I’ve become curious about some of the discussion about the need to be cautious about training extremely large AI models that demand a lot of compute power. Joe Biden’s executive order introduces a threshold above which there are disclosure requirements.
What was the basis for that threshold?
With an appeal to scientific history, I dive in.
As always, you’re better off watching the video that reading the summary 😄
Johannes Kepler is justly celebrated as one of the greatest scientists. In the 17th century, he formulated laws of planetary motion based on empirical data instead of mere speculation. His work extended beyond the laws of ellipses, equal areas, and harmonies to significant advancements in optics that enhanced telescope functionality, leading to his role as a practicing astronomer.
Kepler’s beginnings, however, lay in astrology. He provided horoscopes and various services to the elite, attempting to reconcile astrology with astronomy. Yet, as he delved deeper into the scientific aspects of astronomy, he gradually abandoned the unsubstantiated practices of astrology.
There is an analogous process, but in reverse, unfolding with AI.
