Calling this a discovery make it sound like we've discovered that neurons actually do this. It's rather lacking in experimental support--especially the kind of support that it is following the free energy principle rather than something else. (For instance, a more simpleminded information maximization--see, for instance, https://pubmed.ncbi.nlm.nih.gov/10896164/--which has been shown to sometimes but not always be a pretty good approximation.)