Never before has it been easier to clone a human voice. New AI tools can take a voice sample, process it, copy it, and say anything in the voice of the original. It's been a thing since as early as 2018, but modern tools can do it faster, more accurately, and with greater ease. OpenAI, the artificial intelligence company behind ChatGPT, presented a project this year that showed how it's possible to clone a voice with nothing more than a 15-second recording. OpenAI's tool isn't yet publicly available and it's said to have security measures in place to prevent misuse.

This week, we're helping you round out your summer reading lists by asking some of our favorite authors to step in as Prudie for the day and give you advice. This is part of our Guest Prudie series. Today's columnist is American author and "King of Horror" Stephen King, whose renowned for his horror, supernatural fiction, suspense, crime, science-fiction, and fantasy novels, including It, The Shining, Carrie, and many more. His iconic books and stories have been adapted into numerous films and television series--including The Boogeyman which was released just last month. His new novel, Holly, hits shelves this coming September.

Chain-of-Thought (CoT) prompting can dramatically improve the multi-step reasoning abilities of large language models (LLMs). CoT explicitly encourages the LLM to generate intermediate rationales for solving a problem, by providing a series of reasoning steps in the demonstrations. Despite its success, there is still little understanding of what makes CoT prompting effective and which aspects of the demonstrated reasoning steps contribute to its performance. In this paper, we show that CoT reasoning is possible even with invalid demonstrations - prompting with invalid reasoning steps can achieve over 80-90% of the performance obtained using CoT under various metrics, while still generating coherent lines of reasoning during inference. Further experiments show that other aspects of the rationales, such as being relevant to the query and correctly ordering the reasoning steps, are much more important for effective CoT reasoning. Overall, these findings both deepen our understanding of CoT prompting, and open up new questions regarding LLMs' capability to learn to reason in context.

In 1977, Shigeru Miyamoto joined Nintendo, a company then known for selling toys, playing cards, and trivial novelties. Miyamoto was twenty-four, fresh out of art school. His employer, inspired by the success of a California company named Atari, was hoping to expand into video games. Miyamoto began tinkering with a story about a carpenter, a damsel in distress, and a giant ape. Four years later, Miyamoto had turned the carpenter into a plumber; Mario, and the Super Mario Bros. franchise, had arrived.

Imagine you wired hundreds of thousands of dollars somewhere based on a call from your boss, whose voice you recognized, only to find out you were talking to a machine and the money is lost. One company executive doesn't have to imagine it happening. He and his company were victims of what some experts say is one of the first cases of voice-mimicking software, a form of artificial intelligence (AI), being used in a scam. In a common version of the Business Email Compromise scam, an employee in a company's accounting department wires money somewhere based on what appears to be a legitimate email from the CEO, CFO or other high-ranking executive. I wrote a column last year noting that reported losses to the scam had grown from $226 million in 2014 to $676 million in 2017.