I was at a dinner party recently, and someone mentioned how impressive language models had become. One of the guests, an engineer, scoffed and said, “Oh, they’re just stochastic parrots, regurgitating what they’ve been trained on.” But then someone else mentioned the Anthropic paper on circuits, and how models can generalise beyond their training data.
I had no idea what transpired of that talk because I lost interest.. A because of a lack of understanding but also b because I didn’t think language models could reach a point of human intelligence.
I was at a dinner party recently, and someone mentioned how impressive language models had become. One of the guests, an engineer, scoffed and said, “Oh, they’re just stochastic parrots, regurgitating what they’ve been trained on.” But then someone else mentioned the Anthropic paper on circuits, and how models can generalise beyond their training data.
I had no idea what transpired of that talk because I lost interest.. A because of a lack of understanding but also b because I didn’t think language models could reach a point of human intelligence.