Blog: Product Release
Read about Vectara's latest product release "Grounded Generation"Read Blog
Gen AI has changed everything. Have conversations with your data. Get summarized answers. Say goodbye to hallucinations. Say hello to "Grounded Generation." Or better yet - ask it a question.
Some LLMs train their models on your data. Some hallucinate when they don’t know answers to your questions. And some lexical searches provide more relevant answers than solely semantic searches. Grounded Generation remedies all of this.
Whether it's conversational chatbots, user generated content (UGC), or research & analysis libraries, your users get the absolute most relevant answers without hallucinations. Vectara has you grounded.
Vectara uses hybrid search - large language model (LLM) based retrieval (i.e. "Semantic Search") and boolean exact match to find the most relevant products, support cases, and documents that answer your users’ questions first. This means that your Vectara powered chatbots, Q&A systems, and conversational apps and websites can base their responses on the information you and your users care about – and none of what they don’t.
Often, the answers to a user’s question don’t fall neatly in a single FAQ or a single webpage. Vectara doesn’t just summarize individual documents, pages, or products, but helps answer the questions your users have by looking across them all before answering. This means your users spend less time trying to find the answers to their questions and more time enjoying what really makes your product or service unique.
We’ve all seen chatbots making things up, telling users it loves them in the most awkward way possible, and selling competitor's products. In the industry, these are called “hallucinations,” and in part, they happen because most generative systems understand language but they don’t understand your business or when to just say that they don’t know. Vectara uses Grounded Generation, which means that its generative system only relies on facts and data you provide.
Many generative systems are training their systems on their users’ data, leading to PII, business/trade secrets, and even authentication details constituting part of the training. Vectara doesn’t train models on your data: we frankly just don’t need to. Our zero-shot large language models are highly tuned to understand many different types of business data and the questions that can be asked of it. Grounded Generation removes the necessity of retraining models based on your data to accurately generate summaries and chatbot-style responses, because the very act of retrieval has already fine-tuned the data available to the generative model.
“Vectara recognizes that generative AI creates value not simply by automating tasks but by empowering users to engage with their organization's data in new ways. Vectara’s platform delivers conversational AI experiences that nearly always produce relevant, accurate responses. This team is on the cutting edge of developing responsible, highly valuable AI solutions. Their platform promises to mitigate LLM’s key shortcomings while broadening the range of potentially transformative use cases.”
Institute for Experiential AI, Northeastern University