OpenAI Introducing GPTs

OpenAI introduces custom versions of ChatGPT called GPTs

OpenAI has released GPTs, which allow users to create tailored versions of ChatGPT for specific tasks, workflows, or personal needs without requiring any coding knowledge. This update enables anyone to build a custom AI assistant by providing instructions, extra knowledge, and selecting capabilities such as web searching, image generation, and data analysis.

No-code customization and community creation

GPTs are designed to be accessible to non-developers, allowing educators, coaches, and other community members to share their expertise through AI tools. The creation process is as simple as having a conversation with the ChatGPT interface to define the GPT's purpose and behavior.

The GPT Store and monetization

OpenAI is rolling out the GPT Store later in November 2023, which will feature creations from verified builders. This store will make GPTs searchable and include leaderboards and curated categories such as productivity and education. OpenAI has also announced that builders will be able to earn money based on the usage of their GPTs in the coming months.

Technical capabilities and extensibility

GPTs provide a range of built-in capabilities and the ability to extend their functionality through custom actions.

Built-in tools

Users can now access DALL·E, web browsing, and data analysis within a single interface without needing to switch between models. Additionally, ChatGPT Plus now includes updated information up to April 2023 and supports file attachments for searching PDFs and other document types.

Custom actions for developers

Developers can connect GPTs to the real world by defining custom actions via APIs. This allows GPTs to integrate external data or interact with real-world systems, such as connecting to databases, email inboxes, or e-commerce order systems. This system builds on the plugins beta, allowing developers to use existing plugin manifests to migrate their actions.

Enterprise and internal deployment

ChatGPT Enterprise customers can create internal-only GPTs tailored to specific departments, proprietary datasets, or business use cases. Companies such as Amgen, Bain, and Square are using these tools to create brand-consistent marketing materials, assist support staff, and aid software engineer onboarding.

Enterprise admins have control over how GPTs are shared and within the workspace, and OpenAI does not use conversations with Enterprise GPTs to improve its models.

Privacy and safety frameworks

OpenAI has implemented several layers of privacy and safety controls for GPTs:

Data Privacy: User chats with GPTs are not shared with the builders. Users choose whether data is sent to third-party APIs if a GPT uses them.
Model Training: Builders can decide if user chats with their GPT are used to train OpenAI's models, though users can opt their entire account out of training via existing privacy controls.
Safety Mitigations: New review systems are in place to prevent the sharing of harmful GPTs involving fraudulent activity, adult themes, or hateful content. OpenAI also allows builders to verify their identity to increase trust.
Reporting: A reporting feature on the shared GPT page allows users to notify OpenAI of problematic content.

Path toward AI agents

OpenAI describes GPTs as an incremental step toward "agents"—systems that can take on real tasks in the real world. The company states that this transition will be handled incrementally to allow for technical and safety work and societal adaptation.