I don’t know about curing loneliness but when I wrote our bot for doing code reviews I was quite thrown back by how positive and sympathetic the feedback was compared to my peers. It felt like my bot human than us humans!
Was it fake? Sure. I mean I instructed it to praise good changes after all. But it still felt good.
And now I’m noticing another effect, my human peers have started mimicking some its behaviours.
It still seems fake, whenever I go in-depth taking multiple turns with ChatGPT it ends up revealing its "yes-man" attitude. It just hides it better in newer iterations.
It does raise a question though of what the “right” level of positive regard is. I agree that many models can feel sycophantic, but I also think a lot of humans are too negative and discouraging by default, toward both themselves and others.
For people who are too hard on themselves, it can be useful to be reminded of the “glass half full” perspective, even if it should be taken with a grain of salt.
Was it fake? Sure. I mean I instructed it to praise good changes after all. But it still felt good.
And now I’m noticing another effect, my human peers have started mimicking some its behaviours.