Hacker News Reader

Comment by CraigJPerry | Hacker News Reader

I disagree, tight feedback loops are always better in every scenario than loose ones. What you do with the new information matters though.

E.g. you would always choose to read an accelerometer sensor in a tight loop in a flight control system - however, you wouldn’t just blindly apply the latest reading as an input to your system (accelerometers are extremely noisy, the data looks like complete trash on the front end before applying some kind of averaging filter).

Further down the chain on a flight control computer is also a good counterpoint to this idea that you should prefer a more lossy sample rate.

Control theory is an established field of study. Better to adopt filtering & PID control strategies than to under sample an input.

View on HN

It's better to adopt filtering instead of reducing your sample rate, only if the opportunity cost of sampling and processing is sufficiently cheap (given scenario). If it's relatively expensive to sample and process (which I think apply to many business type scenarios),, just decreasing your sampling rate to more appropriately match your ability to manipulate outcomes make sense.

The old destructive vs non destructive testing idea. We measured the screw but we destroyed the screw. It had good quality. Now, how many screws do we need to sacrifice to be on top of our quality?

> are always better in every scenario than loose ones.

Sounds a little unconvincing.

> applying some kind of averaging filter

How do you choose the size of your averaging window?

How do you make sure you don’t catch a systematic bias due to some unrelated event?

Control theory may help in some cases (flight control) but possibly less so in others (stock market).

>> Sounds a little unconvincing.

Yeah, it is pretty wooly. How about taking it from the opposite position, almost like a proof by contradiction, e.g. the opposite position might be “you can see better if you look less often”.

I feel like you're conflating "sampling rate" with "tight feedback loop". Yes, a higher sampling rate (assuming each sample remains at or above the level of quality of your lower rate samples) is always better. But a tighter feedback loop is one where you are making decisions with that data at a faster clip too. As pointed out in the article (and by yourself), this is not always good. If you are using some averaging filter, as you say, then that is being "looser" about your feedback loop. You are recognizing that just because the accelerometer is telling you that you are going down doesn't mean you are about to crash. Someone with an overly tight feedback loop would not be so careful, and might immediately course-correct as soon as they sense they are going down unintentionally.

Depends on the assumptions, tight feedback loops can be based on less information each loop. For example an image sensor collecting light.

In that example you can have excessively tight feedback loops where looking less often let’s you see better.

I would read the opposing position as "you can see better if you look for longer periods of time".

You're talking about feedback loops in Control Theory, which is an established scientific field with strong mathematical underpinnings.

The article is talking about "feedback loops" in some handwavy business-marketing-whatever bullshit advice sense.

The article is talking about feedback loops in terms of learning, not "business-marketing-whatever bullshit advice" and I think the author makes this clear.

This paragraph sums up the motivation of the article quite nicely:

"But after achieving proficiency in a field, tight feedback loops are useless. That’s because initially, the learning environment is gentle. The path is well-travelled, there are easily accessible guides, things work according to common sense. As we become proficient, the environment becomes harsher, with noisier feedback. Improving is not as easy. At higher levels of skill, further progression depends on self-learning, on discovering or inventing new practices and knowledge. As skill increases, the gap between optimizing for metrics and optimizing for mastery widens."

Reply on news.ycombinator.com