katie on Nostr: Pretty much everything under the umbrella of spiritually, for one. Beliefs about what ...
Pretty much everything under the umbrella of spiritually, for one. Beliefs about what a family should look like. What someone believes is right for their children. What someone believes are appropriate boundaries or limitations for engagement with others. Most things in life aren’t black and white, and it’s ok if we don’t all have the same definitions/goals/beliefs. In fact, battling to push beliefs as universal ideals have created a significant amount of harm historically.
I think of ML models such as deep nets are a good metaphor for this, sometimes the weights converge in roughly the same spots every time if left to train long enough. Most of the time, they don’t - the data is too complex, or there are unknowns. The patterns it picks up are valid, but they’re a byproduct of their journey. They get stuck in local minima and sometimes it takes awhile to get out of that space, sometimes they never do. One model may be more “right” than another.. but only if we can agree on the metric that we use to assess that, which usually there is not a single one or it’s measuring the wrong thing, or our target variable is flawed. Especially when it comes to encoding, which is what humans are doing every day.
Most probability density functions are simplified estimates based on limited observations. You can use something like a KDE to more accurately model distributions, but they’re highly sensitive to the data they’ve observed. Universal truth is more rare than not.
Even gravity, which we had a mathematical equation for on earth, was proven to be only valid… on earth. And the “true” model is more sophisticated when you looked outside of earth - which is when the theory of relativity was created. We are all estimating reality through our own observations and those that we collect from others. It’s unreasonable to assume we all have or should have identical encodings.
I think of ML models such as deep nets are a good metaphor for this, sometimes the weights converge in roughly the same spots every time if left to train long enough. Most of the time, they don’t - the data is too complex, or there are unknowns. The patterns it picks up are valid, but they’re a byproduct of their journey. They get stuck in local minima and sometimes it takes awhile to get out of that space, sometimes they never do. One model may be more “right” than another.. but only if we can agree on the metric that we use to assess that, which usually there is not a single one or it’s measuring the wrong thing, or our target variable is flawed. Especially when it comes to encoding, which is what humans are doing every day.
Most probability density functions are simplified estimates based on limited observations. You can use something like a KDE to more accurately model distributions, but they’re highly sensitive to the data they’ve observed. Universal truth is more rare than not.
Even gravity, which we had a mathematical equation for on earth, was proven to be only valid… on earth. And the “true” model is more sophisticated when you looked outside of earth - which is when the theory of relativity was created. We are all estimating reality through our own observations and those that we collect from others. It’s unreasonable to assume we all have or should have identical encodings.