AI Roundup - Friday, November 10th 2023
OpenAI Data Partnerships
OpenAI has announced OpenAI Data Partnerships, a new initiative in which organizations can collaborate with OpenAI to create public and private datasets for training AI models. The goal is to develop AI systems that deeply understand various subject matters, industries, cultures, and languages. OpenAI believes that broad and diverse training datasets are necessary to build artificial general intelligence (AGI) that is safe and beneficial for humanity. By including their content in AI models, organizations can make these models more helpful and improve their understanding of specific domains. OpenAI has already partnered with several organizations, such as the Icelandic Government and Miðeind ehf, to enhance GPT-4's ability to speak Icelandic. They have also collaborated with the Free Law Project to integrate a large collection of legal documents into AI training. Through Data Partnerships, OpenAI aims to involve more organizations in shaping the future of AI research and offer them models that are more relevant and useful to their specific needs.
Introducing GPTs
ChatGPT puts you in control of your data, with chats not shared with builders. Third-party API usage is also under your control. If builders customize their own GPTs, they choose whether user chats can be used for model training. Privacy controls include the option to opt out of model training entirely. New systems have been implemented to review GPTs against usage policies and prevent harmful and fraudulent content. User trust is being built by allowing builders to verify their identity. Safety measures will continue to be monitored and strengthened. Specific GPTs can be reported using the reporting feature. The goal is to eventually allow GPTs to perform real-world tasks, but this will require careful technical and safety considerations. Societal implications are being deeply considered, with further analysis to be shared soon.
New models and developer products announced at DevDay
OpenAI has launched its Assistants API, which enables developers to create AI-powered applications with built-in assistant capabilities. The API offers features such as Code Interpreter, Retrieval, and function calling, which help streamline the development process and enhance the capabilities of AI apps. With the Assistants API, developers can create purpose-built AI assistants that can perform specific tasks, leverage additional knowledge, and call models and tools to accomplish objectives. This flexibility opens the door to a wide range of applications, from natural language data analysis apps to coding assistants, voice-controlled DJs, and more. One of the key changes introduced by the Assistants API is the ability to have persistent and infinitely long threads, allowing developers to hand off thread state management to OpenAI and work around context window constraints. Additionally, the API provides access to tools such as the Code Interpreter, which can write and run Python code in a sandboxed execution environment, generate graphs and charts, and process diverse data formats. The Retrieval feature augments the assistant with external knowledge, while function calling enables assistants to invoke user-defined functions and incorporate their responses in messages. OpenAI emphasizes that data and files passed to the API are not used to train models, and developers have control over deleting the data when desired. The Assistants API beta is available for experimentation in the Assistants playground.