OpenAI DevDay, Opening Keynote

GPT 4-Turbo

• Now available through the API: You can now use GPT 4-Turbo via the API.
• Offers an extended 128,000-token context: Previously, it was 32,000 tokens.
• Over 50% reduction in cost compared to GPT-4.
• Knowledge updated up to April 2023: Previously, it was only up to September 2021.
• Outperforms GPT-4 in performance.
• API now supports simultaneous text and image inputs.
• New JSON mode for pure JSON responses from GPT.
• More relaxed rate limits.

Custom GPTs

• Users can build "Custom GPTs" for specific tasks.
• Create Custom GPTs using natural language, no coding required, and upload files for context.
• Businesses can create proprietary Custom-GPTs for their company and organization.
• OpenAI provides two sample Custom GPTs: Canva and ZapierAI.

Custom GPT Store

• Users can upload their Custom GPTs to the store for others to use.
• OpenAI will offer a revenue-sharing program, allowing model authors to earn income.

Assistant API

• The Assistant API allows you to build autonomous agents with access to tools.
• OpenAI currently offers three tools: Code Interpreter (programming), Retrieval (custom knowledge), and Function Calls.
• Define their roles through custom instructions, just like using a regular API.

High-Quality Text-to-Speech Synthesis

• OpenAI has released the tts-1 and tts-1-hd models.
• tts-1 model prioritizes speed, while the tts-1-hd model prioritizes quality.
• Choose from six voice types and create lifelike voices through the API.

Copyright Protection

• The copyright protection feature shields you and your company from copyright claims when using OpenAI's products.

Whisper V3

• Whisper is OpenAI's speech-to-text model that transcribes audio into text.
• Whisper is open-source, and V3 is also released in open-source form.
• Whisper v3 is not yet available through the API (paid).

Enterprise Custom Models

• OpenAI's research team will create enterprise custom models with specific domain knowledge for specific companies.