Buy low, sell high.
That’s what everyone says.
Problem is, it’s not as easy as it sounds:
People get so excited playing the game they wind up buying high and selling low.
Don’t let it be you.
Use the whole brain,
The temporal lobe, the rostrolateral cortex, the PFC,
not just the emotional limbic system.
Count the cost, run the numbers,
Watch it unfold in your mind’s eye until you reach the end game
And then, if you still imagine a regret-free retrospective, go for it!

When it comes to machine learning, investment comes in the form of r&d focus,
You find a problem, get an idea, write the code, machine learn, and iterate.
Train right and you’ll see dividends return
Train wrong, you’ll still learn something useful.

But sadly there are other outcomes that too much R&D falls into:
Ideas that were never written down,
Partially written codebases that never saw the light of GitHub,
Models that never saw multiple trials,
And the myriads of other stages that development stalls in.

Problem is, there’s just too much to do,
Too many papers to read,
too many new ideas to try
People see everyone on arxiv having fun, and they want to join too,
But that’s not where your fun is, isn’t it?

You gotta keep your entropy low
trajectories straight and narrow
Focus unswerving
Until you see it through to the end

Use what works. Don’t write the library++ for every import you make.
Don’t get caught doing insider trading every time the latest and greatest paper gets published,
Predict your dynamics,
Follow the trajectory,
Let your actions control your observations and not other way around.

And then your reward system will be on fire.
It’ll be blazing with activations.
You’ll be optimizing faster than Adam!
You’ll reach minima people told you didn’t exist!
And even if that wasn’t your origonal goal,
The collective intelligence will remember you for how are you contributed to it.

So don’t give up when it’s rough or tough,
Go far beyond what you think is enough.
Persevere in whatever you try,
And buy low, sell high