanese? Close enough, and most of the changes are for regional issues, such as whether the family name is provided first or last.
What does this mean? Word documents, PDFs, Youtube videos: if it is in electronic form, these tools can help. What does that mean to improving quality of issues related to:
- Financial statements and tax returns – across versions and periods
- Policies and procedures – especially analyzing what must change and where it will be relevant
- Contracts and proposal management
- Systems lifecycles, controlling code and specifications.
I do not mean to leave AI/generative AI in isolation. I would be remiss to not mention the benefits of standardization here. One of the great version control projects we were involved in deals with the US government’s U.S. Standard General Ledger (USSGL). This massive undertaking lays out the different budget and reporting general ledger accounts, how they work together to represent key transactions, and how they link to different reports (“cross walks”). For 2024, the document is 1412 pages, and pages 4 through 30 represent in summary the changes from the prior version. Some 17 years ago, we attempted to show how representing the USSGL with XBRL GL would permit automated consumption and analysis of the annual changes and analyze the impact on existing systems. AI can help compensate for the lack of standardization, but the two together would be an amazing tool to keep systems up-to-date and relevant.
Whether your goal is keeping up with change, improving quality, minimizing mistakes, or finding out what Junji Inagawa is doing today, generative AI tools like ChatGPT may be an important tool in your version control arsenal.
Comments
- No comments found
Leave a comment