Discussing whether search is a solved problem reminds me of a talk Hynek Hermansky gave at ICSLP in Sydney in ’98 on why speech recognition isn’t a solved problem. Hynek made an analogy to flight, where at any point between balloons and jet airplanes, flight might’ve been considered solved.
SearchMe.com is great. They provide reduced resolution page views with overlayed snippets, and it’s fast. This just feels right for navigation searches (e.g. typing “lingpipe” to try to find our home page or blog). And the page views add more value than I could’ve imagined to the snippets. Like Páraic, I’m so enamored of it that it’s replacing Google as my default search engine on Firefox (just click on the link in the upper right of searchme.com’s home page). I sure hope they can scale as more people find out about them.
Cuil.com, by focusing on recall (and marketing), seems less useful, even if they get the bugs ironed out. Despite the fact that we’re focusing on recall for genomics information extraction tasks, I’ve never felt recall was an issue for most web searches. I could use more approximate and contextual matching, perhaps, but the index size has never seemed an issue.
I miss Excite.com, which used to run TF/IDF rather than social-network-based search ranking algorithms. I missed Excite even as I was starting to use Google for many searches. But then again, if Excite had been successful, we wouldn’t have the Apache Lucene search engine.
PowerSet.com was focusing on some kind of precision and question answering (and marketing), which I also felt was of questionable value (for me as a searcher; it clearly worked for their VCs and founders) compared to using Google. Plus, they never showed (at least to the public) that their tech scaled either in complexity (different page types, multiple pages for entities) and size (number of web pages).