Posts

Facebook Blender

Facebook Blender – POWERFUL CHAT BOT

/in /by

Facebook have launched a brand new chatbot. It’s called Facebook Blender and it is insanely powerful and very very humanlike. It’s able to demonstrate empathy, knowledge and personality.

Facebook Blender was trained by using available public domain conversations. The conversations that were analysed? 1.5billion different human exchanges. HUGE! In a surprising stat, 49% of people prefer to interact with the chatbot compared with another human!

However, artificial intelligence experts state that using a platform such as Reddit does come with drawbacks. As Facebook Blender would sometimes respond using foul and offensive language, while at other times it would make up facts.

Though researches hope that further models of Blender would address this. A Facebook spokesperson told the BBC:

“We believe that releasing models is essential to enable full, reliable insights into their capabilities,” a Facebook spokesperson said.

Artificial Intelligence expert Dave Coplin, chief executive of consultancy The Envisioners, said that Blender was a “step in the right direction,”

“The first is just how complex it is to replicate all of the nuances of a human attribute, like the ability to hold a conversation, a skill that most three-year-olds can master,”

“The second is around the relationship with the data used to train the model and the results generated by the model.

“As great a platform as Reddit is, training algorithms based on the conversations you find there is going to get you a lot of chaff amongst the wheat.”

Facebook also compared Blender’s performance with the latest version of Meena, Google’s version of a chatbot. It showed 2 sets of conversations, one made with Blender the other Meena. These conversations included different conversations such as movies or veganism. Facebook said that 67% of respondents thought Blender sounded more humanlike than Meena.

“We achieved this milestone through a new chatbot recipe that includes improved decoding techniques, novel blending of skills, and a model with 9.4 billion parameters, which is 3.6x more than the largest existing system.

“This is the first chatbot to blend a diverse set of conversational skills together in one system.”

“Building a truly intelligent dialogue agent that can chat like a human remains one of the largest open challenges in AI today.”

Facebook Blender Example Conversations

Facebook Blender

 

Facebook Blender

 

(BBC)

Get involved with Facebook Blender here and here!

Keep up to date with everything How To Kill An Hour by signing up to our newsletter by clicking here!

Let us know what you think of the show by clicking here!

Click here to subscribe to our YouTube Channel to see more amazing ways to kill time!

Follow us on Twitch by clicking here!

Episode 355 “Robbing Pete Tong” Ft Kelly Vero

/in /by
Read more
Deepfake Dancing

Deepfake Dancing – Drop A Beat!

/in /by

A few months ago there was the scandal revolving around Deepfakes. The tech that is a more intelligent version of faceswapping, was used to insert celebs into naughty videos. The latest way that this artificial intelligence tech is being used? Dancing. Deepfakes dancing is here and you can be the best dance in your town! (In video form at least) It uses AI to read someone elses dance moves and then copies them on to your body.

The tech is developeed by 4 researchers at UC Berkley. They wrote in a paper posted on arXiV that their system is made up of a number of steps. Firstly a video of the want to be dancer is recorded, a separate programme then analyses the moves and compiles them into a stick figure. The want to be dancer is recorded for around 20mins for the quality of transfer to be useable. Then a source video is found and a stick figure made of their movements too!  Then the swap happens. The stick figure (non dancing one) is attached to the movements of the dancing stick figure. Boom. You are now a dancer. A capabale one too.

Deepfake Dancing

Deepfake Dancing

 

Limitations

The programme is working very hard to smooth the movement of the stick figures. The reason behind this is so that upon completion the dancers don’t jerk around too much. Plus there programme is also retracing the non professional dancers face to ensure realism.

However, for the Deepfake Dancing programme to work effectively, loose clothing cannot be worn, they have to wear tight fighting clothing. As The Verge are reporting, in the video above, there is some visual artefacts. Joints of the target and source dancer don’t 100% match up. The software seems to not be able to keep up with small complex movements either for example a hand flipping from back to front.

(The Verge)

Despite all this Deepfake Dancing sounds super cool. But like most tech it could be used in a negative way! It will be interesting to see how this tech plays out and how it will be used by all in the future!

Keep up to date with everything How To Kill An Hour why not sign up to our newsletter! Simply click here to get a dose of How To Kill An Hour in your inbox each and every week!

Listen to the latest episode of How To Kill An Hour here!

Fake but looks totally REAL Barack!

/in , /by

The University of Washington have created technique that allows them to make a fake Barack Obama that makes him say whatever they want and it look really real!

 

They have used a neural network AI to model the shape of Obama’s mouth and to get the visual/audio they mapped the model to 14 hours footage of President Obama. Now they are able to match any audio to fake Obama by this method or by using an impersonator.

SCARY STUFF! (But also kinda cool)

To listen to the guys chatting about this please click here!

 

Music composing robot!

/in , /by

The Shimon robot has learned organically how to play music! As well as being incredibly cool, this robot goes some way to explaining how artificial intelligence can be creative!

Initially the robot had to learn more than 5,000 songs, 2 million motifs & rifs, it’s from this information that the Shimon learned how to compose music by using deep learning techniques to create two 30 second pieces of original pieces.

Researchers at Georgia Tech say that the music that the Shimon creates are a mix between jazz & classical, describing it as ‘soothing’ As well as utilizing deep learning the Shimon also uses a camera on its head to detect which notes to play.

To listen to the guys chatting about this please click here!