Ask HN: Complete prompt: Fool me once, shame on you, fool me twice, ____

I am concerned about pervasive deployment of language model assisted applications to the public at large. Prompt engineering required for effective applications are ultimately a form of (crowd sourced) data labeling. The aggregate data collected by the LLM provider from all of these apps that will inevitably pop up is precious gold dust -- practically labeled matter ranging from "my mood" to "what my friends say" to "my wife likes" etc. Couple this with data from other sources (surveillance 1.0) and what do you have? Which of these corporations are going to resist (at the minimum) holding on to this data? This is data that will be fed into the next generation of AI. Who will throw it away? Either we have to move this capability to the edge devices (so as to take the central players out of the loop), or, we somehow have to figure out how to use these LLMs with obfuscated data. As it is, we will be spilling our guts (or your users') to an unprecedented degree, in a semi-regimented and semi-formal manner via prompts, to a handful of corporations. It would be good to have discussions around things such as an ethical guidelines for developers, howtos on practical sanitizing of data handed over to AI providers, etc. This is powerful technology that can have a significant positive impact for humanity but it needs to be used responsibly, as it can equally have a significant negative impact. And lets not wait 10 years after the excitement has worn off. We can not afford that.