Something I find very interesting in the age of AI is my own change with respect to these models. As LLMs have gotten more powerful, more autonomous, I have started to get more and more frustrated and even sometimes angry when they do the wrong thing. Kinda like an uncanny valley for intelligence - the models appear and sound very competent, but when you look closer, they are all smoke and no fire.
One of the reasons I find this interesting is that I think it represents a change in my approach to these models. I would never have gotten frustrated and angry at a computer program in the old world. At the person who wrote it (often myself) certainly, and definitely some level of self-hatred and negative self-talk would also seep in (“why can’t I figure this out?”). But not at the thing itself. However, that’s quite different with LLMs - I find myself genuinely pissed off if a frontier LLM says it has done something that on closer inspection it clearly has not done; when it is clearly being sycophantic and simply parrotting things back to me; sometimes, when it doesn’t read my mind. Just as I might with a co-worker, say.
It’s interesting, the ways the anthropomorphising gets you.