1/2 "This is kind of shocking." Mikhail Belkin of @UCSD on how adding the same set of vectors to the activations of an LLM's transformer blocks can steer its output from, say, either English to Chinese, or Python to C++, at the Simons Institute. Video: simons.berkeley.edu/talks/mikhail-…

2/2 Mikhail Belkin spoke of how to determine such "steering" vectors using white-box LLMs and recursive feature machines at the Simons Institute's Smale@95: A Conference in Honor of Steve Smale. Video: simons.berkeley.edu/talks/mikhail-…

“We have tacitly abandoned certain public spaces to the most disordered and depraved among us because enforcing the law feels mean and makes us uncomfortable,” writes @katrosenfield.