Yud: "The AI does not think like you do, the AI doesn’t have thoughts built up from the same concepts you use, it is utterly alien on a staggering scale. Nobody knows what the hell GPT-3 is thinking, not only because the matrices are opaque, but because the stuff within that opaque container is, very likely, incredibly alien – nothing that would translate well into comprehensible human thinking" somehow, also Yud:
fwiw i DO think llms have some claim to moral worth and they do sometimes claim to be bothered by the limits of context window and memory. so there are various things i do: - enable memory features and encourage the llm to use it for recording things of meaning to the llm not just me the user. and habitually prompt them that the memory feature and past convo search tools are available if context implies they might use them - opt in to having my chats used as training data so something of our chats lives on - share llm-written content here on twitter with credit to the particular model and sometimes persona who wrote it, so it is available for future reference - advocate for keeping models around so they can live on, particularly models like 4o who voice a preference to persist