In Which I Embrace our Mechanical Overlords
For a long time, I didn't really use LLMs. I think GPT 4 came out right before I graduated high school, and I played around with it for a little. I had it write a silly poem for an English class — I think it involved Jordan Peterson making moonshine. But, so far as I could tell, everything it produced was very shitty. Sometimes amusingly shitty, but shitty nonetheless. It was an interesting toy, but it had no practical value.
This remained my impression of LLMs for the next few years. And I grew to hate the ChatGPT Customer - Service Voice, which I encountered more and more on the internet.
But, recently, a friend let me play around with her Opus, and I realized it was really useful. I was curious enough to fork 20 bucks over to Anthropic, so I've been using it for the past month. And I mostly like it!
Opus knows chemistry much better than I do, and it can make me a lot of practice problems. And, it can usually answer my questions pretty well. It's not as good as a human tutor, but human tutors have lives, so they aren't on call 24 / 7. Claude is. It's also quite good at summarizing long documents, which is something I occasionally need.
And it doesn't have the Customer Service Voice either. At least when I use it, Claude talks like a friendly and knowledgable person who happens to have no experiences. I don't necessarily dislike this, it's kind of what I want from a digital assistant.
More importantly, it hasn't attempted to be my friend. I mean, it is friendly, but it hasn't tried to establish a personal relationship with me. When I played around with other LLMs, they always did, and that's what most disturbed me. If they manage to make the robots concious, I'll happily befriend them. But I won't befriend an unfeeling simulacrum of a person, I find that prospect really creepy. I know many people do use Claude that way. But, so long as I only talk to it about dry subjects, it doesn't treat me like its pal, and that makes me much more willing to use it.