There could be several reasons, I'm sure. One particular explanation that sticks out to me is a concept called 'feature integration theory'. I mention this because the things you say you remember -- what the guy in a video looked like, a time when your spelling was auto-corrected, general mundane details or 'features' -- are in line with this theory. You might find this document helpful, but as it is rather long, I'll provide a shorter explanation.
Feature integration theory is a theory of attention developed in 1980 by Anne Treisman and Garry Gelade that suggests that when perceiving a stimulus, features are "registered early, automatically, and in parallel, while objects are identified separately" and at a later stage in processing.
In other words, our perception of features are quicker and more automatic, registered early on. The object itself is recognized later, and thus we do not always fully integrate it into our attention. We must provide ourselves enough time to integrate the features that we have perceived into one main object or idea, and if we shift our attention away during this mandatory time of attention, we will not recognize the object as one larger entity or idea, but rather a collection of the features that we noticed automatically. This also might explain why you retain less information while multi-tasking.
Treisman distinguishes between two kinds of visual search tasks, "feature search" and "conjunction search". Feature searches can be performed fast and pre-attentively for targets defined by only one feature, such as color, shape, perceived direction of lighting, movement, or orientation. Features should "pop out" during search and should be able to form illusory conjunctions. Conversely, conjunction searches occur with the combination of two or more features and are identified serially. Conjunction search is much slower than feature search and requires conscious attention and effort. In multiple experiments, some referenced in this article, Treisman concluded that color, orientation, and intensity are features for which feature searches may be performed.
Indeed, this theory is supported by the existence of Balint's syndrome [1], a not-quite-understood condition in which damage to the parietal lobe renders the patient with a variety of symptoms, including simultanagnosia, or the 'inability to perceive simultaneous events or objects in one's visual field.'
This spatial disorder of visual attention—the ability to identify local elements of a scene, but not the global whole—has been referred to as a constriction of the individual's global gestalt window—their visual "window" of attention. People fixate their eyes to specific images in social scenes because they are informative to the meaning of the scene. Any forthcoming recovery in simultanagnosia may be related to somehow expanding the restricted attentional window that characterizes this disorder.
Simultanagnosia is a profound visual deficit. It impairs the ability to perceive multiple items in a visual display, while preserving the ability to recognize single objects. One study suggests that simultanagnosia may result from an extreme form of competition between objects which makes it difficult for attention to be disengaged from an object once it has been selected. Patients with simultanagnosia have a restricted spatial window of visual attention and cannot see more than one object at a time. They see their world in a patchy, spotty manner. Therefore, they pick out a single object, or even components of an individual object, without being able to see the global "big picture."
Sources:
[1] Udesen H, Madsen AL (May 1992). "[Balint's syndrome--visual disorientation]". Ugeskr. Laeg. (in Danish) 154 (21): 1492–4.
