On the limitations of quantum machine learning, watermarking GPT outputs, AI safety, the orthogonality thesis, and doomerist sentiment.
Love Aaronson's commitment to science!!! It was a pleasure to listen to him explain the limits of quantum computing, the prospects for AI safety, and his work at OpenAI. It is promising to hear that OpenAI is actively seeking some sort of computational altruism to embed into their LLMs. Such a funny idea for a deep humanities guy like myself! But why not? Computational altruism are just words we are using to describe a function or a process---some sort of stop gap measure. But as Aaronson and Winograd remind us, any trapdoor can be used for both good and bad. Ugh... human being!!!