YouTube is being overrun with AI slop. And it probably doesn't help that YouTube itself, and owner Google, is where a lot of it is coming from. Using AI-powered tools to "enhance" videos, without telling anyone--including the creators who made said videos. YouTube viewers and video producers like Rhett Shull have noticed a certain sheen and smoothness to some videos that wasn't intentionally done by the original uploaders. This sort of filtering isn't new--in fact, you've probably seen it over-applied to old movie clips uploaded to TikTok and YouTube shorts, giving them an unnaturally smooth motion and overly glossy look for things like human skin.

This week, in a neatly self-referential story, an algorithm has been developed to help workers ascertain if their jobs are at risk of automation by… an algorithm. Researchers compared robotic and human abilities to identify the most at-risk jobs (meat packers and slaughterers are high risk; astronomers and neurologists are low risk) and also used the comparison to work out which lower-risk jobs people could safely and easily jump to. In other AI-defying news, people are using'algospeak' – language that will not be picked up by AI content moderation systems – to ensure their (sometimes dubious, sometimes legit) content remains live and uncensored. Examples include using'unalive' instead of'dead' and the corn emoji instead of'porn'. As well as screening out properly dodgy content, moderation algorithms have been found to censor already marginalised groups, demonetising and down-ranking posts that include words such as'gay', and'racist'.

Facebook's artificial intelligence researchers have a plan to make algorithms smarter by exposing them to human cunning. They want your help to supply the trickery. Thursday, Facebook's AI lab launched a project called Dynabench that creates a kind of gladiatorial arena in which humans try to trip up AI systems. Challenges include crafting sentences that cause a sentiment-scoring system to misfire, reading a comment as negative when it is actually positive, for example. Another involves tricking a hate speech filter--a potential draw for teens and trolls.