Came across a situation where DevOps-style ML (deep-l ..?) was used in flash financial business. Without too much QA; merely empowering the developer/trader-trained staff to do their thing. Asked about architecture, complexity, legacy and (Taleb-style) fragility.
Got no real answer. Seems like a risky thing. Not as in risky == having some Normal distri, but as in risky == bet the business including the well-being of all employees and their extended families for a generation and a half.
As is (over?)summarised in this tweet:
We’ve gotten to the point where many companies are building faster than they can learn. Velocity is *not* the problem.
— John Cutler (@johncutlefish) April 29, 2019
Anyone out there that has a clue about how to Control these sort of things? I have, I can. Just call.
[’cause I need a refill]