The Human Stream: the future of data feeding LLMs

Far from running out of data - AI is about to be overwhelmed by a flood of new, more valuable and more sensitive data.

I’ve heard recently views expressed that external AI data is becoming commoditized and LLMs will run out of new data soon.

“There’s only one internet” ... “AI is running out data”

I’ve probably said this at some point myself. But now I disagree with it.

And it might be true if you assume that the data that feeds AI is just the stuff that trained the early frontier models - like books, Wikipedia, the ‘common crawl’ of the internet etc.

But we are beginning to see the emergence of a truly multi-modal AI that will see entirely new sources of data come online. Just imagine all the video, sound, health and behavioral signals captured from the billions of smartphones on earth. This is part of the new stream of data that will shape AI.

We can call this The Human Stream.

• This raises big, thorny questions for leaders:

• What would you do if you could access the human stream?

• What shouldn’t be done with it?

• How will you secure access to it?

• Who will control it?