The last thing may result from a hard-coded genetic heuristic learning rate. We can’t update fully Bayesian and a learning rate is an approximation given computational constraints. There is an optimal learning rate, but it depends on context, such as the trust in prior information, esp. the volatility of the environment. And thus it may happen that your genetic prior for your learning rate may not match the dynamics of your current environment. I guess our modern environment changes faster than the ancestral environment and most people update to slowly on new information. Updating much faster is probably adaptive. I also have that.
The last thing may result from a hard-coded genetic heuristic learning rate. We can’t update fully Bayesian and a learning rate is an approximation given computational constraints. There is an optimal learning rate, but it depends on context, such as the trust in prior information, esp. the volatility of the environment. And thus it may happen that your genetic prior for your learning rate may not match the dynamics of your current environment. I guess our modern environment changes faster than the ancestral environment and most people update to slowly on new information. Updating much faster is probably adaptive. I also have that.