jebarker

> LLM coding is like reading a math textbook without trying to solve any of the problems.

Most math textbooks provide the solutions too. So you could choose to just read those and move on and you’d have achieved much less. The same is true with coding. Just because LLMs are available doesn’t mean you have to use them for all coding, especially when the goal is to learn foundational knowledge. I still believe there’s a need for humans to learn much of the same foundational knowledge as before LLMs otherwise we’ll end up with a world of technology that is totally inscrutable. Those who choose to just vibe code everything will make themselves irrelevant quickly.

Based on the post I was under the impression they’re a current CS student. So this is probably a part time job whilst studying

Highly recommend the book "How to tame a fox" about the Russian fur farm fox domestication experiment

I guess the counter-argument is that the world is slowly going to shit and it's only by paying attention and getting involved that we can prevent that. But I don't think that justifies daily news reading.

Is it the Napeoleon Technique or Ostrich Effect to decide not to read the news? I’ve been doing this for the past ~1 year and I’m much happier for it. I’m torn whether I’m just being a bad citizen/selfish or genuinely just not paying attention to things that I have no influence over.

> if I don't see something, I forget it exists.

To some extent this seems like a feature, not a bug. I have many tabs open that I find myself resistant to closing but I'm pretty sure if they just got closed and I forgot they existed my life would be no worse.

I feel similarly. I am happiest and healthiest all round when I focus on the one thing I have chosen to do at any given time rather than figuring out ways to multi-task.

I do however enjoy choosing to do math/coding adjacent activities for leisure or learning sometimes when I'm away from the computer. I've found that it was a net positive in my life to add in puzzles/exercises that I can do with pen and paper in those circumstances.

The issue I find with Chinese cooking is that it often requires very high heat and so (probably due to bad technique on my part) I get a house full of smoke or at least very high VOCs. Any tips to avoid that?

Mike Acton talks about deliberate practice in programming exactly this way. Every day start with a blank sheet and try to build something for an hour (his example is Astroids). Next day, start again and get a little further. Eventually you'll be able to build the whole thing in an hour.

Analogies all the way down.

Multiplying two 100 digit numbers requires the application of a fixed algorithm which can be learned. If you don't know the algorithm it's challenging, if you do it's not.

But that is not true of painting. Painting requires choosing a subject (for its subjective qualities) and then translating what you _want_ to capture about that subject and how you want to represent it in paint on some medium. You will also be applying a theory of mind and perception about the audience of the painting since you probably want it to appeal to them. All of these choices and the skill to combine them into a painting that achieves what you want is vastly more challenging than multiplication.

Multiplication is akin to paint by numbers.

EDIT: it actually strikes me that this conversation gets to the crux of why AI art is so polarizing. It depends whether you view art predominantly as being about the thing that is created or the process of creation.

How does the existence of AI make watercolor painting less of a challenge for a human?

The 3500+ book and 50 year reading list of a gentleman that died in 2025

My top reads this year:

- How to Tame a Fox - great popsci history of a genetics experiment

- "The Sixth Extinction" and "Not the end of the world" - compelling but contrasting takes on climate change

- Through Two Doors at Once - posci history of the double slit experiment

- Alvarez: Adventures of a Physicist - Luis Alvarez should get just as much attention as Feynman does IMO!

- A Matter of Death and Life - the last book I read this year that was touching and made me remember what’s really important in life

I compiled the code with minor changes on my M3 Macbook Air and it found the 19186 unique solutions in about 3hrs 15min, so ~9.2X speedup (over the 30hrs stated in the post). Would indeed be interesting to see how fast you can go with some optimization.

EDIT: I forgot to add -O3 the first time, with that the time to 19186 drops to under and hour, so over 30X speedup.

> At other times in history, leisure was the default state and work existed to enable leisure.

What times/places are you thinking of when you write this?

FWIW I've had a meditation practice for over a decade and do use things like box breathing and journaling too. But sometimes my brain just needs stimulation and this has never gone away. I've learned that the thing I need to avoid is engaging with my monkey mind in spiraling thoughts, so video games are ideal as they require enough attention to prevent me getting distracted but I can also zone out and just play.

I just bought a Switch 2 having not really played games much for decades. I'm finding that occasional breaks for "mindless" gaming noticeably relaxes my always on, work obsessed brain. The challenge is overcoming the feeling that it's an unproductive use of time and I should be reading, coding, exercising etc. But in balance I think it's just the break my mind needs right now.

> Within an hour, Koishi Chan gave an alternate proof deriving the required bound {c(k^2) \geq 1/k} from the original Erdős-Szekeres theorem by a standard “blow-up” argument which we can give here in the Alice-Bob formulation.

Is this an example of the 4 minute mile phenomenon or did the AI proof provide key insights that Chan was able to use in their proof?

This is an interesting paper and I like this kind of mechanistic interpretability work - but I cannot figure out how the paper title "Transformers know more than they can tell" relates to the actual content. In this case what is it that they know and can't tell?

This is how I feel about basically all consumer software

I agree. Wolfram gets so much crap ok this site but I think he’s produced so much interesting science and tech that has built up over decades, that’s very admirable.

I'm not sure how you got that from my comment. CEO is a job title that is easy to get, that was my only point.

Anyone can be a CEO, just start a company.

I haven’t read the paper yet, but I’d imagine the issue is converting the natural language generated by the reasoner into a form where a formal verifier can be applied.

OpenAI has an excellent interactive course on Deep RL: https://spinningup.openai.com/en/latest/

> Note how different, and massively more complex, the spatio-temporal real world of messy analog never-same-twice dynamics is to the 1-D symbolic/discrete world of text that "AI" is currently working on.

I agree that the real world perceived by a human is vastly more complex than a sequence of text tokens. But it’s not obvious to me that it’s actually less full of repeating patterns or that learning to recognize and interpolate those patterns (like an LLM does) is insufficient for impressive generalization. I think it’s too hard to reason about this stuff when the representations in LLMs and the brain are so high-dimensional.

> what evolution has given us is a learning architecture and learning algorithms that generalize well from extremely few samples.

This sounds magical though. My bet is that either the samples aren’t as few as they appear because humans actually operate in a constrained world where they see the same patterns repeat very many times if you use the correct similarity measures. Or, the learning that the brain does during human lifetime is really just a fine-tuning on top of accumulated evolutionary learning encoded in the structure of the brain.

Yes, preserving correctness seems like a good metric. My immediate reaction was to think that the parent comment was saying they’d like to see this comparison because AI will come out ahead. On this metric and based on current AI coding it’s hard to see that being the case or even possible to verify.

As someone that has only dabbled in C++ over the past 10 years or so, it feels like each new release has this messaging of “you have to think of it as a totally new language”. It makes C++ very unapproachable.