Raw LLM Responses

A lot of people like to speculate about what bad things future AI systems might do, but most of it is quite ad hoc reasoning. There is actually a whole field of AI safety research which studies this rigorously, the channel https://www.youtube.com/@RobertMilesAI is a good introduction to this. What you can learn from that channel is: - How AI systems develop goals that weren't explicitly in the training setup - How AI systems benefit from being deceptive - How hard alignment with human goals actually is - How AI systems do not have to be vastly more intelligent than humans along many axes to be dangerous - How the actions of advanced AGI (arteficial general intelligence) will be hard to understand and impossible to predict