AI Roundup - Monday, November 20th 2023
OpenAI announces leadership transition
Sam Altman, CEO of OpenAI, will be departing the company, while Mira Murati, the company's chief technology officer, will serve as interim CEO. The board of directors has decided that Altman's departure is necessary due to his lack of consistent communication, hindering the board's ability to carry out its responsibilities. However, the board is grateful for Altman's contributions to the company and remains committed to its mission of ensuring that artificial general intelligence benefits humanity. Murati, who has been with OpenAI for five years and leads the company's research, product, and safety functions, is considered uniquely qualified for the interim CEO role. Meanwhile, the board will search for a permanent CEO. The majority of the board is independent, and the independent directors do not hold equity in OpenAI. OpenAI was founded as a nonprofit in 2015 and restructured in 2019 to allow for capital raising while preserving its mission and governance.
OpenAI Data Partnerships
OpenAI has announced a new initiative called OpenAI Data Partnerships, through which the company will collaborate with organizations to create public and private datasets for training AI models. OpenAI believes that in order to develop safe and beneficial artificial general intelligence (AGI), AI models must have a deep understanding of various subject matters, cultures, industries, and languages. By partnering with organizations and including their content in training datasets, OpenAI aims to improve AI models' understanding of specific domains and make them more helpful to different industries and communities. The company has already started collaborating with the Icelandic Government and Miðeind ehf to enhance GPT-4's capability to speak Icelandic, and with Free Law Project, a non-profit organization focused on making legal information accessible, by incorporating their collection of legal documents into AI training. OpenAI hopes that Data Partnerships will allow more organizations to contribute to AI research and benefit from models that cater to their specific needs.
Introducing GPTs
OpenAI assures users of ChatGPT that they are in control of their data. The chats with GPTs are not shared with builders, and users can choose whether their data is sent to third-party APIs. If builders customize their own GPT, they can decide if user chats can be used to train the models. OpenAI has implemented new systems to review GPTs against usage policies, focusing on preventing harmful content and activities. They have also taken steps to enhance user trust by allowing builders to verify their identity. OpenAI will continue monitoring how GPTs are used and will strengthen safety measures accordingly. Users can report concerns about specific GPTs using the reporting feature. OpenAI acknowledges that GPTs will continue to improve and become smarter, eventually being able to handle real-world tasks as "agents." However, they emphasize the need for incremental progress and consider the societal implications with careful technical and safety considerations. OpenAI plans to share more analysis on this topic soon.