Happily! I wrote about the idea about 6.5 years ago - all the essays are from back then:

"Details on Reinforcement by Covariance"

https://medium.com/aboveintelligent/details-on-reinforcement-by-covariance-75df1d9f4725

"Mixture of Experts with Covariance"

https://towardsdatascience.com/mixture-of-experts-with-covariance-6d4a167743ad

"Measuring & Learning from Error-Detectors"

https://medium.com/@oaklandthinktank/measuring-learning-from-error-detectors-ad8e835cb220