Including a public preview of the Assistants API, new text-to-speech (TTS) capabilities, upcoming updates to GPT-4 Turbo and GPT-3.5 Turbo models, new embedded models, and updates to the fine-tuning API.
Compared to previous chat completion APIs, the Assistants API is able to remember previous conversations and create persistent and infinitely long threads.
The Assistants API is a new service provided by Azure OpenAI that aims to help developers more easily create high-quality artificial intelligence assistant experiences in their applications. Compared to previous chat completion APIs, the Assistants API provides the following key improvements:
1. Stationary evolution: The Assistants API can remember previous conversations, which means it can maintain context during the conversation and provide a more coherent and personalized communication experience.
2. Persistence and infinite threads: Developers can create persistent conversation threads that can be infinitely long regardless of the size of the model context window. In this way, even long or complex conversations can proceed smoothly.
3. Simplify conversation state management: Since the Assistants API itself supports conversation state management, developers do not need to build their own complex systems to track conversation state or manage threads. This reduces development complexity and makes it easier to create dynamic, interactive AI assistants.
Azure OpenAI Service updates and new features mainly include:
1. Public preview of the Assistants API: This is a new feature designed to allow developers to easily create high-quality experiences similar to the co-pilot in their own apps. The Assistants API supports persistence and infinitely long threads, simplifying developers ‘work in managing conversation state and integrating tools, and improving development efficiency.
2. Text-to-Speech (TTS) function: A new text-to-speech model has been introduced, which can convert text into high-quality human voices with personality and style. It is suitable for customer support, training videos, real-time streaming and other scenarios.
3. Updates to GPT-4 Turbo and GPT-3.5 Turbo models: Updated versions of GPT-4 Turbo and GPT-3.5 Turbo have been released, including improved code generation capabilities, reduced “laziness”, and fixes that affect non-English UTF-8 generation errors, etc. Provides improved instruction compliance, JSON schema, reproducible output, parallel function calls and other features.
5. New embedded models and fine-tuning API updates: New embedded models and fine-tuning API updates have been introduced, including a new model, support for continuous fine-tuning and better pricing. These updates make it easier and economical to build custom models and fine-tune with longer messages.
6. Pricing optimization: Provides a price reduction for the cost of training and hosting fine-tuning models on Azure OpenAI Service, including a 50% reduction in training and hosting costs for GPT-3.5-Turbo.
Details:https://msft.it/6019ihVTb
Video: