Greg Eiden has an interesting idea we thought might be interesting/terrifying/amusing…
What might ChatGPT or other such tools derive from an analysis of the Samizdata blog post archives? To focus this on something timely: ask it to do the analysis and propose policy ideas for the new Trump Administration, ideas consistent with the perspectives and ideas of the Samizdatista community. Or at least the broad consensus of that community about liberty and limited, less regulatory government.
There are other ways this might be posed to an AI, e.g., “which proposed policies of the Trump administration are consistent with the broad governance perspectives at Samizdata.net and why? Which are not and why?” How would you set up such an analysis to have the model have the best chance at producing something useful?
I am not interested in an experiment that assesses if ChatGPT is a good model for this or a good model in general–I want a good analysis! If the product of such an inquiry was not compelling to enough samizdatistas, if it did not pass a laugh test, we stop there.
Maybe there is a better choice of large language model than ChatGPT; maybe ChatGPT is not optimized for this sort of analysis but other models are? If you have some knowledge here, please chime in!
