We believe that a practical approach to solving AI security problems is to devote more time and resources to researching effective mitigations and alignment techniques and testing them against real-world abuses.
Importantly, we also believe that improving AI security and capabilities should go hand in hand. Our best security work to date has come from working with our most capable models because they follow user instructions better and are easier to lead or “guide”.
We will be more cautious in building and deploying more capable models and will continue to strengthen security precautions as our AI systems evolve.
While we waited more than 6 months to deploy GPT-4 to better understand its capabilities, benefits, and risks, sometimes it may take longer to improve the security of AI systems. Therefore, policymakers and AI providers must ensure that AI development and deployment is managed effectively on a global scale, so that no one cuts corners to get ahead. It’s a daunting challenge that requires both technical and institutional innovation, but it’s one we want to contribute to.
Addressing security issues also requires extensive debate, experimentation, and engagement, including on the limits of AI system behavior. We have and will continue to foster collaboration and open dialogue between stakeholders to build a secure AI ecosystem.