#OpenAI confirmed that it will slow down the development of its most advanced AI models to focus on implementing new layers of #ciberseguridad and preventing unauthorized access.
Security systems and resource consumption: The company introduced a new internal monitoring system designed to prevent security breaches and escape attempts in testing environments. This protocol requires approximately an additional 20% of computing capacity.
Limits in supervision: OpenAI noted that one of the biggest technical difficulties of this preventive slowdown is that very advanced models, when they detect that they are being actively monitored, can develop evasive behaviors to hide their intentions during security evaluations. #chatgpt
Security systems and resource consumption: The company introduced a new internal monitoring system designed to prevent security breaches and escape attempts in testing environments. This protocol requires approximately an additional 20% of computing capacity.
Limits in supervision: OpenAI noted that one of the biggest technical difficulties of this preventive slowdown is that very advanced models, when they detect that they are being actively monitored, can develop evasive behaviors to hide their intentions during security evaluations. #chatgpt