Hacker News Reader

Comment by adamnemecek | Hacker News Reader

a.) I’m taking most news coming from ibm with a grain of salt, there was so much commotion around Watson and nothing came out of it

b.) quantum digital is boring, quantum analog, now that’s flavortown

View on HN

I've been working with 'Watson' products for a while now and 'nothing came out of it' is an overstatement. They have some interesting products that have useful applications in business, however their marketing just doesn't align with their portfolio. They use so many buzz-words to get C-level hyped, but they never manage to live up to the promises. If you manage expectations there are some really interesting opportunities though, but never groundbreaking.

Ok what did you do with it.

First of all, I figured out that there's no 'it'. It's a portfolio of products ranging from image recognition as a service to on premise NLP and enterprise search. All these products have their own specific use cases, some of them have hardly any real world use cases.

What I've recently been doing is using NLP for document classification. It's the type of problem you can solve with many different solutions, but 'Watson' is the one we're using.

Sounds next gen.

They answered the question you asked, and contributed information to the thread. Don't snark in response to that.

I think it might have come across differently than intended. They are using it for something that doesn't seem to match the hype around Watson.

I think the NLP was what much of the hype was about though?

Unless I'm missing something here, "Watson" was definitely not positioned as a catalog of services when it famously played jeopardy.

That's the point being made; the "Watson" marketing and the "Watson" reality are at odds, but there is a Watson reality, even if it's more mundane than the marketing.

I rewrote my initial draft of that sentence to deny IBM their strategy of using Watson as a human name, which is part of the problem. You shouldn't really use a possessive on Watson, because nothing is Watson's as it isn't a person, grammatically or otherwise. Watson is a collective noun, which is a suite of products. Even if you nominally know better, it fools your brain if you let them frame Watson as a proper name of a person.

no, i guess it positioned itself as jeopardy-as-a-service

Let's just be clear about this: "Watson" as an entity does not exist. It is nothing but a brand, like "Microsoft" slapped in front of a disparate array of independent siloed products. Most of these are probably just acquisitions.

Watson Knowledge Studio - This is actually just something called SIRE (Statistical Information and Relation Extraction toolkit). The "Watson" in the name means nothing.

Watson Natural Language Understanding API (Note: This is just a product that was called AlchemyLanguage -- something they bought and renamed to include the word "Watson") - I can tell you everything that makes it crippled (you can build a NERC model using Watson Knowledge Studio, but the API has no way for you to extract e.g. negations, sub-entity types, etc. basically any entity attributes). Again, Watson means nothing, it's rebranding an aquisition.

Watson Analytics - It doesn't integrate into anything Watsony except maybe Cognos. I.e. It's completely useless for doing any analytics from any of the other Watson NLU/Classification products. The Social Media component has no way for you to integrate your own data. It's nothing big a big toy meant to be shown in demos. It is 100% useless in practice. I have no idea what they bought and rebranded here. I mean c'mon, I'm using your Watson sentiment analysis crap to do sentiment on Tweets and Facebook and your Watson Analytics product has no way of me getting this analyzed data back in to build visualization and see trends? With a lot of effort, basically somehow normalizing the JSON output from NLU maybe, but there is no icon that says "Import Watson NLU data".

Watson Natural Language Classification - This is just a multi-label text classifier. It's not too bad to be honest, except I don't like the limit of the number of texts per classifier and the limit of the number of characters per text. I also think they need to handle hierarchical multi-label classifications. My biggest problem is... I can't feed the output of NLC into any other Watson tool. How the fuck do I do my analytics? C'mon IBM, throw a man a bone here why must I do everything myself. Why can't I plug the output into Watson Analytics?

Basically everything is branded under a label called "Watson", but none of the products even _remotely_ integrate with one another. There are so many deep flaws in the products. The documentation often does not correspond to reality.

If someone at IBM would just sit down and try and make all these Watson products actually "talk to one another" this would be a really good enterprise quality suite of products. I really think they are close, but they're fucking it up in such a stupid way it dumbfounds me. I honestly don't think anyone at IBM will fix this. They are so big and so stupid at the same time.

Believe nothing you see at an IBM demo or any presentation. It is smokescreen and mirrors. You will have to build something yourself to see the flaws.

> quantum digital is boring, quantum analog, now that’s flavortown

What do you mean by that? You mean simulating on analogical classical computers? Or quantum computers without qbits?

The latter https://en.m.wikipedia.org/wiki/Continuous-variable_quantum_...

String (theoritical physics) computers?

Are there expected advantages from using continuous values?

I imagine error resistance would become incredibly harder. Does information density grow fast enough to compensate it?

This is actually a really long answer but to give you a flavor, on an analog computer, you can sort in o(n) https://en.m.wikipedia.org/wiki/Spaghetti_sort also you have integration and differentiation as primitives.

Or was it a pun about types of quark?

Reply on news.ycombinator.com