At Arcurve, our core value is partnering with our clients to solve their complex business problems using technology. Because we’re with our clients on every step of their digital journeys, when a new technology becomes a hot topic, our talented team examines that technology and figures out how it can apply to and add value to their businesses. Enter ChatGPT and other Large Language Models. Like many out there, you might be wondering how they can help (or hinder) your business. That’s where we come in....
The File Store: Where Knowledge Goes to be Buried…
Picture this: your enterprise stores the entirety of its knowledge base in a cloud-based file store, Wiki, or document management system. We can bet that these systems might resemble the closing scene from Indiana Jones: Raiders of the Lost Ark, in which the Ark of the Covenant is wheeled into a gigantic warehouse to be literally and metaphorically ‘buried’ for all time.
Despite your digital files being nicely indexed with naming conventions, categorization schemes, and tagging standards, these important elements are oftentimes lacking—and aren’t much help—when trying to find specific documents (or excerpts) contained in these huge repositories of Enterprise knowledge.
While search technology on the internet has succeeded in making it possible to find just about anything, enterprise search continues to face challenges that have yet to be efficiently solved even in something as seemingly simple as finding that lost document from 2005. Why is it so difficult to harness the rich knowledge that is stranded in these unwieldy data jails?
There is No Enterprise Web
Web-based search has most notably been dominated by Google’s Page Rank system and its successors. Algorithmic improvements have since made the massive body of knowledge known as the World Wide Web more searchable and discoverable than ever.
However, there is no equivalent on which to base a meaningful enterprise algorithmic search. Even if files are hyperlinked, knowledge trapped in complex proprietary file formats such as PDF prevent effective search and discovery.
The inherent siloed nature of the data makes the challenge associated with making trapped knowledge accessible even more difficult to address.
How Large Language Models Help
File folder hierarchies, tags, SharePoint sites and naming conventions are tools people have used over the years to try to find the knowledge they need to do their jobs.
Unfortunately, these have all fallen short of helping people quickly discover and derive insight from the vast stores of knowledge on which enterprises spend billions every year to maintain.
Large Language Models address the shortfalls of prior methods by taking the knowledge trapped in file stores and combine them with data spread across enterprise databases to train a model that people can use to quickly find the knowledge they are looking for, using intuitive questions.
To be used most effectively and safely, the right LLM model (like ChatGPT) with enterprise level and service should be set up by experts who have experience in the domain of natural language processing. Large Language Models are only as good as the data on which they are trained. It is essential to ensure factually accurate sources are used to construct the model.
But that’s a story for another day.
For today, we’ll leave you with this. While the capabilities of Large Language Models will continue to rapidly expand, improve, and extend, it is possible to take advantage of their current capabilities now, and to be flexible and prepared for ongoing developments.
