Happiness and Ambition
Part 1
D-O-T, the money power respect
The last one's better
Say, it's a lot of goofies with a check
I really thought I would write more this year. A small army of drafted blog posts exists in this folder, and I have four different ideas for technical blog posts needing only to be watered with time.
I did write, early in the year. Wrote my first paper, wrote a lot of thoughts about how I wanted to change my relationship to technology as I thought through it, and wrote (and thought) a lot about my future. It was highly productive, and then winter gave way to spring.
What happened?
Friends and family visited, and I visited friends and family. I was once again part of two DND groups after a hiatus. The weather in Europe was nice, and in Boston too, surprisingly. I caught up with friends, read the excellent Dandelion Dynasty series, as well as many books that were good but not quite as good. I played some video games. I also lost my headphones, and tried to find how long I could go without them (Answer: three weeks and counting) . I went to parties, and went on dates, and while none of them were life changing they were all memorable.
In short, I lived the good life. I was happy, happier than I was last year as wonderful as it was. I lacked the looming pressure of a research project that has gone on too long. During this period I was (and still am) evangelizing my life. The trick is working a little bit less. Prioritizing things outside of work. And that is true.
But there is a balance in all things, and my birthday looms. 24 years. Two dozen. A year of great numerological significance. The karats in gold and hours in the day. The clock is ticking, and the first quarter of my life is almost up. Reader, I tell you this truly. This notion no longer fills me with fear as it did when I was a youth. I look upon my days and see the default they would be and say:
"Not too bad."
But reader, I do not want to be a creature of just sedentary joy, smiling as I float down a lazy river of a life. A fatty dish is better with acid and spice. And any meal is better with interesting company. I can and will work on things that make me happy, spend time with those I care about, and be kind to others. More importantly, I can be content with that. But I can also try for more, to work on that which I think is important, strive to meet interesting people and make more friends, and help the world in more ambitious ways. Happiness can be achieved in myriad forms, at every hour in every place by every person. And happiness is enough. But I don't need to stop at enough. Don't stop at satiety. Let id and thymos run on reason's leash.
Here's what I believe: I should not struggle to be happy, but I should struggle in life. Not for status or money, but for concrete work I think is important, whether that be research or writing or volunteering. I shouldn't be afraid of suffering for those goals, and should in fact expect it. If I'm able to too consistently go to every thing I'm invited to, I either need to make more friends in my area or work more. If every day the weather is nice outside I can take a break and enjoy the weather, I need to have more pressing work. If I'm averaging two books a week read (which I did from February to April) I need to spend less time consuming the written word and more time, well, writing it. The key is that at some point happiness/contentment/fulfillment stop being things you want more of because you are happy/content/fulfilled by what you have. At that point, you might as well work on aspirations.
This is easier said than done.
Part 2: Motivation
Similarly, when you survey people about what motivates them at work, they go “Feeling good about myself! Having freedom, the respect of my coworkers, and opportunities to develop my skills, learn things, and succeed!” When you survey people about what motivates others, they go, “Money and job security!”
There's another bit to this, an element of my life that has existed in the past year and a half that was mostly absent for most of my undergraduate and graduate education. A year and a half ago I wrote about how no matter what you wanted to want it was important to use your conscious mind to drive that focus. Back then I had started to more clearly delineate my intrinsic and extrinsic interests. I had a research direction I actually believed in, hobbies I wanted to work on long term (writing & dance). Covid and a year of truly detesting working on research had defined via negativa what intrinsic motivation was, and so I was much better at recognizing and pursuing it. Over the past year, I was able to refine those goals from star charts to maps.
For much of my adult life I have been searching for golden cities in the distance, hungry for something but unsure where it is or even what. Now while some part of me will always look on that horizon, I have begun panning for copper in the rivers I know. I have started to mine and already struck veins; all that remains is to dig deeper. In the past, especially for work, I often seriously questioned whether (a) the work I did was useful or even usable, and (b) whether the work I did was directionally good, whether that be for my long term ambitions or the enabled technologies. The answer was often no: I was moving the needle in the wrong direction, and I wasn't even doing it well. While I still wonder about these questions, I wonder about them less. This is for a few reasons:
Professional success is no longer motivated by "prestige and status will vaguely solve all your problems," but rather "there is x.y.z. things you may want to do in the future, that will be easier by having these types of success/money/power/respect."
From greater experience I'm better able to reason about the impact of academic research in my field. Usually minimal unless its very much not. This has allowed me to worry less about the impacts of what I work on and more about how what I learn/do can better prepare me for what I want to do/learn in the future. Furthermore, I can focus on areas I do think will have a positive impact.
I have a few specific directions I think are important/compelling/ripe for academia, and the machine learning, systems, and economic chops to reason about them. Architectures optimized for inference at smaller scales (especially at the edge)1 , ,and open technical policy questions2 are all underexplored and important despite half the planet working on NLP3.
I better know my priorities in life. Social dance and social gaming (whether tabletop or video form) are great joys. They will be part of my life, but they must be content to play second fiddle. They also have a bad tendency to ruin my sleep schedule, which makes my greater priorities (Writing, research, in person time with friends) harder.
All this being said, what I also have learned over this past year and a half is that intrinsic motivation is not enough. Vaguely trying to direct your conscious mind to focus on the things you think are important is not enough. When you are following the compass of your own motivation, there is a huge gap between what you will naturally do (with the sweet sweet temptation of {TV/books/tiktok/games/fomo}) and what you could just far more productively do while being just as happy with life4. Furthermore, there's a huge gap between what feels productive and what is productive. Cycling between burnout and laziness might feel productive, as does constantly clocking into work to stare as loss graphs curve downward, but neither of these things is actually the job getting done. Intrinsic motivation allows you to work much harder before burnout. See illustration:
For long term productivity, extrinsic motivation follows intrinsic motivation. It lets you do what your heart and head want to do, not what your brain stem wants you to do. Otherwise, life quality plummets for nothing. Without a sense of desire, of taste, of goodness, how can you make anything good?
Part 3 - Why?
Ah, but a man's reach should exceed his grasp, Or what's a heaven for?
- Robert Browning, in Andrea del Sarto
Some of you will not understand this purpose. You will believe something along the lines of, "ah, but you would be less happy if you never got status from your research success, if you never sated your ambition."
Ambition does not have to be for status or long term happiness. It can be for its own sake, or for truth or beauty or the betterment of humankind, and when it succeeds in those realms it is incomparably better than ambition for the daily goals we humans naturally strive for. I could say that I could be perfectly happy as a stay at home dad, but you wouldn't believe me.
It is a choice to be true to one's aspiration. It is not a choice everyone gets. It is not always the right one. Notice I didn't say your life's purpose, or your "true" aspiration, or anything like that. The point is not to find meaning but make it, to make it after you have achieved all that you need for yourself outside of it. I don't believe in a true aspiration or divine purpose. I believe in achieving fulfillment and creating meaning with one's own hands. To choose to aspire is not for everyone, and it shouldn't be. Strivers are terrible at just keeping society running; we seek novelty, to improve things, to change people. To embrace aspiration and ambition means not embracing constancy and repetition. If you are reading this, you are probably a striver by nature. We often lack constancy and repitition in our lives, and crave it. But recognize when is enough is enough, and you are able to get back on the saddle.
In short, balance happiness and ambition when you have enough of the former. Use intrinsic motivation as a guiding compass, and extrinsic motivation as a motor to grasp beyond your reach. Do it for truth, or beauty, or for no other reason than that you choose to do so.
A large factor in this is that pretty much everyone who knows how to pretrain models well is gobbled up by big industry labs or lacks the GPUs to train models for academic research. I have the fortune of being able to avoid both of those problems due to intellectual snobbery and a collaboration with IBM with generous terms. Side note: its an extremely good exercise to try to write something like nanoGPT fully on your own in pure pytorch, and see how far you can push up performance. For those of you who want a challenge, repeat the above with distributed training, tensor parallelism, and custom layertypes and dataset.
Shocking how little discussion in my circles I have seen over how much scaling laws have changed in the past two years due to algorithmic improvements, or how one should measure this. This might be the most important question in our field right now? Hoping this is “I’m wrong on the internet question and get corrected”. Low key the best tackle at this was cramming
I intellectually understand that the vast majority of people are working on niche long tail problems not meant for me. I still hate scrolling through 500 papers a week on arxiv
Part of this is defining happiness not as momentary joy but as that longer and slower thing which is fulfillment/contentment. Joy is an important part, inseparable from the greater whole, don’t get me wrong: a life without joy is a painting done entirely in pencil. But once you are fulfilled in life, let joy come when it comes. Don’t chase for it every moment.