You Have to Feel It

Smart move by Mitchell to omit (in his opinion) _why_ you have to feel it, as evidenced by the spread interpretation in the comments.

In my opinion, you have to “feel it” in order to do your best work.

However(!), and also in my opinion, you shouldn’t always strive to be in a position where you “feel it”. While it is important to spend most of one’s life feeling it / doing their best work in order to be fulfilled, the hazard of insatiably “feeling it” is that you can much more quickly burn out.

Working with passion fuels a level of intensity and emotional involvement that can take a while to recover from if you don’t get the result (read: success) you desired.

But yes, you do indeed mostly have to feel it.

Feeling some moderate positive emotion as a result of your work is not incompatible with 9 to 5 work.

The bigger problem is usually the opposite: nagging negative emotions, feeling annoyed, feeling contempt towards some parts of the work that one is bound to do. These emotions are unbecoming, so the psychological defenses hide them, as if there's no feeling at all. This is what "mind-numbing" work often is.

Yeah, do not disagree with either of those statements.

  > Working with passion fuels a level of intensity and emotional involvement that can take a while to recover from if you don’t get the result (read: success) you desired.

I'm reminded of a phrase: Passion is worth 10 IQ points.

The number of IQ points doesn't really matter but this is about feel. With passion you're much more likely to dig in. By digging in you're more likely to see subtle issues that can result in drastically different outcomes (the more complex something is, the more likely such issues exist). You care about the thing working and so you care about finding out when it doesn't work.

On the other hand, if you have no passion you just go through the motions. You spend less time thinking. It passes the tests? Okay great, let's move on, "it works, so who cares?" In this situation you care less about the thing working and more about getting the task done.

I feel like the second attitude is becoming much more common. I'm sure there are a ton of reasons why but I feel like one of these is that complexity has just exploded. An unfortunate fact is that you can make things too simple. Little errors compound to become big errors that are difficult to wrangle. I think we've gotten to a point where there's so much (often hidden) complexity that we are constantly being overwhelmed, making it harder to care, creating a dangerous feedback loop.

Every good problem solver knows that the best way to tackle a problem is to break it down into bite sized and simpler pieces. But the flip side of this is that every big problem is caused by the accumulation of many little problems. For some reason we have a much harder time thinking in this direction. For this reason I think we need to stress the importance of the little things[0]. It is also important to remember that when solving the big problem that solving each little problem is not enough. That only works if they are independent. You may want to start out treating them as such but that's why this tends to become an iterative process, because as you converge to solving the larger problem these hidden complexities start to reveal themselves. So solving small problems is a defensive strategy.

[0] This can easily be misread. I am not insisting that everyone be a perfectionist. What I've said is far easier said than done. Perfection does not exist, there is always something wrong. The question is much more about bounding that error and keeping it small. It is about recognizing these issues and keeping track of them. More important than solving problems is the recognition of them. After all, it is incredibly difficult to solve problems you don't know exist. By keeping track of these things you can better triage tasks. Even a few comments in the code stating what assumptions are made or stating the conditions that the code is expected to run on will save you tons of headaches in the future. A trivial amount of work in the moment can pay enormous dividends when given enough time.

There's a related problem I see in academic review, but I think it applies far more broadly. The easy part of review is recognizing flaws. One should always acknowledge the flaws, but the authors tend to already be aware of them[0]. The difficult part is determining if the flaws undermine the research or if despite the limitations that the work pushes progress forward. (All but a few works are incremental)

I think this applies much more broadly because even in conversations people are quick to latch onto a subtle inconsequential detail and then dismiss the rest. Being able to read the words does not make one literate, it is the interpretation of them that does. I think this example is quite prolific with internet conversations, enough that we can circle back to sarreph's mention of this in their first sentence. But I think another great example was from this post from a week ago[1]. Most comments are responding to the headline, but many even looked at the post and missed the entire point (which isn't about work being interrupted).

[0] Authors may not acknowledge them in the work because the review process is too adversarial and such acknowledgements can be used as ammunition against them (thanks lazy reviewers ;), because solving those flaws is a good followup and they don't want to get scooped, or many other reasons.

[1] https://news.ycombinator.com/item?id=44999373

you're not my supervisor

Reviewing is also becoming an utter mess in hyped areas such as AI. It's unpaid, thankless work, where most of what you must evaluate is increasingly some gamed, partially AI-generated paper with strategically hidden aspects, rampant cherry picking, etc. Conference sizes are exploding. AAAI went from 15k submissions last year to 30k submissions, with 75k unique authors. There are close to 30k reviewers. It's becoming an assembly line factory with people not knowing each other any more (even in an anonymous review process, if you value the community, you behave differently). It's similar to what old professors lament when referring back to university 50 years ago. When groups were so small, profs knew individually the students. Then came the sausage factory model with 10x the students, students becoming just an ID number, reduction of oral exams in favor or written, gaming the rules more, increased cheating etc.

The academic conference situation will implode under its own weight. As more and more people enter, the quality of reviews suffers, there is less and less illusions about any form of duty, responsibility, honor etc. and it's all a game of turf wars, citations rings, just slipping through the cracks by submitting enormous numbers of papers and resubmitting in a few months those that get rejected, without even fixing the typos that the first reviewers pointed out. And observing this just makes the few who did care also become jaded and put in less energy. Just as you have people in college who want to put in the absolute minimum work in order to get the piece of paper, you have the same happening in academia. Pump out the most papers with the least works, and the reviewers are just an obstacle in that view.

Tell me about it... I almost didn't graduate because of this stupidity... I can say that the review process wasn't great when I started but there was a quick decline after GPT 3 came out. Everything became so short sighted and I've (and many others) even been screaming about this wall we were going to hit for years. We absolutely could have avoided this slowdown if we just understood you can put your eggs in more than one basket and stopped saying "fuck theorists". In betting on "scale is all you need" they destroyed the upstream research pipeline that got us to where we are today... And then we can't figure out that there'll be an exponential increase in papers if we keep adding more people, publish 5 papers a year, and reject 80% of them. How do you expect this to be sustainable?

I think at a minimum academia needs to start tracking abuse. Because right now, when we find plagiarism we hide it, usually just disappearing during blind review. These people should get a one year ban on first offense and a 5 year ban for every subsequent occurrence. Make it a citation network, and everyone you've co-authored with in the last year gets flagged for extra review. And FFS, let's not give best fucking paper of the year at the number one ML conference to a fucking intern who hacked the company you interned at. Not only is that just encouraging more highly unethical behavior, but how the fuck can you even trust the results of that paper? I love this field, but how did we get infiltrated by a cult?