Hacker News Reader

Comment by Joe8Bit | Hacker News Reader

I'm not sure if my expectations were overly optimistic, but using the 'Ask Watson' feature in the beer demo gave REALLY disappointingly inaccurate answers to very simple questions about beer.

e.g. "What is beer made from?" resulted in what appeared to just be a link to a wikipedia article on Brewerys. Not sure if this is implementation issues or representative of the underlying Watson API.

View on HN

The current beer corpus used by Brewery Master only contains a small number of documents. Due to its small size, its very likely the the answer Watson is searching for doesn't exist.

Totally see your point, but it makes me wonder why IBM point to this demo as an example of Watson API usage (and how good it is) when the demo doesn't live up to (anything like?) Watson's true potential.

Strikes me as a little counterproductive from a marketing POV.

I helped a friend's company integrate Watson question answering into their product. My personal opinion is that Watson NLP question answering product will probably be awesome sometime in the future but to me it seems like an evolving product still in its infancy.

BTW, the beer app runs on IBM's BlueMix PaaS, which is pretty nice: generous free tier and it is built on Cloud Foundry technology so I had no learning curve for using it. Anyway, if you have an app that does not quite fit into the free tiers for Google or AWS, BlueMix has a dev friendly pricing model.

I had the same experience.

Who makes a saison near washington, dc?

Watson: Wikipedia on low alcohol beer

Google: 1) Washington Capitols schedule 2) Atlas Brew Works (they make a saison in DC)

Kind of funny that Google silently substitutes "season" for "saison" which is where the first result comes from.

well, 'saison' is french for 'season'.

True, but if I just search for 'saison', all the results are for the beer style.

Agreed - I am starting to understand that as stunning as it was to see Watson win at Jeopardy, the actual range of inputs was fairly limited in comparison to the massive spectrum of inputs in unrestricted natural language.

Not trying to hate on IBM, huge respect for Watson developers, I just think there are some growing pains translating this to real world.

Reply on news.ycombinator.com