| What is consider efficient if it yields less best results? I understand time constraints and end goals may prohibit the 'best' approach. What I do not understand is how can you say something is more 'efficient' if the yield in understanding is less than what you would get with another method. Hopefully this will clarify my thought process here: If 'Best' is to teach others and requires 10,000 hours to yield 90-99% understanding. In contrast, 'Efficient' method requires 2,500 hours to yield 30-40% understanding. However, there is diminishing returns meaning that doubling your hours to 5,000 does not return you with 60-80% understanding, rather maybe closer to 50-60% understanding. With 7,500 hours closer to 65-70% and 10,000 hours may around 75-89% understanding. Here you've spent the same amount of time but did not achieve the same level of understanding. I think you may have a dynamic 'Best' vs 'Efficient' curve and to switch between those options to optimize maximizing your level of understanding in the least amount of time. |
Consider these two scenarios -
Goal: remember where to look up information when it comes up in $JOB Metric: how much you remember, how quickly you find the info
Goal: discover new hyper-efficient method of training an AI (or insert popular ML topic here). Metric: percent improvement vs current pubished best practice (deliberately vague) Required understanding to make progress: "like a Ph.D. from Stanford"
Now you can possibly measure something.
The idea achievement of "90% understanding" is VERY topic dependent. Simple topic? Sure 100% understanding, I remembered the Latin names of all of the plants in my house. Complicated topic? The information for "100% understanding" might not even be written in the textbook - it probably includes things like seeing the interconnections between the topics and being able to apply them in slightly different contexts.
Make sure you read the studies so you know what they're talking about. In this area, I think summaries are frequently misleading. You have to know what the real evidence is that substantiates the claims. (I cannot tell you how many times I have looked at the evidence and just rolled my eyes - obviously not applicable in settings where I wanted it to be.)
Edit: See this comment (not me) - https://news.ycombinator.com/item?id=41275869