Sam Altman has returned as CEO of OpenAI, accompanied by fundamental leadership changes and a newly formed initial board. Altman expressed his excitement about the future and gratitude for the team's hard work during a challenging period. The new initial board comprises Bret Taylor as Chair, Larry Summers, and Adam D'Angelo.
Altman thanked the outgoing CEO, Ilya Sutskever, emphasizing his respect and admiration for Sutskever's contributions to the field. Ilya will no longer serve on the board, but discussions are ongoing for his continued involvement with OpenAI.
Acknowledging the efforts of various team members, Altman highlighted Mira Murati's outstanding leadership as CTO and expressed gratitude to Adam, Tasha, and Helen for their collaborative efforts in reaching a solution that aligns with OpenAI's mission.
Altman reassured the team that the company was in excellent hands, praising the leadership team's readiness to run the company independently. He also appreciated the support from partners and users, noting that the outpouring of support helped the team get through the challenging period.
The new board, consisting of Bret, Larry, and Adam, will focus on crucial tasks, including building a diverse board, improving governance, and overseeing an independent review of recent events. Altman outlined three immediate priorities for OpenAI: advancing the research plan, investing in full-stack safety efforts, improving and deploying products, and serving customers.
Bret Taylor, in a message to the company on behalf of the OpenAI Board, expressed gratitude for the collective efforts of the OpenAI community in finding a path forward. The board is committed to strengthening OpenAI's corporate governance, including building a qualified and diverse board, stabilizing the organization, and enhancing the governance structure for the trust of all stakeholders.
The return of Altman, coupled with the leadership changes and governance enhancements, signals a new chapter for OpenAI as it continues its mission to ensure artificial general intelligence benefits humanity.
OpenAI has announced the integration of new voice and image capabilities in ChatGPT, offering users a more immersive and interactive experience. This development allows users to engage in voice conversations with ChatGPT and share images, opening up new possibilities for communication and problem-solving.
The voice feature lets users initiate back-and-forth conversations with ChatGPT, making it a versatile assistant for on-the-go interactions. Users can request information, ask questions, or even have ChatGPT narrate stories. The voice capability is available on iOS and Android, with users having the option to choose from five different voices. The new text-to-speech model powering this feature creates human-like audio based on text input and a short speech sample, ensuring a natural and engaging conversation.
For the image feature, users can now show ChatGPT images to troubleshoot issues, analyze data, or discuss various topics. This capability is powered by multimodal models, specifically GPT-3.5 and GPT-4, which leverage language reasoning skills to interpret images alongside text. The drawing tool in the mobile app allows users to focus on particular parts of an image, enhancing communication and collaboration.
Over the next two weeks, OpenAI will gradually roll out these voice and image capabilities to Plus and Enterprise users. Voice is available on iOS and Android platforms, while images can be used across all platforms.
The deployment of these advanced features aligns with OpenAI's strategy of gradual release to refine risk mitigations, gather user feedback, and ensure responsible usage. The voice technology, capable of crafting realistic synthetic voices, is initially focused on voice chat applications, leveraging collaboration with voice actors and partners like Spotify for specific use cases like Voice Translation in podcasts.
Similarly, image input undergoes careful testing and red teaming to address risks related to model interpretation and privacy concerns. OpenAI emphasizes the importance of real-world usage and feedback to enhance safeguards while maintaining the tool's utility.
These features will expand beyond Plus and Enterprise users, reaching developers and other user groups shortly.
OpenAI has introduced an exciting new feature in ChatGPT, enabling Plus and Enterprise users to create unique images through conversations with the powerful DALL·E 3 image model. This innovative capability allows users to describe their vision, and ChatGPT will generate visuals, offering a selection for users to refine and iterate upon within the chat.
Whether for a science project, website design, business logo, or any creative endeavor, ChatGPT with DALL·E 3 can bring your ideas to life. Users can specify their requirements, and the model provides a variety of images for users to choose from. The process allows for collaboration and refinement directly within the chat, making it an intuitive and interactive experience.
DALL·E 3 represents a significant advancement in image generation, incorporating research advancements within and outside OpenAI. Compared to its predecessor, DALL·E 3 produces visually striking and more detailed images, capturing intricate details such as text, hands, and faces. The model is responsive to extensive and detailed prompts, supporting both landscape and portrait aspect ratios.
The responsible development and deployment of DALL·E 3 involve a multi-tiered safety system to prevent the generation of potentially harmful content. Safety checks are conducted on user prompts and the resulting imagery, and feedback from early users and red-team evaluations have contributed to addressing gaps in safety systems. OpenAI has taken steps to limit the likelihood of generating content resembling living artists' styles and images of public figures while improving demographic representation across generated images.
OpenAI is also exploring a provenance classifier, a tool that can identify whether an image was generated by DALL·E 3. While the classifier shows promising results in internal testing, OpenAI emphasizes collaboration across the AI value chain to develop effective techniques for identifying AI-generated content.
Additionally, DALL·E 3 is equipped with creative controls to decline requests for images in the style of living artists, and creators can opt their images out from training future image generation models. This commitment to responsible AI development aligns with OpenAI's mission to provide robust and creative tools while prioritizing safety and ethical considerations.