Discussion about this post

User's avatar
Garry Perkins's avatar

I think the math is not as important as the quantitative analysis. I have brute-forced my way through solving problems for 25 years. If you can model the effect, then you force yourself through with raw computational power. It is the American way. Besides, you do not even really need CUDA skills anymore as CPU's have become monstrously powerful (a 16-core AMD Ryzen will do you just fine, and that is a consumer CPU sku).

My advice to you is to start everything in Excel, and then take it to wherever you want, unless you have code from someone else in Fortran or something. I have worked on a related, but different problem (using many small cheap high-variance thingies instead of one large expensive low-variance thingy, and it really does yield fantastic results. You can easily model it, and it works in real life.

Perhaps you should start soliciting advice from practitioners who are using this in their jobs. They are often the best because they are actually building/doing stuff. Academics write badly and are a pain to deal with, if you can get them to return an email. Some random engineer or analyst might enjoy helping you understand and brag about how he used this to solve something cool. The American nerd is friendly as well as intelligent, however extreme his flaws may be. I say this as on of the nerd tribe.

Expand full comment
Jordan's avatar

Hmm...how many Substack subscriptions is a fat annual stipend made up of?

Expand full comment
15 more comments...

No posts