In the past few weeks, business articles and Linked-In posts have been replete with question marks as to the potential of Large Language Models and ChatGPT and their real applicability within the realm of business as opposed to just consumer. In this article I will aim to explain the key differences between the two, the first I will shorten to B2B AI and the later to B2C AI. But before I do that, perhaps it might be useful to cite some interesting cases that have popped up that emphasize the point.
* Patent pending
One was reported in the New York Times in May titled: A Man Sued Avianca Airline: His Lawyer Used ChatGPT. As it turns out, when prompted, GPT had invented out of thin air all the precedents cited by the unsuspecting prosecution lawyer. When the case judge and defense lawyers searched but couldn’t find any reference to the cited precedents, the lawyer admitted he had simply copied GPT. The judge proceeded to throw out the case; and the lawyer apologized (As of this writing, he was being considered for disbarment). In Europe, similar cases are being brought up that has lawsuits claiming violation of Intellectual Property and GDPR rules (dealing with privacy). Other lawsuits have to do with inaccurate or outright erroneous information.
So, when I was recently asked about Chat GPT, the best answer I could muster was comparing it to a baby’s brain. Sure, it is wired with billions and billions of neural links. And yes, Large Language Models, such as Chat GPT, could probably begin to process and synthesize data in fairly short order, not unlike a human brain. And yet, I argued that a high school with the ability to read, write, and synthesize, does not automatically make them a Doctor or a Lawyer, unless they delved into contextualized learning on that specific topic. Similarly, Large Language Models have proven to be certainly good readers and writers; and they have a lot of general knowledge that is available in the public domain. They can synthesize it all and can regenerate content. But they aren’t lawyers or Doctors, nor can they be considered experts on any one topic (We will ignore the question of can they become in the future for now).
Expertise fundamentally comes from contextualized learning, experience, and from relationships.
Several reasons come to light, all of which shed light on a divide emerging between B2C AI and B2B AI.
Most of the critical information in business sits protected behind firewalls and is not accessible via either crawlers or LLMs. Patient records, business contracts, financials, project data, product and engineering specs, research and support documentation, partner documentation, you name it. If LLMs don’t have access to it, they know nothing about it. In ecosystems that TIDWIT powers, there could be tens of thousands of content files that are “off the grid” so to speak and available only within the ecosystem.
LLMs can be very reliable if they are provided the right context. TIDWIT’s concept of ecosystem “Walled Gardens” curates content that is relevant, readable, and updated, providing the necessary context for smarter results. But sometimes, the content in an enterprise can be complex and in an unreadable format, which would weaken AI’s effectiveness. TIDWIT’s vast experience in handling and processing millions of enablement digital assets over the years optimizes readability and “AI friendliness” for optimal content processing, which in turn will have a direct impact on the quality of the results that any AI query will render.
Most organizations will not open their firewalls to AI. The evidence is easy. If after almost 3.5 decades of Internet, most large organizations increasingly secure their intellectual property against cyber-threats, what are the odds of them opening it up to something as powerful as Artificial Intelligence? The answer is ZERO likelihood. Yes, they may open parts as they to search engine crawlers, but that’s pretty much the extent as it will go when it comes to B2C AI.
The architecture of LLMs continue a long lineage going as far back as Darpa, which is generally flat and does not factor in complex inter-organizational structures. Google made billions in B2C Search, but they could never quite penetrate B2B. Why? Because their architecture wasn’t designed to complex multi-tiered ecosystem structures across organizations. Rather it was B2C with a simple front-end screen with a prompt.
Connect with us today to request a TIDWIT Ecosystem AI demo an learn how you can address these challenges and make the most of your Ecosystem content and knowledge.
© 2024 TIDWIT Inc. – All rights reserved