As a engineer, I sympathize with him completely. I would react the same if I had worked on a really really tough problem and made an incremental advance, which I thought was a breakthrough, but everybody dished it because they didn't understand it. Not to say what Cuil is doing a great job. But Machine learning and NLP is a tough thing.
I would react the same if I had worked on a really really tough problem and made an incremental advance, which I thought was a breakthrough, but everybody dished it because they didn't understand it.
I was once part of a company that thought it had superior technology that everyone else "didn't understand." It was true, it had marvelous technology. It was also true that lots of people didn't understand it.
Here's what it also had: 1) a deployment system everyone agrees is a pain in the @@@, which they never paid much attention to, 2) user interfaces which didn't repaint themselves like native widgets and which looked terrible and had terrible APIs, 3) intermittent bugs the customers kept reporting and getting disregarded and blamed for, 4) a less than open, "not invented here" culture with a big dose of "us versus them."
It's cool to make an incremental advance. The hard part is getting people to understand it. The hard part is reaching your audience. For this, it's more productive to enlist their help, not to blame them.
I think it may be the format the result is presented in. It acts as if it can give the world then gives very little. If they were upfront about the level it is at by making the design more about listing pieces of information rather than trying to assume they all mesh together nicely they may get better reviews.
How it's presented is a huge part of why people reacted so negatively. They present the articles as prose, so people expect it to flow like prose. This makes them very sensitive to disorganization.
I think they're deserving of the criticism they've received. Cpedia is an interesting idea, and I assume they've done a lot of really good work we're not grasping. But it's not ready for public consumption yet, and they shouldn't have pretended it was.
They were over-zealous. It happens. What's important is how they move ahead. I'm with wheels on this one. We shouldn't write them off yet.
No doubt it's a huge problem area, been researching around areas that would be involved in this like clustering and summarization,
Google Squared attempts to be less ambitious and still feels very incomplete and they probably have some very smart people and the backing of the most complete index of the internet and probably the most computational power afforded to any work in the area.
It might work better if they just presented a list of results, and users could up/down vote the results for relevancy. Overtime they could use that feedback to float more relevant "snippets" to the top. Kinda like wikipedia but without all the hassle of typing.
How about using mouse tracking like the page analytics guys do. What people are hovering their mouse over is what their reading ... and that constitutes a "vote".
I appreciate that it's a hard project and I appreciate the work that's gone into it. NLP is hard, and any progress in it is always amazing and a worthwhile endevor. However, the measure of a product is not how hard you worked on it, but how well it works.
Cuil/cpedia doesn't work. They shouldn't be releasing it as if it were a product. Progress towards the solution is great, but unless it reaches or almost reaches the solution, it should be published in an academic journal.
By releasing it in the wild, you set an expectation for how much progress has been made. Blaming the user for not understanding the problem is incorrect; the user is not supposed to see the problem in the first place. If your users need to understand the problems that you've had under the hood, then your product isn't ready for prime time.
