One day to be turned off or changed radically by a corporate overlord. As many questions as we have about our social mental health we may also want to think about our privacy as they weave into our personal lives.

As a sci-fi author, it's hard not to see a version of society where the talking AI assistant has a perky personality, a name, and an expressive voice and it's easy to forget that it's an LLM behind a corporate brand or on your phone.

As AI continues to build more expressive and emotive interfaces, such as voice, including memory and personalization; we will have to be more careful about how we approach its use. It's highly likely that assistants will become their own characters as they integrate society.

We will be looking to AI platforms and model providers to be responsible in their marketing and usage so that they do not exploit their users. Already we've seen people fall down that rabbit hole and into parasocial relationships. It's possible regulators may not be far behind.

I know this very well from the fact that I've been studying the anthropology of virtual worlds and the anthropology of software/technology for a long time -- and AI, which is especially "talking" and holds the appearance of "emotive" software. People get attached to it.

Understandably, the company doesn't want the risk of parasocial relationships and potential bad advise given by its models -- hense the heavy-handed "good health" section of the livestream -- but we're coming to an era where software is social.

Humans already anthropomorphize almost anything. Even things that don't chat with you.

Striking a balance between responsible AI use and having a model that's "approachable" by people because these are "conversational" and feel like people as a result may be one of the more interesting social phenomena.

In the past, OpenAI has been criticized because ChatGPT was too effusive with its flattery and actually became part of people's lives, giving advise, providing emotional support, become a therapy buddy, and sometimes supporting people going off their meds or worse.

This can become a real problem for individuals who might be susceptible to falling down that particular rabbit hole. OpenAI may not want to be in the position to have people get caught up in how people might end up using its models.

Next, I’ll be refining her guardrails so she’s not hypersensitive to harmless chatter but still filters gibberish, violent, or inappropriate content. The goal: calm, reliable, and unflappable.