AI Roundup - Monday, November 13th 2023
OpenAI Data Partnerships
OpenAI is launching Data Partnerships, a program that encourages collaboration with organizations to develop public and private datasets for training AI models. The goal is to create AI models that possess a deep understanding of various subjects, industries, languages, and cultures in order to ensure the development of safe and beneficial artificial general intelligence (AGI). OpenAI aims to increase the usefulness of AI models for individuals by incorporating their specific domain expertise into the training data. They have already formed partnerships with diverse entities such as the Icelandic Government and Miðeind ehf to enhance GPT-4's ability to speak Icelandic, as well as with the non-profit Free Law Project to include a broad collection of legal documents. OpenAI believes there are many more organizations eager to contribute unique datasets and shape the future of AI research. The Data Partnerships initiative aims to facilitate such collaborations and enable more organizations to benefit from AI models that better align with their needs and interests.
Introducing GPTs
OpenAI reaffirms their commitment to user privacy and data control with ChatGPT. They assure users that their chats with GPTs are not shared with builders. Users have the choice to allow or disallow data transmission to third-party APIs used by GPTs. Builders, who customize their own GPTs, can decide whether user chats can contribute to model training. OpenAI also implements systems to review GPTs for compliance with usage policies and to prevent the sharing of harmful GPTs involving fraudulent, hateful, or adult content. Steps have been taken to enhance user trust, including offering identity verification for builders. OpenAI will continue to improve safety measures based on user feedback and adapt to the evolving ways people use GPTs. They acknowledge the potential for GPTs to become more capable "agents" in the future but emphasize the need for cautious progress, technical advancements, safety considerations, and societal adaptation. OpenAI plans to share further analysis on the societal implications soon.
New models and developer products announced at DevDay
OpenAI has released the Assistants API, which allows developers to build agent-like AI experiences in their applications. An assistant is an AI that follows specific instructions, utilizes additional knowledge, and is capable of performing tasks using models and tools. The Assistants API offers features like Code Interpreter and Retrieval, making it easier for developers to create high-quality AI apps. The API is flexible and can be used for various applications such as natural language-based data analysis, coding assistance, vacation planning, voice-controlled DJing, and more. A significant change introduced by the Assistants API is persistent and infinitely long threads, enabling developers to delegate thread state management to OpenAI and work around context window limitations. The API provides access to tools like Code Interpreter, which writes and runs Python code, Retrieval, which incorporates external knowledge into the assistant, and Function Calling, which allows assistants to invoke user-defined functions. As with other OpenAI products, data and files shared with the API are not used to train their models, and developers can delete data when needed. The Assistants API beta can be tested using the Assistants playground.