How many US VAERS deaths were due to COVID19?

We can now use GPT4 to help estimate this number

Summary:

• Daniel (Truth_in_number) on Twitter posted a long thread on VAERS in mid-2023, seemingly based on this comical premise - unlike him, no one else had looked at VAERS extensively¹

  • While Daniel did read 120 reports, he hasn’t yet posted the VAERS IDs for them (to the best of my knowledge), so the people who already agree with him are just taking his word in terms of final analysis

• I refuted most of his central points anyway using large scale text analysis of VAERS reports (for example there are hundreds of reports of sudden death in VAERS, while Daniel claims he did not find any “true” sudden death reports)

• The one major point he made² - that 61% of the deaths he read were caused by COVID19 - was much harder to refute till now because it required large scale text analysis of 16000+ reports which did not follow any well defined text pattern

• I used a recent feature provided by GPT4 to automate this process³, and found that only 39% of 16K deaths even tested positive for COVID19. In other words, less than 39% of US VAERS deaths could actually be due to COVID19, leaving the other 60%+ as potential vaccine induced deaths.

In a previous article I mentioned that LLMs can be used to extract cause of death from a VAERS report.

When I wrote that article, LLMs weren’t quite good enough for the task.

But things are moving quite rapidly in this field, and recently Large Language Models have added a feature to extract Structured Output from a given prompt.

So I wrote some code to send the writeup and the lab data information to GPT4, and ask⁴ it to extract some information and return the result in a structured format.

And not just cause of death, I used the same approach to also extract other information:

• did the patient test positive

• was the patient hospitalized?

• how many days from symptom onset to death?

• dose number

As you can see, I appended a bunch of columns to the original VAERS DATA CSV file to make it easy to filter for all the reports where the patient tested positive for COVID19. I also added a link to the MedAlerts History page to make it easy for people to verify for themselves whether GPT4 extracted this information properly.

Click here to see the results

I wanted to provide some sample code to explain how I did this, but I remembered this tweet from Daniel when I challenged him about the 61% number on Twitter.

Link to tweet

Since Daniel is quite sure there is no skillset deficit, I will leave it as an exercise for Daniel and the other vaccine pushers to figure out how I got GPT4 to give me this information.

What is the percentage?

As you can see, GPT4 was able to read the VAERS writeups, and make a reasonable⁵ inference about whether the patient ever tested positive for COVID19.

From the original list of 16520 death reports, only 6350 patients tested positive for COVID19 at some point.

6350/16526 = 38.4%

In other words, less than 40% of the 16500+ US VAERS deaths state that the patient even tested positive, let alone died from COVID19.

While Daniel claims that 61% of the deaths in US VAERS is due to COVID19, the number is far less.

What about the other 60%?

While the vaccine pushers don’t seem to have much curiosity⁶ about this question, you can see that even death reports in US VAERS are often too sparse to answer this question properly.

This is of course in direct contrast to foreign VAERS reports where you can expect serious reports to contain copious documentation. In my next article I will do a similar analysis for foreign VAERS reports and report my results.

1. Which is false, obviously. Not only did some researchers read a large sample of the VAERS death reports by April 2021, they already read double the number that Daniel did, and these researchers were all clinically trained!

The one consistent theme among the vaccine pushers is that they often don’t know what they don’t know, but get annoyed if people point this out :-)

2. And was mindlessly parroted by other vaccine pushers on Twitter, making it seem like a large majority of VAERS death reports were actually COVID19 deaths

3. You can now use GPT4 as well as Gemini to extract specific structured information from not just VAERS reports, but from any kind of dataset which has a lot of text. I don’t like to hype up AI, but this is actually a game changer.

4. The total cost was under $50, which isn’t too bad considering we were able to get a fairly definitive answer to this question. I predict that the ability to use LLMs to extract such information at an ever-declining cost is soon going to become Big Pharma’s WORST nightmare! :-)

5. In fact I added a field for GPT4 to “explain” why it decided True or False, and you can read the explanation and also read the VAERS report itself to verify if it is accurate.

6. For example, EVEN IF DANIEL was right about the 61% number, that still means over 1/3rd of the people died due to other reasons. What were those reasons? Out of 16K, that is nearly 6000 reports for which we need to find the cause of death. Why didn’t Daniel do the logical next step and try to classify them into (say) the top 5 causes. That’s what an unbiased scientist would do.