Keep pulling the thread on Mira Murati.
A foundational, medium-term goal for OpenAI is to build a "science of prediction" to forecast the capabilities and risks of future AI models before they are trained.
OpenAI has invested significant compute and research resources into a "super alignment" team focused on aligning future AGI or ASI systems.
OpenAI spent six months on safety work, including extensive red teaming and evaluations, before releasing GPT-4.
OpenAI has been advocating for government regulation focused on "frontier" AI systems.
OpenAI plans to launch a marketplace where creators can publish their custom GPTs and be compensated.
OpenAI's current research direction involves integrating multiple modalities, such as text, images, voice, and video, into its large-scale models.
One of OpenAI's core alignment research agendas is to build an "AI alignment researcher," an AI system that can help advance alignment research.
OpenAI has not open-sourced models like GPT-4 because it would be unable to enforce the safety guardrails that were developed over a long period.
OpenAI's new product interface for creating custom GPTs is a step in the direction of building AI agents.
AI agents will become ubiquitous in creative and work spaces and will be something people won't be able to live without.
OpenAI's strategy is to move carefully and iteratively towards a world with autonomous AI agents to manage risks and societal implications.
OpenAI's strategy of iterative deployment for new products like custom GPTs is designed to learn from user behavior and allow for continuous societal adaptation.