22 comments

[ 2.6 ms ] story [ 68.6 ms ] thread
Being too successful is a funny way of describing premature optimization and the risks behind great yet narrow specialization.

Specialists are by definition exceptionally successful in exceptional circumstances. I don't think that makes them too successful. It just makes them high risk-high reward takers.

The example cited is not death by too much success but rather death by not diversifying the product line.
This is a good example of hacker writing in 10 paragraphs the sort of thing that has often been diluted into a 250-page business book.
The amazing thing is that there is a lot of repetition even at 10 paragraphs. Some of the other comments here sum it up in twitter-sized chunks.

The missing piece is how to identify over-specialization, and it's missing because it can't be usefully generalized.

"The missing piece is how to identify over-specialization, and it's missing because it can't be usefully generalized."

If you read the literature on selective breeding they have extensive methodologies for avoiding genetic problems. If you just want a quick overview, Temple Grandin's book "Animals in Translation" has a couple of really entertaining chapters on the subject.

That book looks hugely interesting--thanks for the recommendation!

We've been using biological metaphors in this discussion, but I'm not sure we can extend them this far.

Didn't Andy Grove say something like this? However he couched it in terms of paranoia.
I can condense it down into one sentence:

If you're not careful you will specialize yourself into a corner.

It's also a good example of someone taking 10 paragraphs to say "don't overfit"...which might be a bit terse for non-AI people, but the article still feels long-winded.
Yeah. "Overfit" is more of a ML concept than AI concept though. Many ML people don't believe in AI at all.
I didn't mean Spielberg AI...rather, AI as the umbrella term for things like ML, vision, heuristic search, some parts of robotics, etc. But yes, more specifically, it's a ML term.
The post made some interesting points about technical success not equating to real world success, but the whole article suffers from a poor definition of the word.
Funny, at Toastmasters last night our table topic was "what if you had a different mother?" and we got to pick moms from a random list. My random mom was Angelina Jolie, and I think I made a well-reasoned case that I would have fucked myself to death.

My point is that parent's success probably outdoes your own success.

Thanks, I'm going to use that Table Topics theme next time it's my turn.
Lisp Machines weren't that successful. I was their target user, and I ditched Symbolics to use Lucid on Suns at the first opportunity.

The Lisp Machine software was appallingly baroque. Everything had every possible feature. There was no design, just implementation, and lots of it. The manuals took up a whole shelf, and it was generally faster to write something yourself than to find the predefined function that did it.

A lot of (arguably most of) the badness of Common Lisp is Lisp Machine culture showing through.

(comment deleted)
high degrees of optimization lead to fragility. look at the current financial situation for an example.
(comment deleted)
Anyone know more about the author Erik Naggum? I have run into seen some old posts from him on the Python list and now this. A simple Google search reveals he is an academic/writer from Norway but not much more.
He is a huge presence on comp.lang.lisp . Not sure about his prowess as a programmer however.
"Worse Is Better" again? =)