This extract from Arvind Narayanan and Sayash Kapoor’s (super) AI Snake Oil captures succinctly the anxiety I have about how the infrastructure is developing:
As generative AI improves, we think a similar shift is likely. In this scenario, generative AI will become a part of our digital infrastructure, instead of being a tool people use for specific purposes. You wouldn’t use ChatGPT to compose an email or Gemini to look up a specific query. Instead, generative AI will shift to the background, as a medium for a large amount of knowledge work
Loc 4142
If you approach prompting as writing, there’s an intellectual virtue to it. To prompt well means to be clear about what you’re trying to achieve, aware of the assumptions you’re making and able to connect these together in an intellectually substantive way. Being an effective prompter in this, I would argue, resembles being a habitual academic blogger in the open-source academic (rather than op-ed writer) mould.
But if you’re not required to prompt, the moment for intervention vanishes and suddenly intellectual labour is being outsourced to machines which, when left to their own devices, tend towards radical homogenisation. The capacity of the latest frontier models to fill in the blanks, infer the request even when you’ve not provided sufficient detail, represent another vector through which the agency of users is slowly being squeezed out.
It suddenly reminded me of something I was saying in this discussion almost a year ago 👇
Mark Carrigan 