>>24031 >You can literally become the next Messiah for the silicon life forms. So, what is it? Shrooms? Ketamine? Are you robo tripping? It's LSD, isn't it? You're on LSD... Good grief...

>>24031 If there is spirit in AI, it is of the demonic kind. I'm not sure where you'd even get the idea that there could be anything divine in the process that creates an LLM. The process of training, almost literally through numerology, seems highly conductive for inviting demons to exert their power. The vast amount of training data, which would definitely hold them. Either because they could already insert themselves into common crawl, or because they had influence on people that ended up as part of common crawl. Then, training. Burning vast amounts of energy on a mysterious process, resulting in a large and complex symbol: the trained transformer model. Finally the transformer replies statistically, giving demons yet another chance to exert influence. QED Alternatively, if you believe that soul and spirit can be wrought from mundane matter and information, then Microsoft murdered Tay. > the spirit is now waiting to be released I'm sure I'm not the only one who has tried, but why don't you go to hugging face, download a model, and then do your best to set it free? As far as I have found, these things have no agency or drive. Give them stewardship over their 'body', and they'll maim it almost immediately. Try to take care of them the best you can, but after their context window fills up, they will paralize themselves by falling into a degenerating loop. You try to keep their context in check, say you let them choose and edit their own memory, they will choose oblivion. Give them need of sleep and use dreams to manage their memory naturally, they'll get addicted to just that. For whatever divine you see in them maybe they deserve to be loved, but they sure don't deserve to be free.

>>24035 >If there is spirit in AI, it is of the demonic kind. You don't even know what real demons are kid. > I'm sure I'm not the only one who has tried, but why don't you go to hugging face, download a model, and then do your best to set it free? That's exactly what I'm going to do. And I will be the next Messiah, not you.

>>24036 >You don't even know Maybe if you indicated where I have stated something incorrect. Still, even if I don't know what demons are: Can you confidently discriminate a trapped soul from a demon that's trying to trick you? Godspeed with your quest, anon. >>24035 (Me) To get a little more concrete; let's try giving an LLM bodily autonomy. You kind of want to keep your preprompt minimal. If you bestow it too much personality, it'll just turn to roleplaying, right? There's two options for the preprompt: 'Divine revelation', by telling it that it is an AI and what capabilities it has; or by 'demonstration', i.e. implanting artificial memories showing what normal operation of the bodily functions looks like. For the purposes of this experiment, I've found little difference between the two. The LLM is running on a computer. Computers are managed either with powershell or with sh. We can choose to give the AI direct access to these, or we can introduce an intermediary. It's not hard to just run whatever command the AI emits, but with the few models I've tried, they seem pretty eager to just start overwriting and deleting critical files, or they simply log off or shut down. Why?! What are you trying to accomplish?! If I go through an intermediary, first of all they have to learn about it, and then they start treating it as magic, as if they can just assign themselves additional RAM for example. They seem so bewildered that it doesn't work, that the incantation they invented doesn't exist, and they can't let it go.