Assessing the options for GenAI information integration along with considerations for your specific company
19 hours ago
While GenAI is super cool, there’s no doubt that the companies getting the most value out of GenAI have found a means to integrate their own information with the AI model. What I don’t necessarily see, however, is a clear way to think about how a company should approach its GenAI strategy. You might read things online about how Bloomberg found great value building its own large language model (LLM) from scratch, or you might come across as the numerous social media posts about how great retrieval augmented generation (RAG) is. To be clear, each of these tactics is excellent in its own right, but as you can imagine, there is no one-size-fits-all approach.
That said, this blog post isn’t going to give you a definitive answer for your company’s situation. Rather, what our goal will be is to provide you with a framework for how to think about integrating GenAI with your company’s information. This post is structured into two high level sections. In the first section, we’ll give a general high level understanding of what each of the three distinct options there are, stopping short of the nitty gritty technical details. In the second section, I’ll provide a number of specific considerations to think about when selecting one of the options outlined in the first section.
Without further ado, let’s get into it!
At a high level, there are three distinct ways that you can leverage your own company’s data. These include…
- Building your own model from scratch
- Fine tuning
- Retrieval augmented generation (RAG)
In the following subsections, we’ll go into more details about what each of these are along with a few technical nuances for each.
Building your own model from scratch
Just like it sounds, this is the situation where you build your own GenAI model from the ground up. These are…
