How I got GPT-2 small to run in a stylesheet (and a whole lot of fonts), with zero JavaScript.

Implementing GPT-2 Entirely In CSS

Thoughts on how to proceed in times of chaos

Nobody Knows What's Happening

We lose something quite important as we grow up

What is Innocence?

How is the one thing I enjoyed while growing up so unappealing now?

Pure Software Is Agonizingly Boring Now


When we were children, we used to build sandcastles simply for the love of building sand castles

As one grows up, they inevitably learn that *Hey, doing [XYZ thing] can make me rich/famous/popular/well-liked!* The activity no longer becomes the point, it's rather a means to an end.

I'd classify this as the true loss of innocence, the adulteration of our natural wish to act in the world with higher "stakes." 

You somewhat see this when people caution you not to turn your passion into your job. Once you attach higher stakes (money) to the thing that you simply love doing, you end up a) becoming quite sedentary (as any action either incurs you a gain or a loss) and b) losing the pure joy of the activity that you once had.

Simply due to having so much input data as one grows up (you start seeing the factors that "matter" or are "appealing" in the world), it becomes quite easy to get caught up in these often inconsequential matters. Superficial, is another good word here.

I used to code quite a lot, just for the hell of it. I'd make websites and games just simply because I wanted to, and felt myself naturally gravitating towards those activities. I had no concept of "I want to make this so I look impressive" or "I want to make this so I can become rich." My time simply consisted of me coding because I wanted to code. 

Once you attach higher stakes to an action, moving in any direction becomes quite daunting, since it either correlates with an increase or decrease in the stakes you've attached. 

As children, we weren't quick to attach meaning to our actions. This made us much more free to act, simply because **nothing actually mattered.** Think about it, we didn't register that any outcomes were attached to our actions. I'd make a game simply because I wanted to. I'd root my Android phone simply because I wanted to.

It's hard to describe this motivation to one who is so conditioned to having their actions as fitting into "a bigger system." *You need to do this for money because it means something! You must do this to make your resume stand out!* The action ends up becoming a means to an end. A slave to some bigger goal that you may not even be fully aligned with deep down. Your life ends up becoming a means to an end for that goal. 

Think about it, even social scenarios have become this means to an end. "Your network is your net worth" chanters have made everyone associate the simple (and most innate thing a human can do!) with superficial factors like prestige and money. With this mindset, one becomes less free to socialize, as their every move corresponds to a gain or loss, and the pressure to converse comes from opportunity cost. 

A kid doesn't care whether they're conversing with a billionaire or the average Joe, they'll simply speak their mind as they wish.


> *Nobody knows what's happening.*

The experts will tell you that they know what's going to happen. They don't know what's happening.

The ones making a product will tell you they know what's happening. They, again, have no clue what's happening.

Hell, even the guy down the street will have some elaborate view on how the world will change. He clearly doesn't know what's happening.

The big labs and their leaders will tell you how the world is going to look like in 5 years. They're especially incentivized to paint this future in a way that pads their pockets. Consequently, they also don't have a clue what's going to happen.

What we do know is that now is a period of vast amounts of change. We do know that the next 5 years are going to look completely incomprehensible. Before, it was quite obvious to see that a college degree would yield you certain paths that would be ultimately fruitful. Now, that is not as clear. This is not to say that college degrees will be useful or not useful (again, I also am no closer to understanding what's happening to make this judgement).

Reasoning by analogy here is tricky. 

> _Oh but this guy did this thing 10 years ago! That's why doing this thing is useful now!_

Well, of course, that was 10 years ago. 2016 was a very different time for tech and the world in general compared to 2026.

Anyone with a strong view as to what's going to change is most likely wrong. This is not to discredit the value of having strong views, but simply attaching yourself to one without critically thinking and making a judgement for yourself is a recipe for disaster. Anyone could be wrong, but given that you have the most context on who you are, only you can really trust yourself.

Who knows what a job is going to be in the next 3 years? Is all knowledge work going to be taken over? Once these models are able to learn in unverifiable domain spaces, it might become impossible to prevent them from taking over any of our professions. Is even the oh-so difficult space of robotics going to be solved within a matter of months after that event?

It's all one large soup of chaos, and clinging to old patterns passed on by time (or even by other people preaching their supposed gospel) is risky. Following it might lead you astray. It has always been true that there hasn't been one unified rulebook on how to thrive in the world, and it's especially more apparent in the times of large transformation that we find ourselves in right now, especially with the ever-looming presence of superintelligence around the corner.

Nobody's word is truth, nor does anyone have a clear idea of what the future is going to hold. Leaning into the chaos during times of transition could be the correct move, and grasping for rulesets or any form or order could be the wrong move. I have not a clear picture of the future, nor do I claim to. All I wish is that we truly think for ourselves in such a time of change.


> *Enter, Enter, Enter, Enter*

You get to type a little bit if you're lucky (or hold down `fn` and speak if you're from that crowd). The machine will do all the thinking for you, all the bits you used to enjoy. Just sit back and watch the results stream in.

When did it get so boring? Why do I feel like I'm simply a forklift operator?

I remember the joy of hacking together different APIs to make cool projects, I remember spending hours building UIs from scratch, spending hours adding in the tiny quirks, simply because I enjoyed the process of patching together bits of code (you could say creating? I will get to this in a moment, because I don't think software engineering was about creating).

When did it lose its joy? All I seem to be useful for now is simply planning and guiding these models. I didn't sign up to be a manager, I didn't sign up to simply watch and interpret an employee's (Claude Code) work. I signed up to be in the midst of the action, I signed up to be deeply engrossed in the work ("work" is such a reductive term here) I do.

I don't want to sound like a Luddite, I truly am not. No doubt, coding agents are incredible. They allow me to test a million different things and try out tons of new stuff I hadn't previously thought of before. I use these tools every day, and they are genuinely useful. However, they seem to have taken the fun out of the one real thing that I enjoyed doing while growing up as a teenager. 

## Software was never about "creating" or "building"

What do I mean about software engineering never being about creating in the first place? Well it's quite simple. 

Creation was always the *byproduct* of software engineering, and software engineering being the medium for creation. 

We choose the mediums that we enjoy, an artist chooses their canvas because they feel *at home* playing with brush strokes. A musician chooses to play an instrument because they *feel alive* strumming the chords of their guitar.

You're pulled to certain mediums because they naturally provide an intuitive and delightful way for YOU to express yourself in the world. There are certain qualities that are naturally in-tune with your brain that make it a great fit.

Across all these mediums, the implicit goal was all the same, creation. You would never pick up any of these to _not_ create. It's inscribed in the very nature of what humans do with the tools given to them.

I took to software because some part of it simply made sense to me! I loved it! I loved the way through which I could get my voice and ideas out to the world! The process was enthralling, the way I would patch together tons of different libraries and techniques together to make a cohesive end product was exhilarating to me (no doubt, the way that other mediums provide the same excitement).

## Optimizing for creating

When we assume that our end goal is creation, it obviously makes sense for the intermediate process to be squashed down in terms of time and complexity. But I must ask you, if we were able to boil down the art of music creation to a couple text prompts, where's the involved-ness of that? Where is the beauty in it?

The intuition one gets from throwing themselves in the deep end is something that we're losing. I'm not going to comment on whether these systems can learn that intuition, but there's something terrifying about offloading that intuition away from humans (eventually leading to an atrophy of that muscle).

This is all from an intrinsic perspective, however. Yes, the world needs to keep spinning, money needs to keep being generated, I totally agree with this. I definitely think that my views are restricted to a very certain archetype that enjoys the process, and this definitely doesn't apply to most other people.

Software is now so prevalent within our lives (it's been like this for a while), and is truly a means to an end. It has shifted from being this nerd/hobbyist activity to something that's vital for everyday life. In this scenario, making that software as easy as possible to create is great! There's no longer a barrier to code, no barrier to access this realm that was previously locked to a specific breed of person. The issue now is, nobody is actually coding. The medium has been polluted.

## You're at a loss if you don't comply

Unfortunately, you are it's simply invalid to not adopt newer tools. Thinking you can outpace them or that you don't need them is simply not an option. You will be rendered obsolete if you abstain.

Some may frown at this constant I seemingly just presented out of thin air, but I do think it's one that's worth accepting lest you waste endless time internally conflicted about it to no avail. That's pointless, that debate could go on forever.

## This is a blessing in disguise

I am urging you to move on to harder and better things! This is your sign, this is your opportunity to work on things that are now achievable with the advancements in AI, but still interesting! Humans continuously adapt to new things, don't label your identity to that of a software engineer and you will find yourself amongst the things that do stimulate your soul.

I am currently searching the next mountain that I climb. If you have any suggestions, or any interesting sidequests, do [reach out](mailto:shubhampatilsd@gmail.com).



flying-toasters

Shubham Patil

Writing