Some might argue that I'm easily impressed, but I'm pretty stoked on the new search filters Live Search have just implemented on their image searcch function. Going to http://search.live.com/images, try entering the name of your favourite celeb (or anyone at that), followed by "filter:face" (you can also try filter:bw or filter:portrait).
It's all done via Supervised Multiclass Labelling - basically, the search engine 'knows' what a picture of a face should look like and is able to filter our results accordingly. There's already a wealth of research on facial recognition, so faces are a logical thing to start with. It'll be interesting to see when this type of thing can be applied to video so you can actually search video content itself (and not just the tags and links that are attached to it). I'm sure there's a government agency somewhere that already has this capability :)
Good to see also that we're at least neck and neck with the big traffic search engines. The relevance of search results between MS and the market leader is roughly equal - check out this from Live Search compared to this (tested on 16/8/07). Both return a good number of hits, but in fact the Live Search engine has a greater number of relevant face pictures of Chuck in the top results than the other search provider. Plus, Live Search also provides links to people who are related in some way to Chuck in the right hand column. I'm not nusually a one-eyed Microsoftee, but it feels good to know our search engine is getting better so quickly.