learning curve

[Arne Babenhauserheide]

Am Dienstag, 17. November 2009 22:38:39 schrieb olafBuddenhagen@gmx.net:
> The problem with learning bit by bit is that you only look up things if
> you want to do something new. You never get a complete picture; you
> never learn how you could do things more efficiently, and/or with better
> result; and you often pick up really bad practices.

I tend to disagree here, too. 

You do pick up back practise if you only check what is absolutely necessary to 
get the task at hand done (as I often do for shell scripting). 

If you check deeper issues when you need them, you understand something new 
and you learn to work more efficiently. 

Look for example at the Mercurial guide I wrote. At first you only learn to 
commit and read your log. At that point you already understand that Mercurial 
tracks your changes - and after using it a bit, you also get a feeling for 
what commit does. 

Then you learn how to do nonlinear development, branching and merging at will. 
Committing is already natural at that point, so you only enhance what you 
already know by heart. 

And after that you learn that working together with others is simply nonlinear 
development by exchanging "commits" between repositories. 


In really complex areas that becomes even more evident. 

One example: I'm studying physics, and I learned this summer with the Feynman 
lectures, which hammer home the point that statistics tell us that the 
distribution of particles with certain energies is exp(-E/kT) - that's "e" to 
the potential of minus the energy divided by the temperature (and the 
Boltzmann constant). He explains that for gases at first (energy distribution 
in different heights - only from gravity and random movement energy). The 
distribution says "this many particles with Energy E are there". 

At that point he never talks about the difference between bose particles and 
fermi particles. He also doesn't try to give the whole mechanism, but rather 
gives a central part of the whole picture. 

Now when I got to learning suprafluids and stuff, it was quite easy to 
understand what their slightly different distribution does: 

1 / (exp(E/kT) - 1)

That's almost exp( - E/kT), but for low energies it goes to infinity - because 
the lowest state of a suprafluid can be shared by an arbitrary high number of 
particles - if you only manage to take away enough energy from them. That's 
why it can crawl over walls, ignores rotations of the container and such. 

To really see the implications of that, you already need to know about , 
Heisenbergs uncertainty relation for the gaussian distribution of energies, 
quantum mechanics, energy barriers and stuff. But you don't need to understand 
that to grasp the basic law exp(-E/kt). 

And really understanding the basic law makes it much easier to understand more 
complex stuff later on - understanding everything at once is just not feasible 
for the vast majority of physics students. 
When you already know exp(-E/kT), many later things are "wow, it's really easy 
to see how that works - just a small alteration to the basic distribution". 

(there are more basic principles in physics than this, but that's one which 
currently fascinates me; it is so easy - once you udnerstand it :) 
And Feynman really manages to make physics sound as fascinating as it is, 
while keeping it easy to understand). 

To organize learning that way makes for a very efficient learning curve. 

(actually he starts with "all matter is made of atoms (as long as we don't 
look to deep)" and "we begin with small lies which make it easier to 
understand the basics - but we tell you which laws are final (to our current 
knowledge) and which are simplifications we'll have to revise" and goes onward 
from that). 

> In either case, you can't seriously argue that it's demanding too much,
> that everyone learning how to set the text color, should also learn how
> to set the background color at the same time, and vice versa...

And the button color, and the text field color (almost no site changes that), 
... 

What's missing there is a way to adapt to user settings. What you describe is 
binary again: Either set all or nothing. But that means that it doesn't 
integrate at all or integrates completely - without middle ground. 

But we already had that part of the discussion... 

Best wishes, 
Arne

PS: I think that this can be relevant to the Hurd, because the learning curve 
is something which also affects every program, translator usage, etc. - and so 
it affects how easy it is for people to switch to the Hurd.

signature.asc
Description: This is a digitally signed message part.