I have a lot of feelings about studying generative ai and LLMs. How in order to understand how they work you have to use them, repeatedly and intentionally, and therefore repeatedly and intentionally perpetrate their harms.
I don’t know what those are in the ostensible vacuum of the laboratory. Aside from their environmental costs, which i suppose can be extrapolated on a per-prompt basis (you just havw to deabstract the financial cost, add in the vc subsidy and figure out the electric and water bjlls and then transform those into kilowatts and gallons and few more hops and skips of economic stoichiometry until you get to the right set of units).
But are you generating value for a technocapitalist utopia where the technos and the capitalists are the only ones that matter. Are you laundering their dirty data into something respectable and even useful in its dismantling?
In my early days using Gmail, a friend and I would mention random news stories and historical events at the ends of our regular emails. The goal was to see what topics would cause the ads in the sidebar to disappear.
Death always worked. War, too. They were easy targets, and sensible: no one wanted their ads running alongside a discussion about a sinking cruise ship. Wild to think there were limits, then.