Note that this example only works against one specific version of one specific algorithm. Who knows what tiny tweaks might be needed to get that recognition working again, or what tiny tweaks might be needed to Re-break it again afterwards.<p>Adversarial algorithms are an interesting space to explore, I grant you that. But this is not a general adversarial algorithm that works against all AI vision recognition systems.
This article has me wondering, would a tied and dyed T-shirt or hoodie be effective at fooling a similar AI system? maybe you could add some mirrors on it as well ;D