AI is rapidly changing how everything it touches works. Many of the popular uses have been around content creation and making a better SIRI or Alexa.
There is a potentially bigger use case of adapting, framing and shaping existing content to fit the needs of the end user, or to fit anyone else in between you and some content.
For example, why can’t all the news be happy? Why can’t content reflect your world view? Why can’t you remove content you don’t like, transparently?
Could we rebalance content toward reality? The gap between reality and media is very large:
This is today’s BBC News front page:
Here it is as viewed through a browser plugin I made which sends all the content to GPT for modification. In this case, I want the content to be happy and uplifting:
Here I ask for the content to be negative and bad:
Here, the content blames Trump:
Lastly, I ask the content to allude to aliens and world government:
Here’s the NYT today, biased to allude again to world government and aliens:
Here, the same page but alluding to hidden agendas:
This may all feel very blunt and obvious, but it’s to make a point. This content shaping could be as subtle as you want it to be. Only modify every 10th article, shift this person this way, another person another way. Instead of blocking content, shape it as you see fit.
President Snow:
It won't work. Fear does not work as long as they have hope, and Katniss Everdeen is giving them hope.Plutarch Heavensbee:
She's engaged. Make everything about that. What kind of dress is she gonna wear? - floggings. What's the cake gonna look like? - executions. Whose gonna be there? - fear. Blanket coverage. Shove it in their faces. Show them that she's one of us now. [laughs] They're gonna hate her so much they just might kill her for ya.
There is bottom-up power in shaping content too. Individuals are able to reshape content any which way they like and remove or counter any bias they perceive.
While this is for text, we can do the same thing for audio.
Lex Fridman’s 8 hour podcast interview with Balaji Srinivasan is a great example for content shaping. I took the audio, automatically transcribed it and separated the speakers. Then I shaped the content and re-recorded the podcast, again automatically. For brevity only the fist 10 minutes or so were worked on.
For example, what if you wanted to hear two women instead of two men:
How about you want the podcast simplified?
Or, what if you wanted Balaji to interview Lex?
This isn’t perfect, but soon it will be, and you will be able to listen to or read any content you want in the timeframe, tone, voice and bias of your choosing. Or someone else can shape it for you.
Oh, and someone will do this for video soon I imagine.