Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

We mourn our craft

nolanlawson.com

Post reply on HN
Print view View original

551–560 of 918 posts

Re: We mourn our craft

#551

Post by zzo38computer »

aavci wrote: post #31

LLMs have made a lot of coding challenges less painful: Navigating terrible documentation, copilot detecting typos, setting up boilerplate frontend components, high effort but technically unchallenging code completions. Whenever I attempted LLMs for tools I’m not familiar with I found it to be useful with setting things up but felt like I had to do good old learning the tool and applying developer knowledge to it. I…

Even if it can sometimes help, I think it is not an excuse for writing bad documentation.

Re: We mourn our craft

#552

Post by seanmcdirmid »

davidw wrote: post #543
seanmcdirmid wrote: post #412

Earlier quoted context omitted.

Use local OSS models then? They aren’t as good and you need beefy hardware (either Apple silicon or nvidia GPUs). But they are totally workable, and you avoid your dislikes directly.

"Not as good and costs a lot in hardware" still sounds like I'm at a disadvantage.

$3000 is not that much for hardware (like a refurbished MBP Max with decent amount of RAM), and you'd be surprised how much more useful a thing that is slightly worse than the expensive thing is when you don't have anxiety about token usage.

Re: We mourn our craft

#553

Post by godelski »

stoneforger wrote: post #527
godelski wrote: post #429

Earlier quoted context omitted.

> The answer they always give is that compilers are deterministic and therefore trustworthy in ways that LLMs are not. I don't see this as a frequent answer tbh, but I do frequently see claims that this is the critique. I wrote much more here[0] and honestly I'm on the side of Dijkstra, and it doesn't matter if the LLM is deterministic or probabilistic It may be illuminating to try to imagine what would have happened…

Nah bro I'll just ask the LLM to do better next time /s

It amazes me people say that serially, given in the next breath they'll complain about how their manager doesn't know shit and is leading blind. Or complain about how you don't understand. Maybe we need to install more mirrors

Re: We mourn our craft

#554

Post by stack_framer »

charcircuit wrote: post #445
stack_framer wrote: post #400

Earlier quoted context omitted.

> your world view has fixed into a certain way the world works Yeah, it's weird. I'm fixated on not having bugs in my code. :)

AI can help with that too by automatically fixing bugs.

I find it such a strange cycle to tell AI to write some code, then tell it to fix the bugs in that code. Why didn't the AI just not include those bugs the first time it wrote the code?!

Re: We mourn our craft

#555

Post by bhadass »

hnlmorg wrote: post #484
davidhyde wrote: post #480

Earlier quoted context omitted.

My take on the difference between now and then is “effort”. All those things mentioned above are now effortless but the door to “effort” remains open as it always has been. Take the first point for example. Those little black boxes of AI can be significantly demystified by, for example, watching a bunch of videos ( https://karpathy.ai/zero-to-hero.html ) and spending at least 40 hours of hard cognitive effort learnin…

I’ve worked in the AI space and I understand how LLMs work as a principle. But we don’t know the magic contained within a model after it’s been trained. We understand how to design a model, and how models work at a theoretical level. But we cannot know how well it will be at inference until we test it. So much of AI research is just trial and error with different dials repeated tweaked until we get something desirabl…

but isn't this like a lot of other CS-related "gradient descent"?

when someone invents a new scheduling algorithm or a new concurrent data structure, it's usually based on hunches and empirical results (benchmarks) too. nobody sits down and mathematically proves their new linux scheduler is optimal before shipping it. they test it against representative workloads and see if there is uplift.

we understand transformer architectures at the same theoretical level we understand most complex systems. we know the principles, we have solid intuitions about why certain things work, but the emergent behavior of any sufficiently complex system isn't fully predictable from first principles.

that's true of operating systems, distributed databases, and most software above a certain complexity threshold.

Re: We mourn our craft

#558

Post by MajesticHobo2 »

blibble wrote: post #534
hintymad wrote: post #516

Earlier quoted context omitted.

One thing that I realized was that a lot of our so-called "craft" is converged "know-how". Take the recent news that Anthropic used Claude Code to write a C compiler for example, writing compiler is hard (and fun) for us humans because we indeed need to spend years understanding deeply the compiler theory and learning every minute detail of implementation. That kind of learning is not easily transferrable. Most stude…

writing a C compiler is a 1st year undergrad project C was explicitly designed to make it simple to write a compiler

Third or fourth, maybe, not first.

Re: We mourn our craft

#559

Post by trente »

throwup238 wrote: post #440
orwin wrote: post #417

Earlier quoted context omitted.

But it write mostly useless documentation Which take time to read and decipher. And worse, if you are using it for public documentation, sometimes it hallucinate endpoints (i don't want to say too much here, but it happened recently to a quite used B2B SaaS).

Loop it. Use another agent (from a different company helps) to review the code and documentation and call out any inconsistencies. I run a bunch of jobs weekly to review docs for inconsistencies and write a plan to fix. It still needs humans in the loop if the agents don’t converge after a few turns, but it’s largely automatic (I baby sat it for a few months validating each change).

The problem with looping is that any hallucination or incorrect assumption in an early loop becomes an amplifying garbage-in-garbage-out problem.

Re: We mourn our craft

#560

Post by anon_hack33rr »

ozozozd wrote: post #541
philipwhiuk wrote: post #424

Earlier quoted context omitted.

> the smartest minds in the world Dunning–Kruger is everywhere in the AI grift. People who don't know a field trying to deploy some AI bot that solves the easy 10% of the problem so it looks good on the surface and assumes that just throwing money (which mostly just buys hardware) will solve it. They aren't "the smartest minds in the world". They are slick salesmen.

The other day someone referred to Claude Code as “the most complex terminal app” they’ve seen. Meanwhile folks are rendering videos in the terminal.

And if you didn't know, Claude Code is actually based on React and they are struggling to keep it at 60 FPS, whatever that means in the context of a terminal app

They are writing markup to render monospaced characters in a terminal lol

Post reply on HN