A different perspective on the limits of human ‘knowledge’

Disagreeing with Elon Musk online is a risky thing to do!

But I saw part of a CES interview where he says: “We’ve now exhausted basically the cumulative sum of human knowledge… in AI training”

I don’t agree.

For that to be true, you would have to believe that the internet = “basically the cumulative sum of human knowledge”. However:

Many data sources are not yet publicly available. The University of Cambridge and others made new datasets available last year and I’m sure many more organisations will follow.

Not all countries and languages have digitised their knowledge to the same extent (e.g. South Africa)

Humans have not stopped and will not stop researching and creating new knowledge anytime soon. Over 5m new journal articles are published each year; potentially a lot of new knowledge. (Source)

I believe all of these will be more effective sources for model training sooner than synthetic data.

Getting agreeing access with all the organisations and individuals that hold the data may be complicated. Which perhaps was the part of what he meant with this comment.

Access to quality training data is a challenge for all AI companies and an area of expertise for Nascent Studio. Contact us to find out some of the ways in which we have helped our companies access high quality training data.