Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only