Stochastic Flocks and the Critical Problem of 'Useful' AI

You might not agree, but there’s a growing consensus, even among reluctant users, that LLMs feel different. I don’t think it undermines critique to ask about the stakes of those feelings — and of this new goal of “agential AI.”

In my latest piece for Tech Policy Press, I examine what agentic AI is, what this agentic turn means for the industry, why people are responding to it as they are, and the risks we ought to watch for as the industry pursues this pivot.

There is a real backlash against the idea that this pivot even exists, and some say that acknowledging it is feeding the hype. I don't think critical thinking is ever hype, nor is acknowledging that the industry is building interfaces differently. Just as GPT3's shift to a ChatBot created a different emphasis for social interaction (creating sycophancy, illusions of mind, and centering the user as a subordinate), so I suspect will the shift to multimodal models that emphasize code.