My rebuttal

  • I first posted a rebuttal article to the original thread
    • the analysis I did for the first rebuttal article could be called “regular” data science
  • It was still missing an important analysis – there was no easy way to read the 16K reports and refute the statement that over 60% of the death reports were due to COVID19 itself
  • So I used GPT4 to “read” all the 16K reports and refute a sub-claim
    • I asked GPT4 to verify if the patient ever tested positive for COVID19 (according to the report)
      • if they did not, then obviously they could not have died from the disease
    • I asked it to provide the answer as a structured output field
  • This did cost me more money than I wished to spend for volunteer work, but I realized that this was an approach worth pursuing if
    • LLMs got a lot cheaper (which was already happening then)
    • and they got better at extracting structured data (which has also happened over time)
  • this updated rebuttal article, if done in a more structured way, would be a good example of “text centric” data science