GPT-3 by Sandra Kublik & Shubham Saboo

Author:Sandra Kublik & Shubham Saboo [Sandra Kublik]
Language: eng
Format: epub
Publisher: O'Reilly Media, Inc.
Published: 2022-07-14T16:00:00+00:00

Data Analysis Applications of GPT-3: Viable

The story of the start-up Viable is an example of how much things can change from the moment you start working on a business idea to actually finding a product-market fit and a customer base. Viable helps companies better understand their customers by using GPT-3 to summarize customer feedback.

Viable aggregates feedback such as surveys, help desk tickets, live chat logs, and customer reviews. It then identifies themes, emotions, and sentiments, pulls insights from those results, and provides a summary in a matter of seconds. For example, if asked, “What’s frustrating our customers about the checkout experience?” Viable might respond: “Customers are frustrated with the checkout flow because it takes too long to load. They also want a way to edit their address in checkout and save multiple payment methods.”

Viable’s original business model involved helping early-stage start-up companies find product-market fit using surveys and product roadmaps. Requests started coming in from bigger companies, asking for support in analyzing huge volumes of text, such as “support tickets, social media, app store reviews, and survey responses” that changed everything, says Daniel Erickson. Erickson is the founder and CEO of Viable—and an early adopter of the OpenAI API. He explains, “I spent actually about a month just experimenting, literally just taking our data, putting it into the Playground, figuring out different prompts and things like that. And eventually, I came to the conclusion that [GPT-3] could power a very powerful question and answer system.”

Erickson and his colleagues began using the OpenAI API to interact with and generate insights from the large datasets they were working with. They initially used another NLP model, achieving mediocre results, but when they began working with GPT-3, the team saw “at least a 10% increase across the board. When we’re talking about going from 80% to 90%, that’s a hell of an increase for us.”

Building on that success, they used GPT-3 in combination with other models and systems to create a Q&A feature that allows users to ask a question in plain English and get an answer. Viable converts the question to a complex query that can pull all the relevant feedback from the database. It then runs the data through another series of summarization and analysis models to generate a refined answer.

In addition, Viable’s system provides customers with “a 12-paragraph summary every week…that outlines things like their top complaints, their top compliments, their top requests, and top questions.” As you might expect from customer-feedback specialists, Viable has thumbs up and thumbs down buttons next to every answer the software generates. They use this feedback in retraining the model.

Humans are part of the process, too: Viable has an annotation team whose members are responsible for building training datasets, both for internal models and GPT-3 fine-tuning. They use the current iteration of that fine-tuned model to generate output, which humans then assess for quality. If the output doesn’t make sense or isn’t accurate, they rewrite it. And once they have a list of outputs they are satisfied with, they feed that list back into the next iteration of the training dataset.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.