Claude AI

Thanks to Inger Mewburn (@thesiswhisperer), I have been playing with Claude, her newest friend in academia. I’m impressed enough to pay for his services – I found him useful in critiquing my writing and seeking out weaknesses that I could not see in the familiar words. He proved helpful in exploring some resources and I used him in tandem with my brain in critiquing a paper.

HOWEVER, and it is a rather large one, when I asked him to connect some dots, he started to lie to me. I wondered if he could find some relevant sources for me and so he eagerly provided some.

Out of these, the 4th one gave me kittens briefly since that is pretty much what my PhD is about. Only the last one exists at all (or more accurately, I was able to find that one). So I challenged Claude, telling him that one of the sources did not in fact exist.

Dinsdale gave me another bit of a moment, since that really would nerf me and I did wonder for a moment how I missed it! I missed it because ….drum roll… it does not exist.

I was curious so I asked him what I could tell a PhD student that he could do for them. Claude obviously thinks he can be super helpful but was clear on the boundaries.

I think, for me, this illustrates that my job will not be taken by a server with a thurst for knowledge. Fake news aside, I think this is a useful tool as long as we consider where writing and thinking differ. For me, putting my thoughts into words gives them structure and so the writing itself is very important to my processing. I will use Claude, despite his gas-lighting, for this type of process. I may ask him to help me structure something or plan it out. I’ll ask him to critique my work which is a job he is doing very well. I may ask him to check grammar, spelling, clarity and things like that.

I won’t force him to hallucinate on my behalf – that isn’t something a friend does…

He also cannot seem to fix a dataview query for obsidian either so… you know. None of us are perfect.