Questions and sometime half thoughts

Aging as an evolutionary advantage

Human, throughout our life, gather experience and the older the man, the wiser they become. So these older man will have a much better survival and procreative advantage. If there is no aging and slowing down of the old animal, future generation will be less diverse. This homogeneity make the whole species an easy target for a sweeping disease. Longer the time, higher the risk.

So you can see, taking advantage away bring back the balance and give the species the strength via diversity.

Making money

Man makes money either by realizing a Platoninc ideal or corrupting one. I want to be in the 1st category.

Have you ever seen a failed machine learning project ?

And the answer if yes, you haven’t seeon one, then ask yourself if you have seen the boundaries of what’s possible ? If all that you see are successes, how much of it is just spending time on worn out path doing something everyone has done ?

Dataset drift detection is a silly idea

The whole point of model based machine learning, or modelling as a method of understanding the world, is that your model can generalize. From practical point of view, generalized model can perform well on unseen (even drifted) data. From understanding point of view, genralized model encapsulate what man think how the world works. Focus on making your model genenalizing well.

Writing code that’s breakable with tests

Taking the idea of modern scientific discovery by Karl Popper, a programmer should write code that’s breakable in tests. This is crucial for production grade code, where we asolutely need to know the boundary of the logic of the code and how it can break.

The Art and Science of Cause and Effect - Judea Pearl

When God asks: “Did you eat from that tree?”
This is what Adam replies: “The woman whom you gave to be with me, She handed me the fruit from the tree; and I ate.”
Eve is just as skillful: “The serpent deceived me, and I ate.”
...

explanations are used exclusively for passing responsibilities.

Then would a man, the one who can explain things, feel guilty if the outcome is bad, or would he feel deserved the outcome when it’s good ?

Does god have human emotion ?

Look, He can create all these cool stuff with ease, on a whim. Does he even feel bored ? The man of ultimate wisdom and love, why does he leave all the atrocious things happen on Earth ? Hmm. Does he feel anything ? Yes, no ? If no, why would we pray for his love. If yes, would his feeling be equal for all parties ? If someone do something horrible to someone else with for a blind reason, would he let it happen ? But anyway, he’s dead now and mankind had killed him. Now what do mankind fear ?

Metrics

All metrics are wrong, but some are useful. more

Tech debt

From business perspective, tech debt is a sign of growth, where business functions outgrows softwares. From labor perspective, it creates jobs. From engineering perspective, it creates fun puzzle to solve. What the beef again ?

Study

Confusion should open the door for your curiosity.

Logging

Logging is programmer’s insurance. Larger the program, higher the premium.

True randomness

Is there objective randomness out there, which is not just randomness observable by a man (subjective randomness) ? And how, epistemically speaking, can we answer this question when given an seemmingly random event, ie: can we tell if this random event happen because it’s truly random, or we observe it to be so because of our inability to perceived, or sense the whole of it. Heisenberg princible comes to mind …

Grandma

Pity the one whose grandma doesn’t cook good food

Thinking and calculating

Without gaps in knowledge, thinking is reduced to calculation - Byung-Chul Han

Best technologies

The best kind of technology is the invisibile one, like wifi

###Value of a test set The more you use a test set, the less valuable it becomes. In some cases, we can measure this value by something like information gain i.e how much information you have gained (in bits) from modifying your model using your test set. As a result, the best test set is only used once.

Written on