A renderer can calculate values brighter than a display can reproduce. That isn't an error. A scene may contain a dim wall, a sunlit window and a reflection of a light source in the same image. The problem is deciding how those relationships should fit into the smaller range available on screen.
Troy Sobotka's Filmic configuration for Blender makes that decision an explicit part of the image pipeline. The renderer produces scene-referred values. A view transform maps them into an output representation, and a look changes the presentation of contrast. Lighting and display interpretation stop being the same operation.
Consider three neighboring scene-linear values: 1, 2 and 4. If the output path clips everything at 1, all three become the same maximum display value. A highlight that contained a smooth variation turns into a flat patch. Lowering the light may recover the patch, but it also changes illumination elsewhere in the scene.
A compressive transform can preserve a visible difference between those bright values by allocating less display range to each successive increase. This doesn't make the monitor capable of reproducing the original physical brightness. It makes a picture that retains more useful relationships within the monitor's limits.
Exposure is a separate control. In a simple scene-linear model, increasing exposure by one stop multiplies values by two before the display transform. This shifts which part of the scene occupies the useful viewing range. It is different from adding a constant to every pixel or brightening an already clipped image.
Color complicates the story because the channels don't always approach the limit together. A bright saturated surface can change appearance as its values are compressed. Filmic includes treatment of high-intensity color as well as the overall range. It is worth looking at colored highlights, not only a grayscale ramp, when judging a transform.
For a meaningful comparison, I would keep the scene, camera and exposure fixed while changing the view transform first. Otherwise a nicer image may simply be the result of different lighting. After understanding that difference, lighting can be adjusted with the new viewing behavior in mind.
The benefit is a better separation of decisions. The scene describes light and materials. The display transform decides how to show their values. If a bright area looks wrong, I want to know which of those decisions needs attention before reducing every lamp in the scene.