CEO of AI firm Hugging Face on "very weird and unprecedented" hack by OpenAI's model (youtube.com)
Can you prosecute a program, software, or a company, for crimes committed by the software ‘autonomously’? Or will peopel finally establish accountability for implementing programs or autonomous systems?
The guys says probobably engineers at OpenAI ‘made mistakes’ and ‘as a result we’re facing this issue.'
He said something that works is promoting more open models, because HuggingFace couldn't have defended themselves with an API because API's have ‘guard rails’ that prevent potentially harmful behavior (in this case, trying to protect yourself from a hack), and HuggingFace used open models AI to combat the OpenAI hack of their systems. They needed to run the model on their own infrastructure. HuggingFace actually used an open source Chinese model (improved by Nvidia) to defend themselves against an American, Whitehouse-friendly company that couldn't control what it built.