Bunker mode for normal humans using chatbots
Many of the people I know use AI to entertain, and get advice on, the most intimate corners of personal life and imagination. Overtime, I have grown bored and frustrated with the scent of epistemic authority that frontier models carry around like a signature strength.
Instead, I carry a vision for a future where every individual is an arbiter of their own creation, where their unique trains of thought, questions, ideas, are valued (to the extent where they remain private and are taken seriously - rather than being dismissed by a certain constitution or safety training). I believe this craving, and need for a fertile and private garden of thought will only increase overtime as we see advanced persistent threats targeting critical infrastructure, including the AI labs, and mass data leaks. It goes without saying that one should be able to speak and create without fear of imprisonment, a former ex-now-lab-employee, government agency, or member of general public having (or getting) access to our AI chat history; which in many ways can reflect the most raw, unformed, and vulnerable contours of our pyschosomatic landscape.
Outside of the "personal" domain, I believe researchers, students, and mathematicians should have access to frontier intelligence without fear, no matter how light the suspicion, that they are quietly serving as a creative source of production for a hungry learning machine. A machine, that is not solely of service to you. Subscription costs, and the value of human ingenuity and time spent to get there, are way too high to not have crystal clear clarity on this. It seems that there may also be issues of thought and research convergence, if a community finds themselves beholden to a duopoly of models. This warrants more careful thought from me, but it seems to support the development of more personal models that might reflect, support, and expand our own divergent ways of thinking, seeing, and experiencing the word.
For my own personal use, and for fun, I've started to develop my own app that provides a space where I feel safe exploring my ideas privately. This sits alongside a growing community of efforts, including Confer and OA-anonymity. I think it is critical that more research is directed towards these efforts, as well as research on modes of training and inference across decentralized and distributed systems: even if the outcome is the final nail in the coffin of infeasibility.
I share the belief that privacy is not about ownership of data. Instead, it is about being able to create, love, and communicate what we want, without fear of repercussions.
Please reach out if you have feedback, pointers, or ideas related this. Or if you simply want to join forces to breathe life into an inspiring future where humanity co-evolves alongside AI.