Weaviate announces the release of a generative search module for OpenAI’s GPT-3, and other generative AI models (Cohere, LaMDA) to follow. This module allows Weaviate users and customers to easily integrate with those models and eliminates hurdles that currently limit the utility of such models in business use cases.
OpenAI’s ChatGPT chatbot (based on GPT-3) has captivated the world with its surprisingly human-like ability to respond to queries in natural language. However, such generative models have so far been limited by a centralized and generic knowledge base that leaves them unable to answer business-specific questions.
Weaviate’s generative module removes this limitation by allowing users to specify that the model work from users’ own Weaviate vector database. This best-of-both-worlds solution combines language abilities like those of ChatGPT with a vector database that is relevant, secure and easily updated in real time. Such a solution is also far less prone to hallucination.
“Generative models like the one used to create ChatGPT display very impressive language abilities but their business applications are limited because they are trained on data that’s freely available on the internet.
“Meanwhile, about 80% of the world’s data exists behind firewalls–for good reasons, because it is confidential or proprietary. So, we created a generative AI module to integrate with GPT-3 and other models (like LaMDA), allowing our open-source users and customers to leverage the model’s language abilities with their own Weaviate vector databases that are easily updated and capable of protecting sensitive data.”
– Bob van Luijt, CEO and co-founder, Weaviate
Weaviate operates on an open-source model, so its generative AI module is freely available to download. The new model also integrates with the company’s SaaS and hybrid SaaS products for use by clients with service-level agreements.
The Weaviate vector-search engine is a “third wave” database technology. Data is processed by a machine learning model first, and AI models help process, store, and search through it. As a result, Weaviate excels at searching using natural language but it is not limited to language; Weaviate can also search images, audio, video, or even genetic information. The open-source code has been downloaded over 1.7M times.
Weaviate–which was known until recently as SeMI Technologies–is registered in The Netherlands as Weaviate B.V. and in the U.S. as Weaviate Holdings, Inc.