Welcome again to everybody. Thank you for joining us for the time-honoured tradition of our group, here is the weekly discussion thread.
• Matrix homeserver and space
• Theory discussion group on /c/theory@lemmygrad.ml
• Find theory on ProleWiki, marxists.org, Anna’s Archive
Do you think using AI is a good way to study theory? I’ve been using it as a way to ask questions. For example: answer this from x perspective or what would x author think about this?
In my area of professional expertise I am afraid that the AI hallucinates too much and I am afraid that the same thing will happen when I ask questions to the AI and that I am not realizing it… at the moment it seems to be going well because I only use it for very specific cases, but perhaps it is because I do not have real expertise in Marxist texts.
A better approach would be “in x book author discusses this thing, in the current situation I am examining this other thing and applying the theory to get this understanding? is that a correct application of the theory?” would go way better.
LLMs aren’t good at being people but they are good at being systems of words.
Hallucinations aren’t as big of a concern to me as shallow or skewed readings, imo. Think of all the ways there are to misinterpret Hegel or Marx, and all the reddit comments and YouTube videos and Trotskyist and liberal articles that comprise the training data. I’d be careful asking information-seeking questions.
Some models can probably be useful to help you parse difficult passages, and talk through concepts if you learn better through dialog. Also can find secondary resources if you’re using one that can search the internet.
But as long as they’re only supplemental to actual theory and you don’t get overconfident in your early understanding (which is a problem regardless of LLM usage) I don’t think there’s much harm. No matter how you study you’re gonna have misunderstandings that become refined into understandings through the process of learning and discussion.
No. The best way to study theory is to read the source material directly. Perhaps find a version with annotated commentary, in case you are confused about historical context and need it explained to you. But nothing beats reading the works yourself and forming your own opinion. You don’t need them to be digested and regurgitated for you through someone else’s ideological filter, and certainly not by a glorified auto-complete algorithm that was trained on ideologically biased data to begin with.
i think the most you could use it for is to create study questions for yourself since they can be pretty open-ended, but even that is suspect since again the LLM doesn’t know what to test you for (or what you want to test yourself for), it’s just a system of stringing together words. that’s the most i would use it for, though.