Adaptive exercises analyse user responses in order to pursue a customized learning path, addressing the student's needs and level as these emerge through his/her responses.
[Diagram] [Example]

Auto-customizing exercises adapt themselves to the user and context, allowing multiple interfaces for the same material.
[Example]

Auto-created exercises take content from other sources and generate exercises automatically, based on that content. An example would be a page which allows you to enter a URL or paste in a text, then takes that text and makes a gapfill exercise automatically.
[Example]

Branching exercises allow students to make decisions on the learning path as they work through material; Quandary action mazes are a good example of this.
[Example]

Collaborative exercises solicit additional content from students, who can thus take part in the development of the exercise itself - for instance, by adding new story branches to action mazes. Students doing the same exercise at the same time might be able to “see” each other and discuss the task.
[Example]

Rich exercises contain a wide variety of media types (text, audio, video, etc.) Distributed exercises search out and include this type of content from a variety of sources all over the planet (and beyond). For example, an exercise related to astronomy might include a live feed from the Hubble telescope; texts may be auto-glossing, allowing instant lookup of vocabulary in a range of reference works.
[Diagram]

Standardized structures enable data and exercise content to be shared and repurposed by anyone; it will be easy to extract exercise data from one source and display it on some other device or use it in another way.
[Example]

Unicode means (eventually) no more font or charset problems -- never worry again about the target computer.
[Example]