[philipodonnell]

There’s a trope where the best white hat is a former black hat because they can recognize all the tricks, I wonder if training an LLM to be evil and then fine tuning it to be good will produce more secure code than the opposite?