OpenAI recently unveiled a series of updates and new features at its first conference, OpenAI DevDay. These announcements promise to transform interactions with artificial intelligence models and significantly broaden their accessibility and applications.
GPT-4 Turbo: The Advanced Model within Reach
The major revelation was the GPT-4 Turbo, the most advanced LLM to date, distinguished by a context window of 128K and knowledge of global events up to April 2023. OpenAI has significantly reduced costs, making this model 3 times cheaper for input tokens and 2 times cheaper for output tokens compared to the previous GPT-4. Additionally, the function calling feature has been improved, allowing multiple calls in a single message and ensuring the validity and accuracy of function parameters.
Improvements to GPT-3.5 Turbo
The new gpt-3.5-turbo-1106 now supports a default context of 16K and offers this extension at a reduced cost. Customization of this model is also available, with a significant reduction in operational costs. This model joins GPT-4 Turbo with enhanced function calling and reproducible outputs.
The Assistants API: Effortlessly Create Agent-Like Experiences
The Assistants API, currently in beta, is designed to facilitate the integration of agent-like experiences in your applications. It allows the creation of specialized AI assistants capable of following precise instructions, leveraging additional knowledge, and interacting with models and tools to accomplish various tasks. Developers can also benefit from Persistent Threads for better state management and bypassing context window constraints.
Multimodal Capabilities
GPT-4 Turbo now integrates visual inputs in the Chat Completions API, paving the way for use cases such as caption generation and visual analysis. Vision features will be incorporated into the production version of GPT-4 Turbo later this year. Furthermore, the integration of DALL·E 3 for image generation and new speech synthesis capabilities enrich the range of possible applications.
Customizing GPTs in ChatGPT
OpenAI has launched a new feature called GPTs, allowing the combination of instructions, data, and capabilities into a customized version of ChatGPT. Developers can now control a larger part of the user experience by easily transforming existing plugins into actions.
These advancements mark a turning point in the accessibility and versatility of OpenAI’s AI models, promising deeper and more intuitive integration of artificial intelligence into our everyday and professional lives. The future of AI is not only bright but also within reach.
