GradIEEEnt half decent: The hidden power of imprecise lines [video]

tom7 is always the right mix of genius, madness and fun

Previous post (but different): <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=35780921">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=35780921</a>

:( did not have a patience to watch the entire video while focused. I know you could use rounding errors to replace non-linearity. Can somebody summarize what exactly he does for gradient descent here and how well does it work?