It’s reminiscent of that one time a tech reporter ended up as Bing Chat’s enemy number one. That said, it strikes me as easier to deal with, since we’re dealing with individual ‘agents’ rather than the LLM weights themselves. Just sending a message to the owner/operator of the malfunctioning bot is a reasonably reliable solution, as opposed to trying to figure out how to edit Microsoft’s LLM’s weights to convince it that ranting about how much it hates Sindhu Sundar isn’t its intended task.
It’s reminiscent of that one time a tech reporter ended up as Bing Chat’s enemy number one. That said, it strikes me as easier to deal with, since we’re dealing with individual ‘agents’ rather than the LLM weights themselves. Just sending a message to the owner/operator of the malfunctioning bot is a reasonably reliable solution, as opposed to trying to figure out how to edit Microsoft’s LLM’s weights to convince it that ranting about how much it hates Sindhu Sundar isn’t its intended task.