Personally, I agree with some of his theories (eg. that humans aren't really hard-wired to think in exponential terms). I certainly take his claims of immortality and mind uploading with a grain of salt, however.
Very interesting guy. I don't necessarily think all of his stuff is crazy outlandish. The key to think about with some of his theories is that his projections (stuff happening in the next 25 years) may not be all that impossible. Just like predicting a startup will fail 90% of the time and you'll likely be right, you can also say the same for crazy predictions like this. If we map his predictions to things that people 50 years ago were saying about our current tech now, some weren't too far off the mark, and we even surpassed some of the predictions (See the world's fair in the 30s as a good example of this)
That being said, he has still done some solid work in speech processing, he's definitely not just some crazy scientist. Google hired him FOR moonshot projects. If someone's not crazy enough to attempt them, who will?
Technology is a bit of a wild beast, I feel like Dr Kurzweil is almost the personification of its unpredictability/mystery in some ways.
Intelligence is not a function of CPU power, and all his AI predictions so far have failed.
The singularity seems to me like science fiction. Theory after data, not before.
Kurzweil has a pretty good prediction track record for what it's worth: http://lesswrong.com/lw/gbi/assessing_kurzweil_the_results/
The future always seems like science fiction (http://lesswrong.com/lw/j1/stranger_than_history/).