Profile cover photo
Profile photo
Phil Sung
163 followers
163 followers
About
Phil's posts

Post has attachment
WSJ article featuring this hedcut of... a road sign.

Post has attachment
Shut up and take my money!

Post has attachment

Post has shared content
In this dystopian future, everyone's internet-connected washing machine tries to run at the exact same hour in the middle of the night...
Adaptive marketing has a downside. Pricing electrical power based on demand could lead to chaotic markets. This seemingly obvious way to make the electricity market better may actually make it worse. http://phys.org/news/2015-07-seemingly-obvious-electricity-worse.html

Post has attachment
"[The problem is] that students don't learn the distributive law as a fact about numbers. They learn it as a fact about parentheses."

Post has attachment
A story of "lost in translation" (despite, or maybe because of, things like "Der Suppen-Nazi")

Post has attachment
The state of the art in computerized hallucination. (Also, these are going to give me nightmares.)

Post has attachment

Post has shared content
"In short, we are spending thousands of dollars worth of water to grow hundreds of dollars worth of almonds and that is truly nuts."

Post has shared content
As has been widely reported by now, the Computer Science Department recently sent incorrect admission emails to about 800 applicants to the MS in CS program at CMU.  We had to retract a few hours later, which caused a great deal of understandable distress among the applicants. Here, I would like to explain how this came about as an unfortunate confluence of a single program error, followed by a chain of several human mistakes.

In the 2013-14 cycle, a bug was introduced into the graduate application and review system.  This bug escaped testing, and in retrospect we do not think our testing framework was adequate.  The bug was not exercised at all during the 2013-14 cycle.  It arose when a database field was changed from a number value to a string, and one comparison was not appropriately updated.  Ironically, for a programming language researcher like myself who has dedicated most of his career to developing expressive disciplines of static typing, this should have been flagged as a static type error but the code was written in PHP.

In the 2014-15 cycle, a change in procedure led to this bug corrupting the database by setting the "admit" bit for many applicants when their applications were marked as "complete".

Alert administrators discovered this before PhD admission letters went out in early February.  The bug was fixed, and all affected degree programs except one were manually corrected. The MS in CS program was overlooked.

All program administrators were notified of the problem and were asked to double-check their admit lists.

Some views of the admit list show only the applicants where the "admit" bit was set manually, so the manual check of the admit list missed the remaining spurious "admits" in the MS in CS program.
Just before auto-generating emails, the system displays the number of emails about to go out and asks for confirmation.  This number showed 800+ (instead of 100+) but was overlooked.  About 800 erroneous admit letter went out mid-day Mon, Feb 16.

The issue was discovered relatively quickly when students in the "decline" pile replied to their admission letter.  It took several hours to determine the nature and scope of the problem and identify the students who had received the erroneous letters.  Brief correction emails went out as soon as we had that information.

Dean Andrew Moore and I followed up with further explanation and apologies to all affected students the following day (Tuesday) and continue to reply personally to inquiries since then.

We are currently running an additional layer of formal review before automated admission emails are sent out. We are in the process of re-evaluating our admissions software and processes to determine how to move forward and avoid such disasters in the future.
Wait while more posts are being loaded