DeepFake Video of Eminem Insulting Mark Zuckerberg

This is a review of some artificial intelligence news that happened throughout the week.

With the use of the GPT-3 technology that produces texts simulating human writing, they emulated the rapper’s voice. The union of two YouTube channels with the use of neural networks created a Eminem-style song in which he seems to scold the CEO of Facebook. They also had the help of Google’s Tacotron 2 system which emulates Eminem’s voice. The video looks like a real song by the rapper although some words and phrases are incoherent.

South Korea debuts artificial news anchor

The virtual presenter with artificial intelligence is taken from a real journalist, imitating not only physical details but also her voice, speech, facial expressions and movements. The virtual presenter began to appear since October having conflicting responses from viewers, some applauding the similarity with the original person and others with the prospect of being afraid of him. It should be noted that this has the advantages of reducing labor costs, production in disaster situations and other emergencies.

They create detailed map of air pollution in the United Kingdom created with AI

Thanks to the combination of artificial intelligence and satellite technologies, a detailed picture of the air pollution in Britain today has been produced. This enables precise air pollution concentrations, offering highly advanced detail with daily updates on a 1 x 1 km scale across Britain. Indicating that the southeast is the most polluted region and identifying hot spots in urban and industrial areas.

Deep learning helping to fight fires

Fast and accurate fire detection is of great importance to humanity and the environment. Objects with characteristics similar to fire can give false fire detections with the use of artificial vision. Making it better to identify false positives by increasing deeper visual characteristics is a challenge. Scientists have proposed a model based on the extraction of multiscale features, implicit deep supervision and channel attention mechanisms. The results of this new model showed that the precision reached 95.3% with a model size of only 4.80 MB, facilitating its implementation in devices with limited resources.

Thanks for reading, I hope you have informed yourself of something new. See you in the next edition.