In a Search Refinement, a Chance to Rival Google (NYT article on Powerset) | Hacker News Reader

In a Search Refinement, a Chance to Rival Google (NYT article on Powerset)

nytimes.com

3 pointsbootload5 comments

I attended the Startup school in March and since I'd flown 12 hours to get to SF, I thought I'd try to meet up with some companies to drum up some interest in my startup. Amazingly this ended up with me pitching (my first ever - eek!) to Barney, Ron Kaplan and Lorenzo Thiel - the main guys at Powerset

They liked my idea - not enough to give any cash though:( - and I'm on their beta testing list.

They're keen to use Powerset as platform for other applications to exploit their NLP index via APIs.

I'll update you when I get my first play with it...

P.S. Bit late Paul G, but Barney sends his regards...

I didn't realize Powerset was based on technology licensed from Parc. That probably means they're doomed. In software (unlike biotech) few startups succeed based on licensed ideas. The founders have to be the people who have the ideas.

But aren't there tons of details in NLP that these founders shouldn't think about? Also, they're getting a Parc researcher as their CTO as part of the deal.

I get the feeling these guys are more hype than substance. One thing I don't understand is why drum up _any_ hype before you've actually released anything? Seems like when they finally release, it'll be a huge anti-climax.

'... I get the feeling these guys are more hype than substance. ...'

One thing they don't have is a clue about building a product compared to creating technology. The thing that tips me off is using 'NLP' and 'natural language' as a description of what the product is about. It reads more like ... "we have this great technology from PARC, what can we use it for? Ahhh search is a natural fit."

Just compare this to say google. I don't remember too many press reports sending potential users the message "google is powered by pagerank and is reliability because it uses linux based distributed computing" . Now I remember pagerank was bandied around a lot at the start and mentions of thousands of linux boxes. All this subsided when people started using google and it just worked.

'... Seems like when they finally release, it'll be a huge anti-climax ...'

This will more likely be a result of mis-application of NLP. NLP is a can of worms and highly dependent on context. You need to understand the meaning of things. To understand the meaning you either have to severely constrain the application and get something like Microsoft Clippy ( http://en.wikipedia.org/wiki/Clippy ) or frustrate users with some horribly complex search process.

Is the natural language approach, a bad idea for search? Is this what pg means by 'solving partial solutions' but being 'trapped on a local maximum'?

Lets wait and see the demo.

Reply on news.ycombinator.com