AI Roundup - Friday, November 17th 2023
OpenAI Data Partnerships
OpenAI is launching Data Partnerships to collaborate with organizations in creating public and private datasets for training AI models. The goal is to develop artificial general intelligence (AGI) that is safe and beneficial for humanity, and that requires AI models to have a deep understanding of various subject matters, industries, cultures, and languages. By including diverse content in the training datasets, AI models can become more helpful and knowledgeable. OpenAI already has partnerships with several organizations. For instance, they have teamed up with the Icelandic Government and Miðeind ehf to enhance GPT-4's proficiency in Icelandic by incorporating curated Icelandic datasets. Another collaboration is with the non-profit organization Free Law Project, which provides access to a large collection of legal documents to contribute to AI training. OpenAI believes that there are many more organizations willing to contribute their unique data and shape the future of AI research and development. Data Partnerships aim to facilitate these collaborations and enable organizations to benefit from AI models that are relevant to their specific domains.
Introducing GPTs
ChatGPT has implemented new privacy controls to ensure that user data is protected. Each user has control over their data, and their chats with GPTs are not shared with the builders. Users also have the option to decide whether data can be sent to third-party APIs used by GPTs. Builders have the choice to use user chats to improve and train models when they customize their own GPT with actions or knowledge. Users can opt their entire account out of model training if they prefer. To uphold safety and usage policies, GPTs are reviewed against these policies. The systems in place aim to prevent the sharing of harmful GPTs that involve fraudulent activity, hate speech, or adult content. Builders can also verify their identity to further build user trust. Ongoing monitoring and learning from how GPTs are used help in updating and strengthening safety measures. While GPTs continue to become more intelligent and useful, OpenAI recognizes the need for careful technical and safety work before these systems, referred to as "agents", can take on real-world tasks. OpenAI is deeply considering the societal implications and plans to provide further analysis in the future.
New models and developer products announced at DevDay
OpenAI has launched the Assistants API, aimed at helping developers create AI-powered agent-like experiences in their applications. An assistant, in this context, is a purpose-built AI that can follow specific instructions, leverage additional knowledge, and utilize models and tools to perform tasks. The Assistants API provides new capabilities such as Code Interpreter and Retrieval, along with function calling, to streamline the development process for building high-quality AI applications. The API allows developers to add new messages to an existing conversation, while also granting assistants the ability to call on additional tools, such as the Code Interpreter for running Python code and generating graphs, and the Retrieval feature for incorporating external knowledge. The API also introduces persistent and infinitely long threads, enabling developers to pass thread state management to OpenAI and work around context window limitations. Data and files passed to the API are not used to train OpenAI models, and developers can delete the data as needed. Developers can try the Assistants API beta version in the Assistants playground.