We consider that a practical approach to solving AI safety concerns is to dedicate more time and resources to researching effective mitigations and alignment techniques and testing them against real-world abuse.

Importantly, we also consider that improving AI safety and capabilities should go hand in hand. Our greatest safety work up to now has come from working with our most capable models because they’re higher at following users’ instructions and easier to steer or “guide.”

We might be increasingly cautious with the creation and deployment of more capable models, and can proceed to boost safety precautions as our AI systems evolve.

While we waited over 6 months to deploy GPT-4 to be able to higher understand its capabilities, advantages, and risks, it might sometimes be obligatory to take longer than that to enhance AI systems’ safety. Due to this fact, policymakers and AI providers might want to make sure that AI development and deployment is governed effectively at a world scale, so nobody cuts corners to get ahead. This can be a daunting challenge requiring each technical and institutional innovation, but it surely’s one which we’re desperate to contribute to.

Addressing questions of safety also requires extensive debate, experimentation, and engagement, including on the bounds of AI system behavior. We have now and can proceed to foster collaboration and open dialogue amongst stakeholders to create a secure AI ecosystem.