Programming thread

  • 🔎 If you see an alert for analytics.usips.org, this is a self-hosted system I am running as a test. It is GDPR compliant and collects no invasive information.
  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
Peaches the Wonder Cat said:
This overlaps between a few threads. What's the best "bang for the buck" coding LLM these days? My last employer decided that I couldn't redeem my izzat and I just want to keep my skills sharp.
Click to expand...

Using a coding LLM and keeping your skills sharp are opposing concepts. That said Claude Code is the one that people won't shut up about. I've heard that Grok's equivalent is good too. So decide which coke-fueled weirdo you want to give your money to, Altman or Elon.
 
SchizoDaemon said:
I enjoy the problem solving aspect. But professionally I never asked to solve anything more complex than "when I click this button, thing is bringing up this error". It is very boring work e.g. find error in AWS console, work out what line, write a test, PR, review and then deploy. It is fucking tedious.
Click to expand...
This reminds me of the last time I tried to teach someone about HyperCard and Decker. I told him just how trivial most of this was, but he talks to ChatGPT every single day—he's coincidentally the stupidest motherfucker I know—and retorted that the neural network nonsense now handles the same thing. I and a buddy tried to explain to him that not only is it wasteful, it's not removing any complexity, just dumping on more.
Concentrate Juice said:
I have seen the light and abandoned my old, machine-focused ways. Computation is mathematics, the abstraction of abstractions, and the language of mathematics as we know it is symbolic manipulation. Symbolic computation is what you study if you want to actually know about what computation is, and mathematics plugs directly into it because they operate on the same substrate, symbols. The Curry-Howard correspondence brings tears to my eyes. I was a blind man but now I can see. Learn abstract algebra.
Click to expand...
Welcome to the club. Now, I was already like this beforehand, but I've got a book recommendation that reinvigorated this feeling within me: "The Architecture of Symbolic Computers" by Peter Kogge (ISBN 0-07-035596-7). I'm not at the point yet where he starts going into hardware implementations of parallel hardware operating on symbolic data, but already the book has been so pleasant to read. Most programming texts of this millennium focus too much on contemporary programming systems, to their detriment.
 
Least Concern said:
So decide which coke-fueled weirdo you want to give your money to, Altman or Elon.
Click to expand...
Or Nvidia, AMD or Intel.
I went with Intel for now, with a few bucks in credit in Open Router for the really tough cases.
And Pi Agent for my coding harness which happily deals with any back-end models.
 
Peaches the Wonder Cat said:
This overlaps between a few threads. What's the best "bang for the buck" coding LLM these days? My last employer decided that I couldn't redeem my izzat and I just want to keep my skills sharp.
Click to expand...
I've had good luck with qwen3-coder-next on local hardware (/proc/cpuinfo says AMD RYZEN AI MAX+ 395 w/ Radeon 8060S, system has 128GB unified RAM). It's not a "thinking"-style LLM so it doesn't spend ten minutes babbling to itself before it starts actually producing useful work. I haven't bothered with online services. They're all hardwired to lure you in with just enough free "tokens" to get you interested in using them for productive work, then rug-pulling at the moment genuine productivity is about to happen so you'll bribe the pay wall.

Just run shit locally. For qwen3-coder-next in particular, keep it constrained to a single repo (or even just a subtree) for best results.
 
moocow said:
qwen3-coder-next
Click to expand...
I didn't like it much when it was new. I find the thinking helps the Qwen 3.8 models quite a bit, sure they spend a ton of tokens, but they're free. I think I did a bake-off with Qwen 3.6 some Gemma and Qwen3-Coder-Next and as I recall "Coder" lost the coding test. But as that was over a week ago I have no records. The nice thing is they're all free to download so it's easy to try them all.... until you run out of disk space.
 
Peaches the Wonder Cat said:
This overlaps between a few threads. What's the best "bang for the buck" coding LLM these days? My last employer decided that I couldn't redeem my izzat and I just want to keep my skills sharp.
Click to expand...
Your best bang for buck isn't about the single LLM anyone. It's about using the right harness and tools with multiple LLMs.

I use open code which allows me to assign different LLMs to different tasks. For running, and summarizing output of tests and builds I use the cheapest thing I can find. The same goes for exploratory work. It's Luna right now.

For writing code I have something medium like Claude or Terra do it.

When I want vision analysis, and layouts built from them I use Sol. The same goes for project planning. Although, I hear Fable is good as well.

You should be running RTK on anything you use as a harness. It leads to 90% efficiency gains on some commands.

Edit: as for learning, you should be actively looking for things you don't understand or doesn't make sense in your output. Then research it until they do make sense or don't even more due to hallucinations. I didn't know what NEXT_PUBLIC did until today when I asked and it didn't make sense.
 
Last edited:
Least Concern said:
Using a coding LLM and keeping your skills sharp are opposing concepts. That said Claude Code is the one that people won't shut up about. I've heard that Grok's equivalent is good too. So decide which coke-fueled weirdo you want to give your money to, Altman or Elon.
Click to expand...
It's both nowadays. Last job had me using LLMs all the time for doing the grunt coding while I thought of the bigger picture - architecture, making sure the LLM added tests and not credentials into the codebase, stuff like that. I don't see it as much different as shifting to procedural/object-oriented/frameworks as new programming technologies and methodologies developed.
 
Peaches the Wonder Cat said:
It's both nowadays. Last job had me using LLMs all the time for doing the grunt coding while I thought of the bigger picture - architecture, making sure the LLM added tests and not credentials into the codebase, stuff like that. I don't see it as much different as shifting to procedural/object-oriented/frameworks as new programming technologies and methodologies developed.
Click to expand...
Writing code by hand is still an important skill.
 
Before learning how to code, how many of you dealt with superficial shit takes when trying to understand how a computer works, i.e.: "CPU is the brain of the PC"?

Recently, I was trying to learn how logic gates work and how do they make computers do arithmetic operations and I've remembered when I first started to learn about coding and how much I bypassed the fundamentals which pretty much kept me from seeing a computer as a electronic device that does math, stores data and runs algorithms, instead of a cryptic magic box.
 
Celestial Peace said:
Before learning how to code, how many of you dealt with superficial shit takes when trying to understand how a computer works, i.e.: "CPU is the brain of the PC"?
Click to expand...
That's a really good question. I don't remember by now. The earliest programming I got was in a high-level environment that represented programs as very high-level instructions, switches, and things like that.
Celestial Peace said:
Recently, I was trying to learn how logic gates work and how do they make computers do arithmetic operations and I've remembered when I first started to learn about coding and how much I bypassed the fundamentals which pretty much kept me from seeing a computer as a electronic device that does math, stores data and runs algorithms, instead of a cryptic magic box.
Click to expand...
I bought the Time-Life "Understanding Computers" book series at a used books store years ago, and they're really good reads for anyone who can get a hold of them. Here's a photograph, which I really should retake sometime, from "The Chipmakers" showing how an adder circuit can work:
chipmakers.jpeg
Here are some of the ISBN codes:
  • 0-8094-5679-6
  • 0-8094-5683-4
  • 0-8094-5666-4
  • 0-8094-7618-5
Transistors can be thought of as little switches, or like parts of plumbing, and logic gates are built from them and the wires connecting them. I don't know, it seems really simple when presented in this manner, doesn't it?
 
Celestial Peace said:
Before learning how to code, how many of you dealt with superficial shit takes when trying to understand how a computer works, i.e.: "CPU is the brain of the PC"?

Recently, I was trying to learn how logic gates work and how do they make computers do arithmetic operations and I've remembered when I first started to learn about coding and how much I bypassed the fundamentals which pretty much kept me from seeing a computer as a electronic device that does math, stores data and runs algorithms, instead of a cryptic magic box.
Click to expand...
https://nandgame.com/ and https://store.steampowered.com/app/1444480/Turing_Complete/ are both great
 
Celestial Peace said:
Before learning how to code, how many of you dealt with superficial shit takes when trying to understand how a computer works, i.e.: "CPU is the brain of the PC"?
Click to expand...
The 8-bit microcomputer I had handed down to me had a set of manuals. There was no way to know how to use the computer really unless you had read them. This included a DOS manual, a basic manual and a reference and a bunch of other stuff I have now forgotten. In this they talk about the basics of how the computer operated, e.g., RAM, ROM, CPU, video, etc.

I've unfortunately lost a book in a move that was a very nice book with code listings that would work on various computers at the time. It also had illustrations and explanations of how the computers work; it also had explanations of how Logo Turtles work (both on screen and the motorised ones with the pen).

I was lucky we had a teacher (a woman) that managed to explain how instructions on a computer worked. She did something similar to the infamous peanut butter sandwich video, which I think is some of the best and most wholesome content on YouTube.

 
Last edited:
Celestial Peace said:
Recently, I was trying to learn how logic gates work and how do they make computers do arithmetic operations and I've remembered when I first started to learn about coding and how much I bypassed the fundamentals which pretty much kept me from seeing a computer as a electronic device that does math, stores data and runs algorithms, instead of a cryptic magic box.
Click to expand...
Moving beyond logic gates, Mark Smotherman's page on the history of microprogramming provides an excellent explanation of how sequential processors work and has both hardwired and microcoded implementations of a simple 4-instruction machine. (Of course, since he's a hardware guy, he didn't make the instruction set Turing-complete, when it's been known for decades that "store", "subtract", and "branch if negative" are sufficient for a one-address machine.)
Technically, you just need to be able to subtract and provide a single branch-on-sign instruction, since you can get addition from subtraction (x + y = x - (0 - y)) but not vice versa and you can get branch on (in)equality from branch on signum of difference (e.g., x ≠ 0 ≡ x < 0 ∨ x > 0). This means you can get by with any element of {("subtract", "store"), ("add", "store negative")} × {"branch if negative", "branch if not positive", "branch if positive", "branch if not negative"}.
 
Last edited:
Celestial Peace said:
Before learning how to code, how many of you dealt with superficial shit takes when trying to understand how a computer works, i.e.: "CPU is the brain of the PC"?

Recently, I was trying to learn how logic gates work and how do they make computers do arithmetic operations and I've remembered when I first started to learn about coding and how much I bypassed the fundamentals which pretty much kept me from seeing a computer as a electronic device that does math, stores data and runs algorithms, instead of a cryptic magic box.
Click to expand...
I had some basic understanding of programming first, starting with Ruby in RPG Maker in elementary school, and C++ later as it was True&Honest language for game dev.

Afterwards I got electronics technician degree in Polish "technikum", which is like vocational school combined with high school.
There was theory for analogue/digital circuits and basic computer architectures taught there.
I pivoted back to programming because it pays better.

I don't think it's really necessary for most to learn electronics. But I think it can be interesting.
Ben Eater has nice YouTube series going from logic gates to full 8 bit breadboard computer, going through latches, flip-flops, ALUs etc.
Ben Eater on YouTube
 
Unimaginative Username said:
Moving beyond logic gates, Mark Smotherman's page on the history of microprogramming provides an excellent explanation of how sequential processors work and has both hardwired and microcoded implementations of a simple 4-instruction machine. (Of course, since he's a hardware guy, he didn't make the instruction set Turing-complete, when it's been known for decades that "store", "subtract", and "branch if negative" are sufficient for a one-address machine.)
Technically, you just need to be able to subtract and provide a single branch-on-sign instruction, since you can get addition from subtraction (x + y = x - (0 - y)) but not vice versa and you can get branch on (in)equality from branch on signum of difference (e.g., x ≠ 0 ≡ x < 0 ∨ x > 0). This means you can get by with either any element of {("subtract", "store"), ("add", "store negative")} × {"branch if negative", "branch if not positive", "branch if positive", "branch if not negative"}.
Click to expand...
technically one instruction is enough if you pack subtraction and branching on sign https://en.wikipedia.org/wiki/One-instruction_set_computer
 
Magazines taught me a lot. I don't remember how much the ZX-81 User Guide was helpful, or the Commodore 64 one. But there was a much larger C64 book, likely the Commodore 64 Programmers Reference Guide which I literally read the covers off of. Then Machine Language for Beginners (COMPUTE! 1984) I honestly don't remember when I started to understand gates, maybe my Radio Shack 200-in-1(I think) which had a Flip-Flop and a NAND IC you could use.
Let me re-adjust the onion on my belt.
 
You must log in or register to reply here.
Back
Top Bottom