OpenAI has announced plans to acquire database firm Rockset in a bid to tap into the surging enterprise appetite for retrieval augmented generation (RAG) capabilities.
The firm will use Rockset’s platform to power its “retrieval infrastructures” across OpenAI products, the firm stated, while members of staff from Rockset will join the AI company as part of the deal.
Rockset’s platform is targeted towards developers and enterprises, providing access to real-time information, database functions, and vector search capabilities.
“Rockset’s infrastructure empowers companies to transform their data into actionable intelligence. We’re excited to bring these benefits to our customers by integrating Rockset’s foundation into OpenAI products,” said OpenAI COO Brad Lightcap.
Rockset CEO Venkat Venkataramani added to Lightcap’s comments, expressing Rockset’s desire to collaborate with OpenAI to “empower users, enterprises, and developers” to help them fully leverage their data.
While the size of the deal was not disclosed by either firm, sources familiar with the matter told Reuters that OpenAI purchased the firm as part of a stock deal.
The stock deal reportedly valued Rockset at several hundred million dollars, which would make it one of the largest acquisitions ever undertaken by OpenAI.
Why RAG is critical to OpenAI’s enterprise approach
The acquisition by OpenAI could mark a significant moment for the firm as it looks to tap into the growing enterprise excitement over RAG.
RAG improves “response accuracy”, according to JItterbit CTO Manoj Chaudhary, by retrieving information from “large datasets or sets of documents” to generate more informed and more precise answers.
The main attraction of RAG is that it allows an AI model to generate responses based on data outside of its original training data. In this way, any data can be used as a secondary reference for a model, including internal documentation or industry-specific information such as financial reports.
RAG is therefore a useful alternative for businesses that need heightened accuracy without spending more time training.
“The beauty of RAG is that when new information becomes available, rather than having to retrain the model, all that’s needed is to augment the model’s external knowledge base with the updated information,” Melanie Peterson, Senior Director of TrainAI at RWS, told ITPro.
“This reduces LLM development time and cost, while enhancing the model’s scalability.” Peterson added.
Accuracy is also improved with RAG, Peterson added, and AI is “less likely to hallucinate or generate inaccurate responses” owing to a more grounded dataset which makes a model “fit for business purpose”.
Model hallucinations have been a key talking point over the last 18 months. Last year, OpenAI CEO Sam Altman told attendees at Salesforce's Dreamforce conference that hallucinations were "part of the magic" of generative AI and often unavoidable.
The company has made significant strides in cutting down on the issue, however.
“The better we are at finding and retrieving the right documents that are related to the initial question, the better the result from LLM. Therefore a good LLM is only as good as its retrieval system,” said Michał Skibicki, AI Manager at STX Next.
“OpenAI is known for the performance of its model. Now it’s time to leverage it and add the missing piece to truly serve business users as well as consumers,” he added.
