What approach links Jeremy Howard and Elon Musk? I recently watched Elon's interview on TEDx, and I've also been watching Part 2 of the Fast.AI course by Jeremy Howard, in particular, lesson 9.
Elon was talking about building a 10x solution to boring machines. He broke it down into sub components, such as:
- Boring at the front and reinforcing at the back can give a 2x speed up.
- The cost of boring is proportional to the diameter of the hole, by halving the diameter we can get a 4x cost reduction.
- By using the machine closer to it's thermal and energy limit we can get another 1.5x.
Elon, then multiplied these (2x4x1.5) and claimed that it was approximately a 10x speed improvement.
Jeremy also performed a similar calculation, but instead of 10x'ing a business, he was 10x'ing computation speed. Jeremy's logic was as follows:
- We can use SIMD to get a 6x speed up for computation that has been optimized to use the SIMD instructions.
- We can use Parallel processing to usally get 2-4x speed up on a standard computer.
Jeremy then claims that you can add these (6 + 4) to get a roughly 10x speed up.
It's interesting that they disagree on how to combine the improvement. However, they both follow a similar logical path. I think that this is a good way to build up a 10x advantage where no single technology can get you there.
Also, I think that you should add you multiplicatives, not multiply them, see standard algebra.