Ask IBM's Watson Research Team Anything | Hacker News Reader

Ask IBM's Watson Research Team Anything

reddit.com

106 pointsww52027 comments

This is either a question or a suggestion but.. is there (or can there be) a more technical AMA-type site for engineers and scientists? I like Reddit but I can't help but feel the right person could curate the questions better for technical and scientific stuff rather than relying on the sort of voting patterns Reddit is susceptible to.

Frequent intelligently curated but publicly driven (i.e. not just "interviews") AMAs with scientists and teams working on cool stuff would make for great reading. Even AMAs with the tech teams of the startups we all know and love on HN would be a great start.

I'd do it (the "if you're gunna complain.." approach) but I'm not well enough connected. Imagine someone like Andrew Warner or pg (who could pretty much get anyone at the table) curating a site like that. It'd be a must-visit to both ask questions and read the answers.

UPDATE: Or is this essentially Quora I've reinvented?

Reddit's solution to that is to create your own subreddit and effectively become elitist about what's allowed. Hence a lot of comments as to Reddit demographic issues is met with "remove default subreddits".

Quora is similar but catered to a smaller audience I assume. Both places have moderators but I don't know if either can control the answers (On reddit, only moderators can).

That solution doesn’t work for AMAs which will (like this one) inevitably attract much attention.

I was rather disappointed that a lot of the top rated questions were knee-jerk or obvious.

I am under the impression that the actual IBM has not responded to any of the questions yet. Any knee-jerk or obvious answers would be from your typical Reddit user.

They will answer the ten most popular questions next tuesday.

As much as I love Reddit sometimes, a good number of the upvoted questions are disappointing although not unexpected.

>They will answer

or Watson?

That's almost always the case with reddit AMAs that accept the highest-rated comments.

Try sorting the questions by "best". The quality seems to improve a bit.

I'd rather play "Ask Watson Anything"

I wish they would put him up as a webapp. It may not be economical to have a supercomputer online that can only answer one question at a time -- but they might find enough overlap that he could start caching frequent requests or even queuing requests as necessary. Or better yet, scale slowly by invite only.

Why do you assume he can only answer one question at a time?

It would have been better phrased as: how many questions per second can watson handle.

And I think the answer is, not a whole lot, maybe even less than 1.

That would be great.

Three days ago there was a comment with a back of the envelope calculation to get to the cost of running Watson: http://news.ycombinator.com/item?id=2218212

Did this competition hurt Jeopardy in general because it is now obvious that the winner is mostly decided by buzzer skills among top players?

No, I don't think so. It was always known that buzzer speed was always an important factor.

I don't think I realized the degree and I've been an off-and-on regular Jeopardy watcher over the years. I don't see this subtlety as hurting Jeopardy's brand any though. It does make me wish that there had been a better platform for competition of this sort that would have had a more level playing field.

(The humans actually had some advantage of anticipating the buzz-in signal so I'm not saying that Watson had all the cards. I'm just saying that a lot of the contest hinged on a factor that inherently could not be made apples-to-apples.)

I'm sure the same thing comes out of Jeopardy "Tournament of Champions" games where a number of very good human players come together. It's just that only regular Jeopardy watchers every noticed those games were taking place.

It just seems more pronounced with Watson because it drew in a large crowd of new (but perhaps one-time) viewers.

With three of the best Jeopardy players ever, I thought they might make the questions a bit more challenging. I can see why this might not work in this case because Watson was programmed using regular questions. But I agree, I would have liked to see the skill level rise with the caliber of player.

I doubt it. Jeopardy's a game, not an exam.

Landmark AI event, and you guys are worrying about buzzer speed. *sigh

That's because the buzzer competition confounds the knowledge competition. It's a shame, really.

But I'm sure IBM is happy.

Can you explain why it is you think that the knowledge part is confounded? If Watson is fast but incorrect, it's a failure.

The knowledge part is confounded because buzzer performance preempts knowledge performance.

Can you explain? (I didn't quite get the preempt part). Watson seems to be getting answers correct as well

A Watson that knows 70% of the answers will beat a human who knows 90% of the answers, purely due to better reflexes. Of course, buzzer skill is partly how KenJen won so many games, so I'm not complaining. Just not AS impressed by Watson as some people.

Except that it's not just reflexes. Watson has to compute the answer before the humans. That's no small task. This video is a great technical explanation of how it works:

http://www.youtube.com/watch?v=3G2H3DZ8rNc&feature=youtu...

Reply on news.ycombinator.com