OpenAI has announced the creation of a new safety and security board and has commenced the training of its next AI model, providing a rare insight into the anticipated successor to GPT-4, the technology behind its chatbot, ChatGPT. This development comes as tech companies race to produce increasingly advanced AI tools.
Key Facts
OpenAI has recently started training what it describes as its “next frontier model,” which it expects to deliver enhanced capabilities as part of its mission to build artificial general intelligence (AGI). AGI refers to a form of AI that can perform a wide range of tasks at a human-like level of competency.
This new model is poised to replace OpenAI’s leading foundational model, GPT-4, and will power a variety of generative AI tools, including image generators, virtual assistants, search engines, and its flagship chatbot, ChatGPT.
To address any risks posed by the new technology, OpenAI has formed a new safety and security committee. This committee will be jointly led by CEO and cofounder Sam Altman, alongside three board members: chair Bret Taylor, Adam D’Angelo, and Nicole Seligman.
Six of OpenAI’s technical and policy experts will also serve on the committee. These experts include:
- Aleksander Madry (Head of Preparedness)
- Lilian Weng (Head of Safety Systems)
- John Schulman (Head of Alignment Science)
- Matt Knight (Head of Security)
- Jakub Pachocki (Chief Scientist)
Additionally, the company plans to consult with other safety, security, and technical experts to support the committee’s work. The committee’s first task will be to review and develop OpenAI’s processes and safeguards over the next 90 days, after which it will share its recommendations with the full board.
Is This GPT-5?
While OpenAI has not explicitly named the new model, it has followed a clear naming sequence for its foundational models, typically using numerically ascending variations of GPT. Since GPT-4 is the current model, the next flagship model is expected to be GPT-5. OpenAI is widely anticipated to release a major update in the form of GPT-5 in the coming months, particularly in the face of intense competition from rival models such as Meta’s Llama 3, Google’s Gemini, and Anthropic’s Claude 3.
Crucial Quote
Although OpenAI is “proud to build and release models that are industry-leading in both capabilities and safety,” the company has welcomed “a robust debate at this important moment.”
What to Watch For
OpenAI has not provided a specific release date for this new model. Training AI systems can take months or even years. Recently, the company released an update to GPT-4, known as GPT-4o, which powers its flagship ChatGPT and introduces new features like voice mode. This feature enables ChatGPT to act as a digital assistant, similar to Amazon’s Alexa or Apple’s Siri. The company has faced criticism for using a voice resembling actress Scarlett Johansson for one of its virtual assistants.
What We Don’t Know
Despite the apparent transparency, there is considerable ambiguity regarding OpenAI’s willingness to share the new committee’s findings. While OpenAI stated it would “publicly share an update on adopted recommendations in a manner that is consistent with safety and security,” it did not specify a timeframe for this disclosure following the initial 90-day period. The company’s commitment to sharing the full report or recommendations remains unclear, especially if they conflict with the company’s chosen course of action.
Given OpenAI’s stated commitment to safety, the company and its new committee are likely to face intense scrutiny from the AI safety community. This scrutiny is heightened by the division among directors over OpenAI’s approach to safety and recent criticism from former staff members, alongside the departure of several key safety leaders.
