Latest Updates and Features in OpenAI
- **GPT-4o mini** — Launched on July 18, 2024, surpasses GPT-3.5 Turbo in academic benchmarks and supports multiple languages. It also excels in function calling and long-context performance.
- **New Responsible AI Content Filtering** — The `DefaultV2` policy includes Prompt Shields for jailbreak attacks and protected material detection for text and code.
- **Global Deployment of GPT-4o** — Now available in multiple regions including Australia East, Brazil South, Canada East, and more.
- **Token-Based Billing for Fine-Tuning** — Fine-tuning billing is now based on the number of tokens in the training file, making cost estimation easier.
- **Assistants API Updates** — New features include file search tool, vector storage, and support for various parameters like `temperature` and `top_p`.
- **Data Analysis Improvements** — Enhancements include uploading files from Google Drive and OneDrive, interacting with tables and charts, and customizing presentation-ready charts.
- **Memory Feature in ChatGPT** — Available to Plus users, allowing ChatGPT to remember information across chats.
- **ChatGPT Team Plan** — A new plan for teams providing a secure, collaborative workspace with access to advanced models and tools.
Top Search Results
GPT-4o Mini
- **Launch Date** — Introduced on July 18, 2024.
- **Performance** — Surpasses GPT-3.5 Turbo in academic benchmarks and supports multiple languages.
- **Function Calling** — Demonstrates strong performance in function calling, enabling developers to build applications that fetch data or take actions with external systems.
- **Long-Context Performance** — Improved long-context performance compared to GPT-3.5 Turbo.
- **Language Support** — Supports the same range of languages as GPT-4o.
From learn.microsoft.com
GPT-4o mini surpasses GPT-3.5 Turbo and other small models on academic benchmarks across both textual intelligence and multimodal reasoning, and supports the same range of languages as GPT-4o.
learn.microsoft.comVerified
Arc Search read websites across the internet to make you this perfect tab.
Try it for freeResponsible AI Content Filtering
- **Policy Name** — `DefaultV2` policy.
- **Prompt Shields** — Protects against jailbreak attacks on user prompts.
- **Protected Material Detection** — Detects protected material in text and code completions.
- **Automatic Inheritance** — New resources and GPT deployments automatically inherit the new content filtering policy.
- **Custom Configurations** — Customers can switch between safety defaults and create custom content filtering configurations.
From learn.microsoft.com
The new default content filtering policy `DefaultV2` delivers the latest safety and security mitigations for the GPT model series (text), including: Prompt Shields for jailbreak attacks on user prompts (filter), Protected material detection for text (filter) on model completions, Protected material detection for code (annotate) on model completions.
learn.microsoft.comVerified
Assistants API Updates
- **File Search Tool** — Allows up to 10,000 files per assistant.
- **Vector Storage** — New support for vector storage.
- **Parameter Support** — Includes `temperature`, `top_p`, `response_format`, and more.
- **Streaming and Polling** — Helper functions in the Python SDK for creating runs and streaming responses.
- **Logic Apps and Function Calling** — Integration with Azure OpenAI Studio for invoking functions based on user prompts.
A refresh of the Assistants API is now publicly available. It contains the following updates: File search tool and vector storage, Max completion and max prompt token support for managing token usage, `tool_choice` parameter for forcing the Assistant to use a specified tool.
learn.microsoft.comVerified
