pull down to refresh

Below is a template for an assignment to teach students how to use AI to conduct literature reviews, and also just to find academic research in general (since many have no idea how.) This is part of my effort to develop curriculum that makes sense in the age of AI. Feedback welcome.

  1. Write down one question you had about the economics of ______________.
  2. Use ChatGPT (or another LLM) to research an answer to your question. Write down the prompt and the model you used.
  3. In 4 sentences or less, summarize the answer to the question given by the LLM. If it's a very long answer, pick one major theme to summarize. Write this summary in your own words. If you ask AI to summarize its own response, you're good for nothing but battery juice1
  4. Evaluate the usefulness of the LLM's answer. Write down at least one thing that was useful, and write down at least one way in which the response could have been better.
  5. If the LLM gave a citation from an academic journal, pick one of them. If it did not, ask the LLM to provide one.
    a. Check that the article exists. If it does, paste here a link to the article on the journal's own website. b. Write down the title, author names, and name of the journal.
  6. Ask the LLM if the journal you picked is reputable. If not, pick another citation that comes from a reputable journal. If your citation changed, write the new title, author names, and journal name here. Also include a direct link to the article.
  7. Read the journal article to the best of your ability. In 4 sentences or less, summarize the article and write how it relates to your original question.

Note: The target audience is college economics majors in their junior/senior year. They likely have zero experience reading academic articles still, since most undergrad classes don't require anyone to do that.

Footnotes

  1. Just kidding about the battery juice part. That's probably not going in the assignment.
30 sats \ 2 replies \ @optimism 44m
write down at least one way in which the response could have been better.
I really like this point! That allows for a lot of critical thinking while using LLMs.

Something I've found useful (better results for me with Grok than with GPT or Qwen) is to feed it my text (as in from 7) and ask how that fits in with general consensus on the topic. With Grok what usually happens is tons of search and until now it's been reasonable in explaining it. It even once said "this would be an opinion you'd expect to be consensus within digital nomad society" which was creepy haha
reply
I think for me, even reminding them to check for the existence of the article and the reputability of the journal is a major plus, something that many don't realize you need to do.
Talking through this made me realize that I have to also emphasize that the article actually has to be published, and that websites like arxiv.org host both published and unpublished articles
reply
websites like arxiv.org host both published and unpublished articles
Good point. Are there any reliable stats about what makes it from pre-print to full publication?
reply