Lisp developer walks away with Google AI contest • Hacker News Board

Re: Lisp developer walks away with Google AI contest

#51

Post by krmboya » Thu, Dec 02, 2010, 3:09 PM UTC

Zak wrote: post #43

pietrofmaggi wrote: post #32

Lisp was clearly a fine tool for this kind of challenge, but the programmer was, as usual, the important piece of the solution. Otherwise I'll expect to have some other of the 33 lisp based entries in the top spot (the second lisp entry is ranked at the #280 spot: http://ai-contest.com/language_profile.php?lang=Lisp ). What about the top ten solutions? http://ai-contest.com/rankings.php 4 Java 3 C++ 2 C# 1 Lisp So is…

I find the mix of languages found in the top 100 interesting. Smart is obviously a big factor in determining the winners. Speed was a factor in the contest as well, with a 1 second per move time limit. Based on that, I'd expect to see the faster of what Paul Graham calls "languages for smart people" near the top. Lisp, Haskell, ML and Lua come to mind. That didn't happen. What follows are observations about things I…

'Speed wasn't that important: Python made up 20% of the top 100.'

In the article, speedy execution can be taken to mean the speed/ease with which you translate you idea into code and not the speed at which the resulting code runs

Re: Lisp developer walks away with Google AI contest

#52

Post by jimbokun » Thu, Dec 02, 2010, 3:13 PM UTC

thomas11 wrote: post #45

Zak wrote: post #43

Earlier quoted context omitted.

I find the mix of languages found in the top 100 interesting. Smart is obviously a big factor in determining the winners. Speed was a factor in the contest as well, with a 1 second per move time limit. Based on that, I'd expect to see the faster of what Paul Graham calls "languages for smart people" near the top. Lisp, Haskell, ML and Lua come to mind. That didn't happen. What follows are observations about things I…

I think a lot of participants might be in the same situation I'm in: I like new and exciting languages and play with them now and then in my free time, but I work in one of the "boring" languages. Having way more experience with the boring language, it's the one I turn to when I have a deadline or the problem is hard enough that I can't afford to spend brain cycles on the language itself.

I feel like Clojure has crossed that threshold for me.

I program in Java at work because it's what my team understands and it's what the rest of the application is written in. But if I need to do some exploratory programming or build tools to increase my personal productivity, I turn to Clojure first. It's much more productive because I can test each little piece at the REPL and it takes so much less code than it does to do the same things in Java (even when it is just invoking methods on Java objects).

If I need to, it is then a straight forward step to expand the code out into its verbose Java form.

Re: Lisp developer walks away with Google AI contest

#53

Post by Xurinos » Thu, Dec 02, 2010, 3:45 PM UTC

There are many comments here that he won because of his programming/thinking skills versus using the Lisp language, and I think this true in comparison to the other contestants. However, one thing he mentions that is an interesting piece of the puzzle -- and a common theme in language competitions that include Lisp -- is:

Lisp doesn’t get in my way as much as other languages. After getting hooked I learned to value the interactive development environment, the object system, etc.. What’s good about Lisp is rapid development and speedy execution.

This might be a small effect, and you can have good programmers in every language, but if he was just as familiar with Java as he was with Lisp, could he have accomplished the same results in the same time with Java? How would the code look?

Re: Lisp developer walks away with Google AI contest

#54

Post by CWuestefeld » Thu, Dec 02, 2010, 4:33 PM UTC

Zak wrote: post #43

pietrofmaggi wrote: post #32

Lisp was clearly a fine tool for this kind of challenge, but the programmer was, as usual, the important piece of the solution. Otherwise I'll expect to have some other of the 33 lisp based entries in the top spot (the second lisp entry is ranked at the #280 spot: http://ai-contest.com/language_profile.php?lang=Lisp ). What about the top ten solutions? http://ai-contest.com/rankings.php 4 Java 3 C++ 2 C# 1 Lisp So is…

I find the mix of languages found in the top 100 interesting. Smart is obviously a big factor in determining the winners. Speed was a factor in the contest as well, with a 1 second per move time limit. Based on that, I'd expect to see the faster of what Paul Graham calls "languages for smart people" near the top. Lisp, Haskell, ML and Lua come to mind. That didn't happen. What follows are observations about things I…

Unless I'm mistaken, the contest was limited to supported languages. F#, ... don't appear because they weren't supported

I'm not familiar with how the contest was implemented, but I would assume that if C# was supported, then any .Net language would work. It should be trivial to build a C# wrapper for your F# code, if the interface you must support can be implemented in all those other languages.

Re: Lisp developer walks away with Google AI contest

#55

Post by richcollins » Thu, Dec 02, 2010, 4:37 PM UTC

swannodette wrote: post #17

richcollins wrote: post #16

Bocsimako is the proof that Lisp is best for AI algorithms Cargo cultism at its finest

Ugh. Cargo cultism? The full quote is: After seeing Gábor’s performance, many participants commented that they would be taking a new look at the 50 year old language. “Bocsimako is the proof that Lisp is best for AI algorithms,” quipped one player. one player.

Cargo cultism is applicable to one person

Re: Lisp developer walks away with Google AI contest

#56

Post by richcollins » Thu, Dec 02, 2010, 4:40 PM UTC

ohyes wrote: post #19

richcollins wrote: post #16

Bocsimako is the proof that Lisp is best for AI algorithms Cargo cultism at its finest

It seems that any positive inclination towards lisp is cargo cultism. Gabor Melis is a seriously smart guy, and probably would have kicked ass in any language. That said, I'm going to pull out that phrase next time someone says 'Java/C# is good for Enterprise', or 'C++ is good for video games', or 'C is good for systems programming'. Cargo cultism indeed.

I don't think you understand the term "cargo cultism"

That the language is Lisp isn't important. The important consideration is that he ascribed the success of the entry to the language without any evidence linking the two.

Re: Lisp developer walks away with Google AI contest

#57

Post by ohyes » Thu, Dec 02, 2010, 4:48 PM UTC

fhars wrote: post #24

dminor wrote: post #23

Earlier quoted context omitted.

Well, the 33 lisp entrants don't appear to hold the top 33 spots: http://ai-contest.com/language_profile.php?lang=Lisp

So more than a fifth of the lisp entries are in the bottom twentyfifth of the competition. So you could also use the contest results to argue that lisp is five times worse for AI than the average language. (Or, rather, that language alone doesn't guarantee success, but that would be too rational and boring for an online forum).

Your math doesn't seem to work.

For that to be true, more than a quarter would have to be in the bottom 25th percentile. (In fact, it still points to lisp being better than average).

Anyway, The bottom 25th percentile of the competition is relegated to submissions that didn't run.

There was a bug (still is, I believe) in the lisp 'starter pack,' which caused the client to crash (It sent garbage to the server or something). So even the default Java starter submission would beat it.

Given the number of people who submitted once and walked away, I wouldn't read too much into it anyway.

Re: Lisp developer walks away with Google AI contest

#58

Post by ctkrohn » Thu, Dec 02, 2010, 4:50 PM UTC

Zak wrote: post #43

pietrofmaggi wrote: post #32

Lisp was clearly a fine tool for this kind of challenge, but the programmer was, as usual, the important piece of the solution. Otherwise I'll expect to have some other of the 33 lisp based entries in the top spot (the second lisp entry is ranked at the #280 spot: http://ai-contest.com/language_profile.php?lang=Lisp ). What about the top ten solutions? http://ai-contest.com/rankings.php 4 Java 3 C++ 2 C# 1 Lisp So is…

I find the mix of languages found in the top 100 interesting. Smart is obviously a big factor in determining the winners. Speed was a factor in the contest as well, with a 1 second per move time limit. Based on that, I'd expect to see the faster of what Paul Graham calls "languages for smart people" near the top. Lisp, Haskell, ML and Lua come to mind. That didn't happen. What follows are observations about things I…

You could have written a program in any language. It just needed to accept certain command line options and send results to stdout in the proper format. They only provided "starter kits" for a few languages, but if you were willing to write a little extra code, you could have used anything.

Re: Lisp developer walks away with Google AI contest

#59

Post by ctkrohn » Thu, Dec 02, 2010, 4:51 PM UTC

eru wrote: post #49

lmaonade wrote: post #29

Earlier quoted context omitted.

After seeing Gábor’s performance, many participants commented that they would be taking a new look at the 1000 year old country. "Bocsimako is the proof that Hungary is best for AI algorithms", quipped one player.

Being born and educated in Hungary is even better. Compare John von Neumann, Edward Teller, Leo Szilard, and Eugene Wigner. They changed the education system, and no longer used the one they had around 1900. Unfortunately.

Don't forget Paul Erdos!

Re: Lisp developer walks away with Google AI contest

#60

Post by winxordie » Thu, Dec 02, 2010, 5:04 PM UTC

eru wrote: post #49

lmaonade wrote: post #29

Earlier quoted context omitted.

After seeing Gábor’s performance, many participants commented that they would be taking a new look at the 1000 year old country. "Bocsimako is the proof that Hungary is best for AI algorithms", quipped one player.

Being born and educated in Hungary is even better. Compare John von Neumann, Edward Teller, Leo Szilard, and Eugene Wigner. They changed the education system, and no longer used the one they had around 1900. Unfortunately.

I don't know about that. The first wave of 20th century combinatorialists were pretty much all Hungarians (Bollobas, Erdos, Polya, Szekeres). Hungary hasn't stopped churning them out either.

I'd argue that their education system's just fine.