WTF is Retrieval Augmented Generation for AI chatbots and large language models?
This article is a WTF explainer, in which we break down media and marketing’s most confusing terms. More from the series →
AI-powered chatbots are scary smart, thanks to the large language models undergirding them. But they’re only as smart as the data they’re trained on.
Two major factors can inhibit the AI chatbots being created by brands and publishers: the timeliness of the bots’ training data and the inclusion of proprietary information in that data set. In the first case, a company would need a way to pass information in real time to an LLM powering its chatbot. And in the second case, the company would need to be able to safeguard that information so that the LLM is just provided whatever necessary context to respond someone’s prompt. In both cases, a system called “retrieval augmented generation” would provide a framework for bridging the gap, as explained in the video below.
“There is a difference between being trained on public and licensed data and knowing what sits in the cloud environment of a brand,” said Hugo Loriot, head of data and technology integration at The Brandtech Group. He added, “All of the data that is very specific to a brand, this is off-limits for a [large language] model — unless you have a way to feed that data into the model, which is exactly what RAG does.”
More in Marketing
Electronic Arts is betting that in-game ads can out-earn CTV
To make in-game ads stick, EA has built its own stack rather than rent one. Now it wants to shape the standards before anyone else does.
Future of Marketing Briefing: Why Bose is building an entertainment company
Bose has a new entertainment division. Its CMO hasn’t used a creative agency in five years. The two things are related.
The rise of pharma ad tech
Insiders say it comes at the cost of legacy platforms such as DSPs and SSPs.