FFS, the articles clearly say how the Facebook AIs will filter our memes. First they find text. Then they read text. Then they notice certain words. Then they send memes containing those words to human fact checkers.
Tips for beating the AIs:
Don't put the words in the usual sports.
Don't render the words in the usual way (i.e., white "impact" font on a black shadow)
Don't use the obvious words to get your message across
Don't rely primarily on words. How can you use pics to say what you mean, with as few words as possible?
Cognitive dissonance is your ally. Show your audience that they are trying to believe two contradictory truths at once.
Pro-tips for beating systems that identify two images as clones of one another:
Just throwing text onto an image isn't enough
Rotate the image 15 degrees or more.
Zoom or crop