J. Bostock

J Bostock, author of the LessWrong post on steering LLMs, is a researcher analyzing activation steering and eval-awareness. He warns single-vector steering may soon fail. He compares steering models to steering humans and explains why larger systems resist simple fixes. He surveys experiments showing steering harms and outlines more robust alternatives to consider.

J. Bostock, a contributor to LessWrong, explores a troubling reasoning flaw known as 'exhaustive free association,' frequently seen in rationalist communities. He illustrates how this pattern misleads by inducing false conclusions through incomplete logic. Bostock critiques superforecasters for underestimating AI risks and discusses the implications of exhaustive reasoning on welfare estimates. The episode dives into the persuasive nature of bad arguments and emphasizes the importance of challenging this reasoning style to enhance discourse.