OpenAI drops new products for developers - GPTs and Assistants API supercharge building AI apps


At their developer conference, OpenAI announced two major new products - GPTs, which allow developers to build custom AI personalities that can be shared worldwide, and the Assistants API, a stateful API with advanced tools like code execution and knowledge retrieval built-in to help developers easily integrate powerful AI capabilities into their apps.

Timestamped Highlights

๐Ÿค“ GPTs let developers create custom AI personas with configurable instructions, actions connecting external APIs, and knowledge injection
๐Ÿ˜Ž Demo of pirate-themed GPT showing off instruction customization and personality injection
๐Ÿ“ Actions connect GPTs to external APIs, improving capabilities. Easily wrap OpenAI tools too.
๐Ÿคฏ Crazy GPT demo stitching together instructions, actions, knowledge and DALL-E to generate art, lighting mood based on computer vision
๐Ÿ› ๏ธ Introducing the Assistants API with primitives for statefully storing instructions, managing conversations and messages
๐Ÿ“ฅ Tools like Code Interpreter run on OpenAI's end but integrate outputs into API
๐Ÿ“š Knowledge Retrieval tool allows expanding model knowledge with developer and user uploaded files
๐Ÿ’ก Retrieval works on both assistant and thread levels depending on privacy needs
๐Ÿ—๏ธ Function Calling improvements - JSON output mode and parallel function calls

Key Insights

GPTs enable user-generated AI content, unlocking new experiences. Creating GPTs allows developers to build AI apps that feel more responsive and alive - with characters that persistently remember conversations, connect APIs to take actions, and ingest knowledge. This enables user-generated AI content.
Stateful Assistants API simplifies building advanced AI apps. The Assistants API provides the primitives for statefully storing instructions, managing conversations and tool integration so developers can focus on building differentiated AI app experiences vs handling low-level ML challenges.
๐Ÿ› ๏ธ
Integrated tools remove infrastructure burden. With Code Interpreter, Retrieval and Function Calling, developers benefit from OpenAI handling the execution, scaling and optimization - and the tools integrate outputs back into the API for easy consumption without setting up infrastructure.
Easy to get started, lots of control over customization. Whether using the Playground or API, it's simple to spin up an assistant. But developers have fine-grained control over instructions, tools configuration and more for advanced customization.
Assistant API enables data minimization. Developers can manage data minimization well - only using data in thread for a single conversation instead of associating all data with a user across conversations.
Build creative and helpful AI apps! These products enable developers to build all sorts of creative AI apps - from productivity helpers to gaming sidekicks to meme generators and more. They remove AI/ML barriers letting developers focus on human-centric innovations.
Lots more Assistant API functionality launching soon! OpenAI shipped a lot today but isn't stopping - much more multi-modal, developer infrastructure improving capabilities and asynchronous support coming soon so the API will continue advancing quickly!.
This blog is a summary of a YouTube video "New Products: A Deep Dive - YouTube" by OpenAI