Some observations and reflections on questioning and thinking after getting used to talking with large language models.
I have noticed that I talk with LLMs more and more these days. They keep getting better, and their knowledge base keeps getting broader. Now I am more used to asking a model directly than opening a search engine and digging through a pile of results first. After it answers, I can keep asking follow-up questions. Compared with clicking through webpages and filtering information myself, this saves a lot of time and energy.
Talking with a model about a topic is another kind of experience. It can point me toward related theories and books and help expand how I understand something. One conversation can feel like opening a new window and seeing the same issue from another angle.
After doing this often enough, I started to have a somewhat extreme thought: in many situations, talking with an LLM feels more valuable than talking with people. Human beings are still the animals in this world that think, of course. But when a model can also organize language and explain a line of thought clearly, human thinking no longer has such an obvious advantage in everyday conversation.
I know that human creativity is still stronger than that of an LLM. But in reality, many people's ability to communicate and understand does not exactly inspire confidence in their capacity for independent thought. On the mobile internet, low-quality opinions are everywhere. Against that backdrop, an LLM that can stay logical and respond in an orderly way starts to feel like a rare conversation partner.
Lately, whenever I have a question, I almost always ask an LLM first. For me, one direct benefit of the AI era is that it saves me the time I used to waste searching all over the place for materials, or arguing with other people without end.
Still, I think talking with people is more valuable in the end. There are many things an LLM cannot give me. A model's knowledge base comes from information that has already been recorded into computers, while a person's "knowledge base" is the whole of that person's lived experience. A machine can neither truly express human feeling nor fully understand the mood I am in when I type something. Only when I am talking with another person do I feel that I am speaking as myself. When I talk with a machine, what remains in the end is often just input and output.