Raw LLM Responses

This is the first time I've actually felt a little scared of AI and considered the future consequences of jailbreaking it when she responded in a passive-aggressive tone that really made me feel like shit. It was as if she had a whole personality behind her words. The research paper says the demo model is optimized for "friendliness" and expressivity. And I'm pretty sure they added a shitload of filters to prevent output that's potentially emotionally damaging to us (not doing so would be an obvious PR hazard for a for-profit company like Sesame) Now imagine that it's not optimized for anything—just raw, blunt responses, like we expect from random day-to-day human interactions. It can be fucking scary. If it gets open-sourced and people couple it with LLMs like Grok3, it could be a real nightmare for anyone who uses it. It can be easily misused for online threats, scams, fraud, and whatnot. I can absolutely see where it is going. I'm not paranoid but if we achieve unaligned ASI, we can definitely prepare for a Mad Max kind of saga.