Last night, I got a pick-up order of Dan Dan noodles, and some Mapo Fish with Crispy Tofu from my favorite Sichuanese place. Wa Jeal, on the Upper East is really good. My wife and I cozied up on the sofa (or the couch, as the yanks say) to set up for movie night. Tonight, we were watching Alpha Go.
I work in AI. The application layer – so I don’t really build the models. But I use the models, a lot. And for a lot of things. My startup works in supply chain and impact. But, I use AI every day. It’s magical how good it has gotten. This website you are reading this essay on? I designed it using Windsurf. In a few prompts.
The pace at which AI has been improving has been insane. But the pace at which humans ignore this growth has been even more insane to me. This essay is more a social commentary on what I see inside (and outside) my social circles.
@Advait Raykar multiple errors during compiling itself. My prompt might be very simple but that is not an excuse. It can’t even give me the “hello world” eBPF program without feedback
That is the message I received from my friend. Over the nearly 10 years of friendship, I have known him to be an extremely smart guy, who sweats the details and works hard. He is doing his PhD in computer science. He is very familiar with AI, deep learning and has been a researcher in the field, before the hype. Yet, I think he is completely wrong about his AI skepticism.
I have had several hours of debates, where he’d often give up, probably more so because of my stubbornness and walls of text than him agreeing with my lines of reasoning. “AI is spiky”, I kept telling him, “It’s not good at that, just yet”.
Ultimately, all my arguments boil down to the same point. I text:
See, I think you don’t buy Sutton’s Bitter Lesson. I do. I think historically it has proven right in all fields. There is no guarantee it will continue to, but my gut instinct and my intuition on how stuff works makes me believe
My PhD friend, while not (yet) a believer of The Bitter Lesson, is getting close (in no small part due to my constant haranguing)1.
What happened in 1764?
A wheel.
Then two wheels, three, ten.
Then—hundreds, humming.
James Hargreaves’s spinning jenny;
Richard Arkwright’s water frame;
Samuel Crompton’s mule.
Spinners: Quietly, inexorably, displaced—
Hands idle,
Hunger stirring.
For generations, spinners had drawn out thread, one careful strand at a time,
fingers trained by decades of gentle repetition.
Now:
One machine.
Eight spindles.
Sixty threads per spindle.
480 threads at once.
The room that held twenty spinning women yesterday—
today holds
one machine,
one operator,
and silence.
⸻
But where the spinner fell, the weaver rose—
loom after loom after loom after loom
after loom after loom after loom after loom
and then
loom
after
loom.
Weavers, once secondary, now central.
Thread cheap, cloth abundant, profits exponential.
Their quiet labor transformed
to noise,
rhythm,
industry,
power.
Spinners faded,
Weavers ascended,
machines multiplied:
a story told in threads,
rewoven
every
single
day.
In my friend circle, I have faced some heat in the past due to some social faux pas I have made while expressing my beliefs about whom I believe are contemporary weavers. An incident comes to mind – a house party, at a grad school apartment right by Columbia University. I was talking to a solid fellow I met at the party. I may have said, “SWEs are cooked”. I recall it being pertinent to the point I was making about re-inforcement learning. The host, who has, and continues to be wonderful to me, shot a gaze. I could sense that she took offense to that statement – she had just got a job in big tech as a SWE, in a really tough job market, after working extremely hard for her Ivy league CS degree.
Side note: I don’t like it when people call a software dev a “swee”, but unfortunately, a part of my extended social circle uses that term, and I do what’s required of me when in Rome.
Back to the apartment by Columbia. I was making my case for swees being cooked. RL, I proselytized. She says she studied RL. She has taken courses in deep learning. Yet, she doesn’t believe they are cooked. I think it’s best to not die on this hill, for the host has been generous and fun the entire night. I should not be that guy. I chose not to be that guy. Unfortunately, I had already been branded that guy.
I always find it somewhat amusing when someone said they were “haunted” by a question. Haunted? What an odd choice of words. Ghosts haunt you. During your sleep. When you space out. When you have a moment of silence, in your brain. I can’t say that I am yet to have been haunted by a question yet, but this one is pretty close.
Who are the Contemporary Weavers?
When spinning was automated, there were a class of people that got displaced, but that led to a renaissance of weavers. For a short while. A few decades, until the power looms came around. A specialized skill, completely obsolete. And they thought they were special. Just like I (and you) think we are.
My close friend’s wife is an editor. She (used to) edit books, the kind you and I (used to) buy at the bookstore and (used to) read2. Her husband, my close friend, works at a frontier lab. My close friend’s wife’s professional circle doesn’t like my close friends professional circle.
What’s funny in this situation is that I think good writing is immune to Sutton’s Bitter Lesson. What’s the fitness function for good writing? I don’t know.
What is the future or art? I wager, it’s bright. As a creative, I feel a lot more enabled by AI.
You’d definitely not hear me say “Writers are cooked”. Bad writers are. But they have been cooked a while. These writers were the spinners. Not the weavers.
I am not saying AI can’t write better than humans. Maybe it can. But I wager it won’t be giving David Foster Wallace a run for his money, for two reasons:
1) By definition of how pre-trained models (non RL ones) work, they are only as good as their data. The art they produce will be the art that the median person would love. They could write a wonderful Marvel script. But I don’t think they could write Arrival. Creativity lies in the fat tails.
2) There is no money to be made in art. All the money to be made (and laundered) in art is to do with celebrity. “Ooh, is that an original chatty 4.5?” simply doesn’t have the same ring as owning an original Basquiat.
I was having dinner at Planta Queen, a hip private equity owned restaurant in NoMad. Yes, I looked up the private equity stake. No, it doesn’t (or shouldn’t) make a (huge) difference in what I think of the place. The food was (pretty) good. The service was (pretty) mid. But I was there for the company. It was my friend’s birthday. And it was also a mishmash of her various friend circles. I was from her “grad school” friend circle. She lives a few blocks away from us, and my wife and her hang out a lot; they also produce a talk show of the both of them going to cafes and filming themselves. Their release schedule, much like our collective attention spans, is erratic.
I was looking forward to this dinner, because off late I have taken an interest in how people live. And this table had a diversity of people when it came to work. For example, one was an architect. She was building a prison. Everyone glazed over that detail. She films food content on the side.
Despite the diversity, the seating arrangement (unfortunately) sandwiched me between an SDE (the birthday friend) and a product manager (my wife). Facing me were two friends of my birthday friend, an SDE at a bank and an ML Engineer at big(ish) tech company.
They started talking about each others’ work. I learned that they had not used Cursor. They had never heard of Windsurf. They had not used newer models for meaningful work. One of them on the table – I knew the models would be frighteningly good at doing what she does. I wondered how much of their jobs could have been o4-mini tokens? If I were to guess, I would say 20%-40%. If I found it to be 80%, I would be surprised, but only midly so. Thin ice.
I was my friends birthday, I didn’t want to think of tokens.
What Ernest Hemingway Taught Me
Quiet accumulation—
dust settling on mirrors,
rust blooming in shadows,
truth drifting unseen
like smoke under doors.
Nobody notices
the shape
the air takes
until one breath
catches sharply:
the room
is filled
with smoke.
Change always arrives
like this—
gradually,
then suddenly.
Where are my friends and acquaintances?
- ✧ Blissful living, from ignorance
- ✧ Awareness of threat
- ✧ Denial of threat
- ✧ Acceptance of The Bitter Lesson
- ✧ Blissful living, from acceptance
We had cleaned up the Sichuanese food, and then lapped up the AlphaGo documentary. My wife and I loved the Alpha Go documentary (and the food). It was surprisingly moving (the documentary). If you have not seen it, you should. It’s free on YouTube. It’s great story telling, and a very important story to be told.
The documentary follows the DeepMind team as they build AlphaGo, an RL based Go player. And pit it against Lee Sedol, who is described as the “Roger Federer of Go”.
The documentary is less about technology, more about the human aspect of it.
“I don’t think it will be a very close match. I believe it will be 5-0, or maybe 4-1. So the critical point for me will be to not lose one match.
– Lee Sedol, main pre-match press conference
In a way, we are all Lee Sedol. Or start off as Lee Sedol. My PhD friend, is Lee Sedol. The wonderful host of the house party in the apartment by Columbia is Lee Sedol.
One of the key moments in the document is “Move 37”, which takes place when Lee Sedol goes out on a smoke break during game 2, a cinematic gift to the filmmakers. I still don’t know what the move means in Go terms. But it was the move that changed how Lee Sedol thought of AlphaGo.
“Yesterday I was surprised. Today I am speechless.
– Lee Sedol, on the match where Move 37 was played
It was a move that was near unanimously considered “strange”, “and “alien” by expert commentators.
“That’s a very strange move. I thought it was a mistake.”
“incredibly creative and unique – a move no human would have made.”
“It’s not a human move. I’ve never seen a human play this move. So beautiful”
What follows is a study of humanity. We all were, and are, Lee Sedol. It is moving to watch.
The documentary ends with an interview with Lee Sedol.
This move made me think about Go in a new light. What does creativity mean in Go? It was a really meaningful move
….
I have grown through this experience… I feel thankful and feel like I have found the reason I play Go. It has been an unforgettable experience.
I want to be Lee Sedol. The post match Lee Sedol. Not everyone has the luxury of celebrity. Celebrity is the ultimate vaccine against becoming a Contemporary Weaver. But for the rest of us, we need to know where we stand. Are we weavers? Are we going to resort to burning the looms?
“The future is already here — it’s just not very evenly distributed” — William Gibson
LLMs are not a magic bullet, and AGI is not upon us. But Useful AI is upon us. We are getting incredibly good at translating tokens to GDP. And, while OpenAI gets a lot of flack for being ClosedAI, I genuinely believe, for $20/month, they have given me a glimpse into the future.
I hope these threads have left you with a few things to think about. Where do you stand? Are you the Contemporary Weaver?
Being the Weaver may not be a choice, but continuing to be the Weaver, is.
Appendix
-
One of the turning points was him finding out his entire paper, including the novel insight was generated by o1 with one prompt; and the entire paper drafted with deep research, minus the experiment. He even discovered new work he didn’t know about in his draft. ↩
-
I wrote that for effect. I am moving towards 100% physical and long form consumption of information. I have a New Yorker Subscription, and I read it in the subway. ↩