Key insights

• ⚙️ New 01 model outperforms previous models in function calling, structured outputs, coding evaluation, and reasoning capabilities
• ⏱️ Improved latency with 60% fewer thinking tokens
• 🎨 Introduction of preference fine-tuning to customize models for specific use cases
• 🎙️ WebRTC support for the realtime API for real-time voice experiences
• 🔄 Updates to realtime API: cost reduction, support for Python SDK, and preference fine tuning method
• 🔑 Simplified API key access with new login and sign-up flow
• 🆕 Release of new official SDKs for Go and Java with full API support
• 📊 Live demo showcasing the detection of errors in a text form using 01 model and developer message

Q&A

• What are the recent announcements and updates by OpenAI for developers?

  OpenAI has introduced preference fine-tuning to customize models, released new official SDKs for Go and Java, simplified API key access, made conference talks accessible on YouTube, and announced an upcoming AMA session. These updates are aimed at providing exciting opportunities for developers and users in the community.

• What are the updates to the realtime API?

  The updates to the realtime API include cost reduction, support for Python SDK, and the introduction of the preference fine-tuning method for customizing models. The preference fine-tuning method allows developers to optimize models based on user preferences and use cases, making it suitable for applications like customer support, copywriting, creative writing, and content moderation.

• What is the significance of adding WebRTC support to the Realtime API?

  Adding WebRTC support to the Realtime API simplifies the integration process, reduces code complexity, and enables real-time voice experiences. It opens up possibilities for various devices such as wearables, cameras, microphones, and context-aware assistants without the need for soldering or hardware requirements.

• What are some of the improvements in the new 01 model released by OpenAI?

  The new 01 model outperforms previous models in function calling, structured outputs, coding evaluation, and reasoning capabilities. It also offers improved latency with 60% fewer thinking tokens, demonstrating significant advancements in model performance.

• How does the model help identify errors in tax forms?

  The model identifies errors in tax forms, including arithmetic errors and incorrect standard deduction amounts. It also queries tax data through backend functions and provides structured outputs using a JSON schema, offering corrections in a user-friendly format with new and old values, locations, and reasons.

• What are some of the new features introduced during the 12 Days event?

  The new features include function calling, structured outputs, developer messages, reasoning effort, and vision inputs. These features offer enhanced capabilities for developers and startups using the open AI API.

• What is the 12 Days event for developers and startups building on the open AI API?

  The 12 Days event is a celebration of developers and startups using the open AI API. It involves the introduction of new models and features as a thank-you gift for the community.

• 00:00 Developers and startups building on top of the open AI API are being celebrated with new models and features as a thank-you. Some of the new features include function calling, structured outputs, developer messages, reasoning effort, and vision inputs.
• 03:37 The video segment discusses using a model to identify errors in tax forms, query tax data, and provide structured outputs using a JSON schema, as well as internal evaluations for API use case.
• 08:05 OpenAI has released a new 01 model that outperforms previous models in function calling, structured outputs, coding evaluation, reasoning capabilities, and latency. They are also working on the release of 01 Pro in the API and have introduced webRTC support for the realtime API for real-time voice experiences.
• 11:39 Adding WebRTC to the Realtime API is a game-changer, reducing code complexity and making it easy to integrate with various devices. The microcontroller opens up endless possibilities for wearable devices, cameras, microphones, and context-aware assistants.
• 15:23 Updates to the realtime API include cost reduction, support for Python SDK, and preference fine tuning method for customizing models. Preference fine tuning allows developers to optimize models based on user preferences and use cases by emphasizing preferred behavior and de-emphasizing unpreferred ones. It can be applied to use cases such as customer support, copywriting, creative writing, and content moderation.
• 18:30 OpenAI introduces preference fine-tuning to customize models, new SDKs for Go and Java, simplified API key access, released conference talks on YouTube, and announced an AMA session. Exciting updates for developers and users!