An Overview of OpenAI's DevDay Announcements
I had the opportunity to attend OpenAI's first DevDay, and I wanted to provide an overview of what they announced. Additionally, the full keynote is available online, and OpenAI's detailed description of the announcements can be found on their blog. If you want to learn more about what OpenAI discussed in their breakout sessions and my key takeaways, I have a blog post titled My Takeaways from OpenAI's DevDay.
The New GPT-4 Turbo Model
A highlight for many, including myself, was the introduction of GPT-4 Turbo. OpenAI expanded the context length to 128,000 tokens which is equivalent to about 300 pages of text, allowing us to push the boundaries of tailored prompts and nuanced few-shot learning.
Better Control
Control is critical in development, and OpenAI took a step forward in enhancing control for developers. The new JSON mode helps ensure structured data as a response, making integrations smoother and more predictable. Also, the ability to call multiple functions simultaneously will help streamline workflows.
Consistent Outputs
Consistency is key, and the new controls for consistent outputs mean we can better replicate AI responses. The enhanced logging functionalities will give us valuable data to refine our applications further and ensure responses are behaving as intended.
Knowledge Base Expansion
OpenAI said they are committed to never letting the knowledge base grow as stale as it previously has. They updated the knowledge base up to April 2023. The announcement will help AI interactions be well informed and as up-to-date as possible.
Whisper V3 and Unified Modalities
The introduction of Whisper V3 and its improved, more realistic voices is cool. But what excited me more was the seamless integration of modalities on ChatGPT. They got rid of the mode picker – the AI now fluidly decides the best modality for the task at hand.
Copyright Shield and Pricing
The Copyright Shield announcement shows OpenAI's commitment to defending its modal's outputs. OpenAI will take on legal costs users get from copyright claims from GPT material. And the new pricing makes AI more accessible, allowing more developers to bring their ideas to life. GPT-4 Turbo is 3x cheaper for prompt tokens and 2x cheaper for completion tokens than GPT-4. They also decreased the price of the GPT-3 Turbo API.
The Rise of GPTs and the GPT Store
Customizable GPTs represent a new frontier for personalization in AI. The upcoming GPT store echoes the early days of the app revolution – a place for discovery and innovation.
The Assistance API
The Assistance API simplifies the integration of AI into our apps, streamlining what used to be complex tasks. It's like having an AI Swiss Army knife, ready to adapt to our project's needs.