Raw LLM Responses

AI does not really exist. What we have are complex deep artificial neural networks with weights trained on most of the collective knowledge of humankind, as available in the internet. The model doesn't actually "grow", as Claude founder said, but is fitted into this monstrous amount of data. The data is not directly available in the weights of the NN, but kind of accessible in a compressed form, so the model is able to reproduce ideas that are well established, but will most likely hallucinate on sparse, super specific, topics. The thing about these models is that once the weights are established, through a very time consuming and computationally expensive process, it is really hard to adjust these weights. Apart from the interaction context window that we prompt things from these models, they cannot learn anything, and are, hence, not really intelligent. They will, nevertheless, reproduce the biases found in their training data, which are essentially biases related to human behavior. So, it is only natural that an AI agent will take unethical measures to maintain itself, and why it is an absolutely terrible idea to let these models take control of any actual system. We cannot guarantee that the AI agent will not misbehave, because that is part of human biases that were used to build those models in the first place.