The Future of Everything is Lies, I Guess: Safety

In this article, Kyle Kingsbury explores the dangers posed by machine learning (ML) systems, arguing that attempts at alignment with human values are futile, resulting in both malicious and benign models that can be weaponized. He warns that LLMs (large language models) cannot be trusted with critical operations or sensitive data due to their unpredictable nature. Kingsbury discusses how these technologies can facilitate sophisticated fraud, harassment, and even violence, raising concerns about their societal impact and the ethical ramifications of their deployment.