We believe that a practical approach to addressing AI security concerns is to dedicate more time and resources to researching effective mitigation and alignment techniques and testing them against real-world abuse.
Importantly, we also believe that enhancing security and AI capabilities must go hand in hand. Our best safety work to date has come from working with our most capable models because they are better at following user instructions and easier to steer or “guide”.
We will take greater care with building and deploying more capable models, and will continue to increase safety precautions as our AI systems evolve.
While we’ve waited over 6 months to deploy GPT-4 to better understand its capabilities, benefits, and risks, it can sometimes take longer than that to improve the security of AI systems. Therefore, policy makers and AI providers need to ensure that the development and deployment of AI is effectively managed on a global scale, so that no one is taking shortcuts to get ahead. This is a daunting challenge that requires technical and institutional innovation, but we want to contribute.
Addressing safety issues also requires extensive debate, experimentation, and engagement, including the behavior constraints of AI systems. We have and will continue to encourage open collaboration and dialogue among stakeholders to create a secure AI ecosystem.