OpenAI has launched a new ChatGPT feature that allows users to explore different tones, styles, and strategies by branching their conversations with the AI chatbot into multiple parallel threads.
Announced on Friday, September 5, this feature functions similarly to creating a duplicate of a user's ChatGPT conversation, enabling exploration of various directions with the chatbot while keeping the original conversation intact.
For example, a marketing team using ChatGPT to brainstorm ad copies can create separate conversation branches to try out formal, informal, or humorous tones while maintaining the original prompt and AI-generated responses. This new branching option is available to all users with a ChatGPT account, provided they are logged in on the web version of the app.
It follows requests from numerous users in the past for similar capabilities. Previously, ChatGPT users who wanted to explore different approaches had to overwrite their existing conversation with the chatbot or start a new chat entirely. With the branching feature, users can more easily experiment with what-if scenarios in their conversations with ChatGPT.
The latest update to ChatGPT has been well-received by several X users, with some comparing it to Git, a version control system that allows programmers to create separate branches of code to test changes without affecting the main codebase.
Instructions for Utilizing the New Feature
To use the new ChatGPT feature, you need to be signed up on the platform and have the web version open on your device.
Follow these steps:
– Hover over any message in the conversation window.
– Click on ‘More Actions’
– Select ‘Branch in new chat’
This will create a new conversation thread containing all the context from the original chat up to the branching point, allowing users to revisit the original conversation without any changes.
A 2024 study found that linear dialogue interfaces for large language models (LLMs) could increase cognitive load and reduce efficiency among users as they are forced to “repeatedly compare, modify, and copy previous content.”
LLMs with such interfaces poorly serve scenarios involving “multiple layers, and many subtasks—such as brainstorming, structured knowledge learning, and large project analysis,” it added.
