Part 2: Here is one of my earliest attempts at summarizing all of Evan Usler's work, see my [Google drive](https://docs.google.com/document/d/1Vhwa6vdphscbS3beajCWBZhRkraPsEwjm8AI497Q8og/edit?usp=sharing). Since then I've improved upon it which I'll share at a later date on Reddit. Here is some of Evan Usler's fantastic work: Minimizing prediction errors: * Adjusting the precision (confidence) of prior beliefs and sensory input. * **By perception (updating prior beliefs)**: when sensory precision is stronger than prior precision (staying still and updating one’s beliefs to align with current sensory input). To do this, one must decrease sensory precision before action. Imprecise prior beliefs may increase sensory precision during speech production. PWS may exhibit imprecise prior beliefs regarding when sensory consequences of action are likely to occur. Predictions include not only expectations of the timing of a sensation but expectations of where in the sensory space they are likely to occur. Imprecise prior beliefs may result in increased trial-by-trial spatial variability of self-generated actions. The difficulty of PWS in predicting the consequences of sensory input is suggestive of imprecise prior beliefs in predicting speech-related sensory input. As a result, sensory precision via attention may increase to foster model updating. This increase in sensory precision could, in turn, prevent the sensory attenuation necessary for syllable initiation. Speakers can only consciously intend their sensory input and attend to their speech subsystems in realizing that sensory input.  * **By action (modifying the environment)**: when prior precision is stronger relative to sensory precision (so that the current sensory input changes to match one’s predictions) * Speech-related sensory input yields sensory prediction errors, which are mitigated by closed-loop motor reflex arcs in the brainstem and spinal cord. * **Prior precision**: It’s the confidence of our prior beliefs about the environment.  * **Sensory precision**: It’s the confidence in the fidelity (i.e., likelihood) of the sensory input. Sensory input: 1) exteroceptive information, including auditory feedback; 2) proprioceptive or somatosensory feedback from speech musculature; and 3) interoceptive feedback associated with internal functioning such as respiration and autonomic activity.  * **Paralysis by analysis** may occur when excessive sensory precision disrupts the efficient action-perception cycling underlying fluent movement. * Attention balances the relative influence of prior beliefs and current sensory input on inference processes, ensuring smooth action-perception cycles. Strong prior precision is associated with low attentional deployment (thus attenuating sensory precision) to more predictable sensory input.  * Initiating action requires disattending (i.e., decreasing sensory precision) to current sensory input at initiation. * Agentic control may be a product of a model’s high-level meta-awareness of the regular and reliable action-perception cycling for efficient prediction error minimization. * Stuttering is reduced during choral reading, because of distraction (i.e., disattending) from the self as speaker (that reduces sensory precision).  * Adaptation effect: Over repeated readings of a passage, the reader may increase precision to prior beliefs regarding incoming sensory input associated with the letters, syllables, words, and sentences (updating of more accurate and precise prior beliefs which reduces attention (sensory precision) to the orthographic features and auditory feedback). **Bayesian Inference** * Constantly updating its beliefs about the world based on incoming sensory data and prior knowledge. Predictive coding is a mechanism through which this Bayesian inference is implemented. * **Belief updating** is facilitated by the precision (i.e., confidence) placed on descending prior beliefs and ascending sensory input. In other words, precision is a second-order prediction of context (e.g., how well you hear an utterance) associated with a speech-related prediction of content (e.g., what utterance you expect to hear).  * **Predictive coding**: Constantly generating and updating predictions about sensory inputs. Predictive coding involves generating predictions about incoming sensory input and then comparing these predictions to the actual input. When there is a mismatch, the brain updates its internal model to minimize future errors. This process helps to reduce uncertainty and maintain a stable internal model of the environment.