Cross‐Referencing Studies - uchicago-bfi-gnlab/lab_manual GitHub Wiki

How to Do:

  1. Break the work into small steps
    • ChatGPT can do a very good job at matching things you might do well.
  2. Save things in files often and never work with pdfs
    • Txt and csv seem to do fine. Here, for example, I take the works cited and ask to get a txt back with one line per entry:
	Aaronson D, Mazumder B, Schechter S. 2010. What is behind the rise in long-term unemployment? Econ. Perspect. 34(2):28–51
	Altmann S, Falk A, Ja ̈ger S, Zimmermann F. 2015. Learning about job search: a field experiment with job seekers in Germany. Discuss. Pap.
	9040, IZA, Bonn
	Amaral P, Ice J. 2014. Reassessing the effects of extending unemployment insurance benefits. Econ. Comment. 2014-23, Fed. Reserve Bank
	Cleveland
	Arranz JM, Muoz-Bulln F, Muro J. 2008. Do unemployment benefit legislative changes affect job finding? Unpub- lished manuscript, Univ.
	Alcala ́
	Baily MN. 1978. Some aspects of optimal unemployment insurance. J. Public Econ. 10:379–402
	This is a list of citations. Can you create a file "svwreferences.txt" file where each line has one entry from this works cited
  1. Use Chunks
    • By trial and error, 19 is usually the best chunk size. If you do more, GPT will often screw something up as below:
	Error analyzing
	It seems that the text for the eighth chunk of citations was not completed, leading to a syntax error. Could you please provide the
	complete list for chunk 8, including all 23 citations? This will enable me to compile them correctly into the text file.
  1. Always provide example output when you have complex output. As below:
	Ok, just a problem with the last names...you have done the first initial of the last author. Names go last then first. For example, with

	Aaronson D, Mazumder B, Schechter S. 2010. What is behind the rise in long-term unemployment? Econ. Perspect. 34(2):28–51

	you should get Aaronson as first author last name. Please do the same thing again with this change
  1. Use those chunks all at once later: image below