Anthropic reveals that as few as '250 malicious documents' are all it takes to poison an LLM's training data, regardless of model size

Claude-creator Anthropic has found that it's actually easier to 'poison' Large Language Models than previously thought. In a ...