🔎 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
Woe, I am still alive. Had the project and the final (80+ questions) due same day. The only other group doing the same thing I did was the grad students. Professor said my implementation of such method was pretty cool, he didn't see the grad students doing that..Now I am okay and the suffering was worth it. I get soo competitive. When I learned the grads were doing this, I knew I had to make sure I did it better (as best I could).
Sometimes I envy people who don't care - I surely can't do that. So agitated sometimes I am, that I can't work properly.
It's as if some people, with the whole of their being, represent all the nasty things I hate about the industry. The antagonism I feel towards them is almost genetic. If we were to meet a couple of thousand years ago, we'd poke each other with spears.
But I got to admit, though, I am just a different side of the same coin - equally autistic, retarded, and narcissistic. Or maybe even more, who knows?
They beat their meat to the sniny new things, they use AI to write their stuff, and I, on the other hand, do just the opposite.
It was so pleasant to work with the old experienced farts. No arrogance, no reliance on AI, no desire to rewrite the project in a thousand microservices with the different language each, no adoration for "cool" complex shit.
People used to admire simplicity back then.
I need to meet more students like this. It's tiring being the only autistic retard. The major has been flooded with normies. They are not even using discord anymore! Many try to make class groupchats with "Groupme"-- as if! Go major in Biology if you want to use Groupme and behave in other such ways that indicate good hygiene. Last year I made a joke about us all being nerds and some (suspected) tranny classmate sneered at me and said "Speak for yourself, I'm here for the 6 figures."
Anywho, your first sentence is funny because I swear I've said the same thing to my professors. You know what the king said: "An idiot admires complexity, a genius admires simplicity."
I'm also into astrology and spent an hour doing time rectification to try and acquire his birth time. If I did everything right then Terry Davis was a Sagittarius Sun, Pisces Moon, Scorpio Rising. I am also a Pisces Moon so I was excited to read that.
You have to appeal to the public to some extent. I don't really like wondering which one of my classmates is FemboyFaggotBread, but I also don't like how limited Groupme is. Don't think I'll join any after this semester anyways.
No one should make the mistake of going into software development with code quality as a priority. Software development is what you do to pay the bills. Management doesn't care about quality, leadership doesn't care about quality and with a very high likelihood your peers are snot-gobbling retards who barely know how to write a for-loop. And should you be self-destructive enough to push for quality, you'll soon discover that any improvement is swallowed up and digested into the blob-like body horror that is your repo, where it is shat out and re-digested over and over again, until the status quo of low grade mediocrity returns. A million monkeys typing with good grammar is still a million monkeys bashing at typewriters.
Your side projects are where you practice your code bonsai.
I've seen so much stuff about what LLMs can and can not do. Sometimes I think about posting my opinion but then I decide to wait until ChudBound 2 is released to say: I barely wrote a single fucking line of code with my sloppy human hands, and it cost me $20 a month
"If the capacity is 100k items per day and the input is 110k items per day, how long will the queue be, on average? Can you model that?" Normally, individuals like that would be out of a job unless they're dedicated Jira shuffles (which that guy was), but with AI you'll be stuck in a maze of some perfectly self-confident retard's AI hallucinations, and when it almost looks that you've arrived at a coherent vision, there'll be more maze.
Or, you're a programmer who understands systems but you're not a sperg who treats writing code and syntax as calligraphy, and you go "hey write me a queue" and "actually make sure this isn't a bitch to grow and persist."
Then you go yap at biz about the SLA so you know if it needs to be persisted and to what degree etc.
IDGAF about writing a single fucking loop unless I have to. "Is it correct? Is it fast?"
I think the common thing here is "I've done this so much I see patterns and instead of memorizing them and then adjusting syntax by language or library I just go HEY PILE OF VECTORS FILL IN THE GAP!! and tell it what numbers to paint by. Better models are generally pretty good at using the right data structure. The problem is they do shit idiomatic to the language, so if you're doing Java you'll get collections of objects when "just use a fucking array" should be relatively obvious. It will do it if you catch it obviously.
AI is more likely to follow trained patterns of misconception even with the direct context showing proof it agrees with to the contrary. I've helped some sci-fi writer nerds with shit and share scripts or actual programs to show why I think you can model this or that.
Here's a recent example that cracked me up:
You know about the JWST heat shield right? YES So you know you can just use that as a model of heat reduction if you wanted to hide IR signatures right? YES So do it PROJECT RHO/REDDIT/ISAAC ARTHUR SAY NO STEALTH IN SPACE You literally just told me how many fractions of a watt would be on the far side of JWST style heat shielding SORRY FOR WASTING YOUR TOKENS LET ME DO THAT NOW.
After doing shit like this, it will then go "hey so you might want to know about scattered sunlight from such shielding" (which got interesting and is extremely overlooked by the spergs who still wank to radiators) and then if we had EUV quality mirror polished sunshields to avoid scattering, it'll easily pull out dust-density in the ecliptic and out-of-plane to integrate how much dust you're catching on your sun shields which will eventually give you away with diffuse scattering.
But it was useful because I know what integration is, and even if I forgot how to do a path integral through the solar system's dust bands, I can make tests from primitives, build it out, and verify it, and go from there. Zodiacal light is from dust, false sunrises in deep dark is neat, you should go see one sometime.
AI with science is a fucking riot. "Hey, I'm confidently wrong about shit you just told me, but here's obscure shit you'd never know otherwise." Then you look up deep space radiation risks and WE FLAGGED U ITS DANGEROUS. "I'm literally asking about how to protect astronauts." OH SORRY.
You need to manage it. If you're not someone who likes to manage little minions, you're gonna hate it. If you do, it's cheaper than grad students. For now anyway.
tl;dr I like getting systems made but I hate writing syntax.
I've seen so much stuff about what LLMs can and can not do. Sometimes I think about posting my opinion but then I decide to wait until ChudBound 2 is released to say: I barely wrote a single fucking line of code with my sloppy human hands, and it cost me $20 a month
I've seen so much stuff about what LLMs can and can not do. Sometimes I think about posting my opinion but then I decide to wait until ChudBound 2 is released to say: I barely wrote a single fucking line of code with my sloppy human hands, and it cost me $20 a month
Absolutely. I think my codebase is kinda special because it's entirely self-contained, with few dependencies, all vendored, no remote black box APIs. Everything is immediately accessible and comprehensible to the LLM. The fundamental architecture of the engine was set long before an LLM touched the codebase, and that core remains untouched.
LLMs will not make good architectural decisions if you don't know what you're doing.
Opus 5 seems to forget basic fucking photonics/radar shit
Hey boss radar wavelength doesn't matter for beam size. Wavelength doesn't apply to optics, wow! ARE YOU FUCKING HIGH??! ...yes. Sorry. Let me check the shit you led me by the nose to make and told me to run instead of shooting from the nonexistent hip. Oh, yes, it does matter. Rayleigh is a cool dude. Etendue continues to exist.
This is precisely what I mean about AI being a multiplier. Which I should have said already but did not.
If you're ignorant or confidently, wrong, it is too. If you can catch shit like that, you're someone who can catch shit like that.
Opus 5 is a DENSE motherfucker. How did it backslide from 4.6? I shouldn't need Fable for this. The whole point of "write down basic boilerplate for me" is that I didn't have to fucking type out "wavelength and beam focusing are proportional" and it should not just know it it should know how to look it up.
I might have to make a scientific quickref and throw it in github, holy shit lol.
No one should make the mistake of going into software development with code quality as a priority. Software development is what you do to pay the bills. Management doesn't care about quality, leadership doesn't care about quality and with a very high likelihood your peers are snot-gobbling retards who barely know how to write a for-loop. And should you be self-destructive enough to push for quality, you'll soon discover that any improvement is swallowed up and digested into the blob-like body horror that is your repo, where it is shat out and re-digested over and over again, until the status quo of low grade mediocrity returns. A million monkeys typing with good grammar is still a million monkeys bashing at typewriters.
Your side projects are where you practice your code bonsai.
It can be. But I've had the 'pleasure' of linking against commercial, closed-source C APIs where the official, paywalled developer documentation had fancy presentation (neatly formatted, searchable, etc.) but its content was pretty much entirely like this:
Function get_reference_edge
Signature:
edge_tag get_reference_edge(int index)
Explanation:
Gets the reference edge.
Arguments:
index: The index.
Gee thanks, that contains exactly no information that wasn't already present in the function's signature from the header file.
What does it actually mean? How is "reference edge" defined? What are the allowed values for "index"? What are the preconditions for calling this? What is the lifetime of the returned edge (e.g. do I have to call a cleanup function later)? How can my code detect, or guard against, failure states? Pfff, such insignificant details are left to the API consumer to reverse-engineer through trial-and-error.
What does it actually mean? How is "reference edge" defined? What are the allowed values for "index"? What are the preconditions for calling this? What is the lifetime of the returned edge (e.g. do I have to call a cleanup function later)? How can my code detect, or guard against, failure states? Pfff, such insignificant details are left to the API consumer to reverse-engineer through trial-and-error.
Losing good technical writers is a big problem that is probably one of the big things that AI has filled in, strangely enough. Except as you can see above, AI fucks up BASIC shit all the time, and a tech writer would remember shit like refraction.
Writing syntactically correct implementations is a distinct skill from knowing how to architect things, be it architecture for human developers or architecture so the code is fast and correct (when to use a hash vs tree vs kind-of-list vs blahdeblahhhhhhhhhhhhhhhh).
Some people really like to hand write I guess. AI is taking a lot from them work wise, which is unfortunate. If they just like it though I mean go do you bro.
<parameters/return value>
<what each parameter is for and why>
<what is the API intended for>
<usage examples>
<notes for what to watch out for>
<list of potential errors>
<related API's>
ideally it also includes a big picture overview of what it does internally
Whenever I walk into a new codebase, I used to look at a large list of libraries/package dependencies and think: Wow, all of these must be super important, and each library is probably written by super smart professionals. Better not mess with those!
Now I just see a list of garbage written by drooling retards that I want to delete quickly and aggressively.
Whenever I walk into a new codebase, I used to look at a large list of libraries/package dependencies and think: Wow, all of these must be super important, and each library is probably written by super smart professionals. Better not mess with those!
Now I just see a list of garbage written by drooling retards that I want to delete quickly and aggressively.
Whenever I walk into a new codebase, I used to look at a large list of libraries/package dependencies and think: Wow, all of these must be super important, and each library is probably written by super smart professionals. Better not mess with those!
Now I just see a list of garbage written by drooling retards that I want to delete quickly and aggressively.
Some people really like to hand write I guess. AI is taking a lot from them work wise, which is unfortunate. If they just like it though I mean go do you bro.
Writing code was never the hard problem of software, which is why I think using LLMs is misguided. If I understand the problem and understand the solution and know what kind of architecture I want to go with, then 99% of issues have been solved. Writing the code is just putting it all together.
I would also argue that not writing your own code deprives you of a discovery process. While writing, you might spot mistakes in your own assumptions, or might spot mistakes elsewhere in code. I've started calling this AI myopia. You only see what the LLM sees. But I guess this doesn't matter if your project is purely LLM-generated and you don't have considerations like efficiency, timing, etc.
Ultimately I blame the whole "learn to code" bullshit. It filled the profession with people who had no business being in the profession, either due to wrong incentives ($) or incompetence. This influx only hindered development, as they required constant handholding and oversight. Now, instead of fixing this by getting rid of useless developers and optimizing ways of working, we are bringing in LLMs to enable the incompetent developers to do even more damage.
And to close this diatribe, I am convinced LLMs will fail in the long term. The hard technical material required to feed them is already running out online and replaced by LLM-generated slop. Even technical books are now often LLM-generated. This will lead to LLMs providing stale, out-dated output, or just experiencing model collapse.
. It's a hard lesson to learn.
You know about the JWST heat shield right?
YES
ARE YOU FUCKING HIGH??!
