BREAKING: US Political Misinfo Twitter Detection AI List
Thread Below 
-
I have obtained exclusive access to documents that have never been publicly revealed. The information I am about to share with you has never been released. This information is of public concern and the below thread includes my reporting on it.
-
I have thought long and hard about whether I should release this information. Ultimately, the public has a right to this information. Therefore, I am releasing it to you, despite my deep concerns about what this means and the possible implications.
-
Unlike the other writers involved in reporting on The Twitter Files, I do not have access to any direct information from Twitter. My information is from a direct source. I do not have the same level of legal protection these other writers have. I am taking the risk anyway.
-
I have never been focused on people or personalities. That type of reporting bores me and does nothing to help people understand the truth. I am focused on data.Therefore, decided to find the data on my own. What I found was shocking.
-
I have received two separate files with thousands of data cells. I will only be releasing a small portion of it and still have a lot to go through. It is important to understand the context of the information.
-
I will not be sharing the tweets that accompany the detection / flagging list for the parameters I am about to share to protect the privacy of those mentioned in them and the original accounts who posted the information.
-
Not all of the tweets flagged were removed. Some are still up. But what is important to understand is that they were flagged at all. Why did certain words trigger the algorithm? Why were these terms being actively monitored?
-
Do not share my reporting or screenshots without properly citing me as the author. All of this will be on Blockchain to show proof of authorship and ownership as well- so don’t even try.
-
What was Twitter searching for?
“These are tweets we flagged for removal algorithmically. Some were held up, some reversed. We also searched for context.”
-
“These are phrases we judged when in the right context were indicators of misinfo. Context was important because a journalist could tweet these & it would be okay. One or more of them was fine. In combination & in context, they were used to judge the tweet as misinformation.”
-
To the left of the detection box you can see the flagged tweeted along with an algorithm and a score. To the right, you can see the rating, annotation, and decision. Not every auto detection of these words resulted in removal. Some stayed up. Some were removed.
-
The real question is: why were the algorithms flagging these phrases and words? These terms were flagged under US political misinformation. Below please find examples from the dataset I received. The world has never seen this data. I am not sure if Musk has seen this either.
-
I will not be sharing the example of the tweets that flagged these detections. I am only sharing the words parsed in the detection list.
Final reminder: not every tweet was removed.
Why were these terms monitored by Twitter as US political misinformation?
-
These are the real Twitter Files. Not curated internal Slack messages. And not screenshots that Twitter executives have shared with me. #RubyFiles
