AI & Traditional SAAS Businesses

I was talking to a freind (@dillonrgardner) about this paper a16z about how software margins and how AI plays will be part of the eco system. '

A few thoughts: I buy the basic premise: AI/ML business don't scale as well as SAAS as a general rule. Things that reduce margins and scale:

NAS search
NAS search

One thing that I think is less important than what the article describes is the issue of inference and Moore's law effects: even in vision applications there are not that many DL models that are bottlenecked by processing power alone. The only exception to this is in a safety critical space like driving where you need every. ounce. of. accuracy.

AI and software are not that different in one key aspect:

FTA:

Handling this huge state space tends to be an ongoing chore. Since the range of possible input values is so large, each new customer deployment is likely to generate data that has never been seen before.

This is a key consideration when thinking about scaling an AI business: what applies for regular software applies ML: complexity kills and you are well-advised to reformulate your problem domain to severely limit possibilities of corner cases as much as possible.

PS: don't you just hate it when someone says AI when the mostly mean DL ? 😊