Researchers Let AI Go Rogue to See If They Could Stop It, They Couldn’t
Metro ‘I think our results indicate that we don’t currently have a good defence against deception in AI systems – either via model poisoning or emergent deception – other than [Read More]