Estimating the Current and Future Number of AI Safety Researchers — LessWrong