Perplexity estimation enables antifragile learning
Personalized language model perplexity approximates surprisal. Similar to flow and stress, there is a sweet-spot of perplexity desired. Completely predictable resources provide little value. Completely unpredictable resources are also not desired, as they’re not pleasant to digest. Resources that are somewhat challenging but not impossible to tackle should be aimed for. This would lead to antifragile learning, an attitude of actively patching your weak spots and developing your thinking by systematically challenging yourself. The edge of chaos hypothesis also supports this approach if keeping in mind entrainment and the brain as a dynamical system.