Hallucination

May 30 , 2023 by Shane Connelly & CJ Cenizal

Generative AI promises to revolutionize how you can benefit from your data, but you need it to provide dependable information without the risk of data leakage. This is why today we’re adding a fundamental capability to our platform to make generative AI safer to use. It enables you to ask your data questions and get reliable, accurate answers by retrieving and summarizing only the relevant information. As a pioneer, Vectara coined the term “Grounded Generation,” while it is generally referred to as “Retrieval Augmented Generation.”  Sign up for a free account to try it on your own data.

How it works

“Garbage in, garbage out,” goes the old computer science saying. The output of a process is subject to the quality of its inputs. If you don’t believe me, just try making ice cream with spoiled milk! We can also observe this relationship in the behavior of generative AI when it tries to seduce New York Times reporters or when it just plain makes stuff up. When Gen AI doesn’t have access to sufficient relevant information it answers with fabrications, commonly known as “hallucinations”.

The way Grounded Generation with Vectara works is we only use valid responses from your data relative to the search query. This dramatically reduces hallucinations in Vectara’s responses. You can try it out on your own on our newly launched AskNews demo to experience Retrieval Augmented Generation (RAG) or register an account to ground generative summaries on your own data.

Figure 1: The AskNews demo application.

 

You can use Vectara’s new Grounded Generation feature to build chatbots that don’t lie and provide website search that provides answers based on your website’s content. Your customer support and customer success agents can use it to respond more quickly to customer needs, and it can enable your e-commerce applications to directly answer questions that your users ask about the products in your catalog.

We’re excited to announce that we’re offering 15,000 grounded generative summaries per month included with our free plan. If you need more, we offer competitive pricing on both the individual and enterprise level.

For more details on how to incorporate Grounded Generation into your application, have a look at our developer documentation.

Hybrid Search

We’re also incredibly excited to share another feature with this launch: Hybrid Search! Neural LLM systems are excellent at understanding the context and meaning of end-user queries, but they can still underperform when matching exact product SKUs, unusual names of people or companies, barcodes, and other text which identifies entities rather than conveying semantics. We’re bridging this gap by introducing a lexical configuration that matches exact keywords, supports Boolean operators, and executes phrase searches, and incorporates the results into our neural search results.

To take advantage of Hybrid Search, all you need to do is to set a non-zero “lambda” value. We recommend users that want to try it out start off with a lambda of 0.025 and then adjust this to find the value that’s best for their data. For more details on this configuration option, have a look at our documentation.

Better, faster, and free

Help your users make faster, better decisions based on your organization’s data by signing up for a free account and trying Grounded Generation and Hybrid Search. If you need help getting up and running, reach out to our team on the Community Forums and our Discord server, or refer to our documentation.