Think Empirically

Concrete examples

Wow! Thinking with empirical examples seems to work way better than just abstract theorizing. I was trying to work out the scientific method using one simple concrete example. Suddenly, I was forced to understand that I had forgotten some important facts about hypotheses. I had forgotten that I needed to have a causal model making the predictions, not just a bunch of predictions by themselves. One concrete example did what a week of theorizing didn’t.

How much time to spend optimizing something?

Look at this XKCD comic: Is It Worth The Time?

Study empirical resources

Test this idea: write down the causal structure for the paper. For example, if it’s a memory paper having a table of error rates and number of training trials, show the putative causal structure: number of trials -> error rate. This way, you can figure out what hypotheses they’re actually trying to test.

Multiple-stage fallacy

Eliezer’s facebook post on this: link.

Breaking down an innocent proposition into (counts) 31 stages and “eliciting” probabilities for the stages engenders a visceral sense of indignation in me. It is a code violation to do that and then tell people they have given their own opinion.

If you want something to improve, you have to work on it

For example, if I want my biceps to be as big as those of my gym buddies, then I have to do biceps exercises every week instead of just skipping them. Ditto for triceps or chest or whatever. Similarly, if I want to get bigger muscles, I have to hit the gym every day like my friends instead of going whenever I feel like it.