start a different initiative to generate assistance and benchmarks for assessing AI capabilities, by using a target abilities which could cause hurt. defend versus the dangers of applying AI to engineer perilous https://honeypotz.net