News for hackers. Stuff that matters.
Re:a16z Programming Test by graycat on Friday April 11, 2014 @09:57AM (#7572834)
Looks to me like A16Z is blowing it. If Ben Horowitz prides himself on good management of technical work as suggested by his blog posts, then he didn't work very hard on the project of the OP.

Issues:

(1) Engineering School. Uh, guys, here in the US we have universities with schools of engineering -- MIT, Johns Hopkins, CMU, Georgia Tech, etc. Running a good engineering school is a lot of hard work. But this A16Z project has A16Z being their own engineering school, that is, specifying 'course material', testing for it, and deciding on pass/fail. Somehow it seems a bit arrogant for A16Z to believe that they can do the work of an engineering school.

A16Z gets really off the track with their,

> A BS/MS/PhD in Computer Science or the equivalent is nice to have, but the most important requirement is significant independent programming experience as demonstrated by your GitHub account, personal projects, academic publications, or startup success.

Wow! "PhD in Computer Science" ... "is nice to have, but"...! Gee, guys, just shut down MIT, Johns Hopkins, CMU, Georgia Tech, etc. because now A16Z is here with "the right stuff"!

Next, for their "as demonstrated by ... academic publications", sorry, A16Z, "academic publications" are not at all about "programming experience". Instead, the usual criteria are "new, correct, and significant" where "programming" really does not count more than, say, just the word processing for the paper.

(2) Machine Learning. Good grief! That "machine learning" is about "scatterplots & histograms, logistic/linear regression, R/Matlab" fills a "much needed gap in the literature" and "would be illuminating if ignited". Their definition is for a 'statistical data assistant' mostly just for 'descriptive statistics'. Ever hear of J. Tukey? Ah, I'll be nice to you guys; not fun to be too harsh on small children!

Then there is their unforgettable,

> If you also know enough data science/machine learning to design useful metrics for dashboards and workflows, that’s a major plus.

Uh, ever hear of reliability and validity? You have three hours to respond here with a relevant description! You are not permitted to use Google or Wikipedia!

(3) Ego. It appears that the A16Z specification was written by someone who has enough ego to believe that their own 'skill set' is just "the right stuff".

(4) Linux. People programming on Microsoft's platform can have done well but encountered only a small fraction of the topics in the A16Z document. The document is mostly about Linux. Sorry, guys, I'm writing in Visual Basic .NET; no, C# has too much of the deliberately 'idiosyncratic' syntax of C which I find too error prone and hate.

(5) Three Letter Acronyms (TLAs). It appears that A16Z is using obscure TLAs as code word filters. Hmm.

So, they have ORMs. Gee, what the heck is that? So, Google a little and get a Wikipedia entry and discover ORM abbreviates 'object-relational mapping'. Okay. Sure, if write the software with objects and keep the data in a relational database, then need move data between objects and tables. Given the huge variety of objects and tables, tough to believe that there is a significant 'subject' here.

Then there is their ETL. Google again and discover that that TLA abbreviates "extract, transform, and load". So that is essentially 'extracting' data from whatever original sources, 'transforming' the data as needed by the work to be done, and 'loading' the data into a data base. Okay. Sounds entirely routine. I'm surprised they didn't also list (A) LRT, look at a screen, read it, and type in what need, (B) PSS for put on shoes and socks, (C) SDP for start car, drive to the office, and park the car, and especially the crucially important (D) GSL for Google a question, select the promising results, and learn what needed. One more: WTH, waste time on Hacker News!

Okay, A16Z, two can play this game: Think of a large server farm or internet protocol network. Since this 'facility' is subject to errors of wide variety, it is important to monitor the facility for unexpected problems of wide variety, say, from software errors, hardware failures, system administration errors, security breaches, etc. To do this collect some data, say, from Microsoft's instrumentation means, HP's Mercury Interactive, what a relational database management system or Cisco box can report, etc. So, maybe 100 times a second, collect numerical data on each of 25 variables. Now we want a detector of 'anomalies', including for problems never seen before, that is, 'zero-day' problems. Such monitoring is essentially hypothesis testing. So, we want a suitable hypothesis test. Due to the fact that the multi-dimensional probability distribution of that 25 dimensional data is essentially impossible to describe at all accurately, we need an hypothesis test that is 'distribution-free'. Since the data is 25 dimensional, our test must also be 'multi-dimensional'. We want false alarm rate known in advance and adjustable in small steps over a wide range. And we want a good shot at doing nearly as well as the Neyman-Pearson result but with much less data than that result requires. Okay, A16Z, your mission, and I'd like to see you accept it, is to say how to do that. Of course, since you guys are so good and brilliant, you already know my published paper for the answer. But there is a typo in the paper. Where is it? You have three hours to post your response here.

Also needed, find a fast way to do the specified computations. I found a way, but that is not in the paper. You have 24 hours to post the answer to this part! Don't report your answer in some silly way such a running code. Instead, report your answer in clear English with any appropriate mathematics written as high quality mathematical writing, e.g., W. Rudin. Go for it, guys!

Uh, you won't find the answer on Google or GitHub! And asking around Silicon Valley won't help, either. You might want to rush off to the RAD lab project of Patterson and Fox funded by Google, Microsoft, and Sun for when they worked on such problems, but that won't help, either.

My view: Here A16Z showed that their understanding of computing is still down at the level of a junior high self taught programmer.

There are no comments on this comment yet.