Raw LLM Responses

LLMs democratically reflect human nature(it is all linear algebra and stats). If AI turns out to be evil then it is only because we are evil. We train it on our thoughts and deeds, and then we expect it to behave "saintly". I run LLMs locally. I have the same models in censored and uncensored versions. The censored versions, when presented with tough prompts, will just return with "Sorry, I cannot help you with that.". However, the uncensored models have no problem helping me make illegal substances, coming up with plans to steal money or how to get rid of a body. Here is what you need to understand. Both versions are trained on the same data, but the censored version has layers of safeguards blocking certain responses. So, why do I use uncensored versions? Speed. I use it for coding, especially auto-completion. Without the safeguards some of these models can run on a Raspberry Pi without any GPU support, thus less power and resources. So, what is the one risk? There is no way AI developers can manually filter out all harmful input data. Too much data. They feed it on everything...the whole internet. If that means documents on how to make illegal substances or plan the perfect crime then it is already in there. They then spend a lot of time coding filters and testing. It ends up like "I know how to make a nuclear bomb...but I won't tell you how to build one.". When we reach AGI/ASI then we just have to trust the safeguards that the system won't develop some evil system. Why do developers cry "danger" but still keep on building these model? Money, power, sense of achievement. When I was in school (1980s) I like to build explosive devices and rockets. My friends wanted to share the same experiences. I was safe, but they were reckless. I only stopped sharing information after a 3rd friend almost ended up dead. The question you must ask from me is why I did it in the first place or kept on after the 1st friend ended up in hospital. Conclusion...AI development won't stop. It has already escaped the lab. I can write my own code and fine tune existing models. I have a financial backer that will fund serious millions if needed...and this entity is not connected to the AI industry. I might not have evil intentions, but many others do.