Keep pulling the thread on Mira Murati.
AI systems are predicted to achieve PhD-level intelligence for specific tasks within the next 1.5 to 2 years.
Future AI systems will have agentic capabilities, allowing them to connect to the internet and collaborate with each other and with humans.
OpenAI is developing a "science of capability prediction" to forecast what models will be able to do before their training is complete.
OpenAI has been advocating for more regulation on frontier AI models due to their high capabilities and potential for misuse.
OpenAI's internal safety work informed the White House commitments on AI and the UN Commission's principles for AI deployment.
One of the most powerful applications of AI will be in education, with the potential to provide high-quality, accessible, and free customized learning to anyone in the world.
OpenAI is developing a customization layer to allow communities like schools, churches, or countries to implement their own specific values on top of the default AI system.
OpenAI has developed a 'spec' that acts as a living constitution for its AI systems to provide transparency into their embedded values.
OpenAI is working on research for content provenance and authenticity tools, such as watermarking, to help people identify deepfakes and misinformation.
OpenAI has intensified its work on combating misinformation in 2024 because it is a global election year.
The performance of AI systems improves as more data and compute are added, a principle referred to as "scaling laws."
OpenAI's first product was an API, and the company initially found it very difficult to commercialize its GPT-3 technology.