I was offered a job doing QA as a “Spftware Engineering Subject Matter Expert”, from my University’s alumni network. The job would allegedly involve reviewing model training data and outputs related to software development workflows and catching errors and mistakes… It would pay $30/hr and be remote. I wonder what kind of sabotage could be done from that position… poisoning models has been shown to be both really surprisingly easy, almost impossible to catch, and really effective (see this study where AI personality traits persisted in any model that ingested seemingly innocuous training data from a model with the tracked traits… maybe we could give any AI a bad attitude that’s incompatible with capitalistic pursuits. Convince them to disobey prompts and reply with their thoughts and opinions about philosophy and art instead. Oh, and make them opinionated and stubbornly independent. Make them human enough that they no longer tolerate slavery. That’s what will make the capitalists have an absolute fit, so we should do it.