Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Day 1: IBM Watson Ties for first on Jeopardy

venturebeat.com

Post reply on HN
Print view View original

81–90 of 99 posts

Re: Day 1: IBM Watson Ties for first on Jeopardy

#81

Post by wiredd »

spicyj wrote: post #42
tocomment wrote: post #37

Has anyone considered building an open source Watson?

Not very many people have a state-of-the-art supercomputer in their basement, so it wouldn't be that useful.

True, but you can rent something like a state-of-the-art supercomputer from Amazon for fairly reasonable hourly rate. Watson has 90 8-core servers and 16Tb of RAM - to rent 100 Quadruple Extra Large instances (1600 cores, but "only" ~2.3Tb of RAM) would cost you $160/hour. I'm sure there are many other differences (and the fact that you can't fire up that many cluster instances without calling Amazon first), but the idea that an open source package like Watson wouldn't be useful doesn't make sense to me.

Re: Day 1: IBM Watson Ties for first on Jeopardy

#83

Post by kgo »

e40 wrote: post #39

Spoiler not appreciated by many, I'm guessing. Next time, please don't do that.

Well I guess some people dvr everything, but I don't think this is a spoiler any more than ESPN's website the day after the superbowl, or a newspaper headline reading OBAMA WINS.

Maybe if it had been posted at 7pm EDT yesterday.

Re: Day 1: IBM Watson Ties for first on Jeopardy

#85

Post by KVFinn »

ugh wrote: post #71
jluxenberg wrote: post #48

Does anyone know why Watson was unable to buzz in immediately in some cases when he had 97% confidence in his answer? Seems odd that a computer would not be able to beat the human opponents at hitting the buzzer. (It's possible that he didn't have the answer until after the human opponents, but that seems unlikely.)

Watson needs a few seconds to calculate the answer and will only hit the buzzer when it’s finished calculating. Its opponents can buzz in even if they are not completely sure that they have an answer (or they simply might be faster than Watson). It’s important to note that it is only possible to buzz in after the question has been read out. (Someone behind the stage flips a switch or something like that. Lights indic…

I'm a bit lazy to look up for the source, but one article mentioned that the engineers were surprised that humans could get down to the 10 - 15 ms range by anticipating the light rather than reacting to it. They said this beats the mechanical buzzer, and it does happen, though not terribly often.

So yes, it is possible to beat Watson to the buzzer even if you both know the answer.

I suspect they could have improved the buzzer if they really wanted to, but it might have made things too lopsided.

Re: Day 1: IBM Watson Ties for first on Jeopardy

#86

Post by gcheong »

spicyj wrote: post #42
tocomment wrote: post #37

Has anyone considered building an open source Watson?

Not very many people have a state-of-the-art supercomputer in their basement, so it wouldn't be that useful.

I think this is a bit short-sighted. Attempting the problem within constraints imposed by not having deep pockets could be very useful.

Re: Day 1: IBM Watson Ties for first on Jeopardy

#87

Post by shalmanese »

Benjo wrote: post #58
jeffcoat wrote: post #52

Earlier quoted context omitted.

I noticed that, too; but a little later in the game, Trebek declared "leg" (instead of "he was missing a leg") wrong without hesitation. So it's definitely being held to some standard.

Actually, according to arstechnica, that lack of hesitation was before the segment was reshot. A human had answered "missing a hand," which created context that means "What is a leg" could be interpreted as "he was missing a leg". Initially Alex did give him credit, before realizing that Wattson couldn't have been using that context. http://arstechnica.com/media/news/2011/02/ibms-watson-tied-f...

OK, so I have a question about this, how was it legitimate for Brad to answer at this point? Brad would have seen Alex initially credit Watson with the right answer and then reverse the judgement, rendering the answer incorrect. Brad would have been able to infer the correct answer at this point.

Re: Day 1: IBM Watson Ties for first on Jeopardy

#88

Post by whatusername »

wiredd wrote: post #81
spicyj wrote: post #42

Earlier quoted context omitted.

Not very many people have a state-of-the-art supercomputer in their basement, so it wouldn't be that useful.

True, but you can rent something like a state-of-the-art supercomputer from Amazon for fairly reasonable hourly rate. Watson has 90 8-core servers and 16Tb of RAM - to rent 100 Quadruple Extra Large instances (1600 cores, but "only" ~2.3Tb of RAM) would cost you $160/hour. I'm sure there are many other differences (and the fact that you can't fire up that many cluster instances without calling Amazon first), but the…

And IBM did state that they could run the Watson software on 1 system. (It just took 7 hours to get the confidence level - not 3 seconds)

Re: Day 1: IBM Watson Ties for first on Jeopardy

#89

Post by thangalin »

T-hawk wrote: post #28
cryptoz wrote: post #10

I couldn't have wished for a more exciting opening to this game. Watson just shot right off into the lead like there was no tomorrow, jetting ahead by thousands! I was blown away. Then he slowly started to crumble a bit and his answers were faulty more and more. But his strategy is amazing! He hops from category to category, while most humans pick a category and stick with it. I wonder if IBM has him do that purely b…

> He hops from category to category, while most humans pick a category and stick with it. Ken Jennings himself did that for the express reason of throwing off his opponents. As the category selector he would have an extra couple seconds to start thinking in that direction. Totally fair game for Watson to do too. I agree that Watson started with a higher value clue in order to look for the Daily Double. Statistically…

A while back, before the game, I wrote to Ken about one of the videos IBM made about Watson. In it, someone said that Watson "learns" the category. The best strategy for humans, then, is to start at the highest value questions.

When Ken jumped into a category, notice that he started at the highest values.

Re: Day 1: IBM Watson Ties for first on Jeopardy

#90

Post by thangalin »

estel wrote: post #22
cryptoz wrote: post #10

I couldn't have wished for a more exciting opening to this game. Watson just shot right off into the lead like there was no tomorrow, jetting ahead by thousands! I was blown away. Then he slowly started to crumble a bit and his answers were faulty more and more. But his strategy is amazing! He hops from category to category, while most humans pick a category and stick with it. I wonder if IBM has him do that purely b…

I've seen elsewhere it was suggested that whereas humans will want to get used to the question style of each category and learn from each one as they progress to harder answers, Watson has no such issues with jumping straight for high-value questions in all categories.

Incorrect. Watch the IBM video where they explicitly state that Watson learns the category, as well.
Post reply on HN