He presents prediction as new, ignoring google suggest; predictive text on phones/tablet keyboards; tab-completion in bash/vim; intellisense etc. And it obscur

He presents prediction as new, ignoring google suggest; predictive text on phones/tablet keyboards; tab-completion in bash/vim; intellisense etc. And it obscures the part that is new: And the second is what actions would lead to useful results. No longer statistical, but based on the (presently) unknown! Not predicting what you are likely to do, but exploring ahead of you. Not based on your past, but your future... A very cool idea. Actually, at present, it seems to be just a secondary ranking filter of statistical suggestions, so it would never consider an original action (i.e. that wasn't already likely). However, that kind of exploration would really be expensive. Another problem is defining "useful results" generally, so it applicable to results not seen before.