It has to be pure ignorance.
I only have used my works stupid llm tool a few times (hey, I have to give it a chance and actually try it before I form opinions)
Holy shit it’s bad. Every single time I use it I waste hours. Even simple tasks, it gets details wrong. I correct it constantly. Then I come back a couple months later, open the same module to do the same task, it gets it wrong again.
These aren’t even tools. They’re just shit. An idiot intern is better.
Its so angering people think this trash is good. Get ready for a lot of buildings and bridges to collapse because of young engineers trusting a slop machine to be accurate on details. We will look back on this as the worst era in computing.
Generally I equate positivity about LLMs with people’s technical ability. I find the more they say AI is good the worse programmer they are.
Technical literacy in general. My friend thinks it’s the greatest thing ever, is an idiot with technology (and life in general).
It also says a lot about their inability to identify bullshit
Might be some dunnig kruger curve there. Not tooting my horn but I know my ways around and I only use ai for programming when I more or less know how it works already. Which means I verify and fix any eventual problems before committing any code. It does speed up the process, it’s a tiny bit simpler than checking stuff out on stack overflow IMO.
Now, if you don’t know your ways around, and “trusts” the outcome on an LLM, boy are you in trouble 😵💫.
every time I use it, i waste hours
Yes, exactly this. It looks good, I ask for it to tweak something. It tweaks, but now something else needs adjustment. Then it comes back unusable.
It ends up taking the same time as doing it myself. There’s some value perhaps in either the novelty or engagement that keeps me focused but it’s not more efficient.
When it does work, I’m always worried it is an illusion I’ve missed something. Like how you send an email and immediately see the typo.
People who love it, love it because they don’t need to or care about having accuracy and precision in their work. Sales and marketing, management, etc. Business idiots.
Ed ed ed!!!
you’re obviously prompting it wrong, and/or not using the latest models /s
People in this thread unironically saying this. Waiting for the “just wait half a year” model that will be so advanced
I still think back of Linkedin post I saw from someone talking about LLMs and throwing in the sentence:
”Apparently I am really great at making super prompts!”
Which is probably something the LLM told them and they have lost all self-reflection, so…
LLM: Wow, you are so good at this. Are you sure this is your first time? …Oh, prompt me harder, daddy.
Oh man I deleted LinkedIn last month, and March has been glorious.
I literally feel like there’s hope for humanity now. Like just a little glimmer of it. I didn’t even really use it, but it somehow sucked my soul and shattered it into 1,000 pieces.
There are right tools and wrong tools depending on the application.
There are right ways to use said tools and wrong ways…like you wouldn’t use a phillips head screwdriver on a flat head.
I guarantee your company’s provided tool is Copilot or OpenAI based, which is already bottom of the barrel for usefulness.
Haha, yes it is
Inuse a flathead (minus) screwdriver on a philips (plus) screw all the time
My experience is that it can work reasonabky well, but you have to waste absurd amount of tokens and have the 1m token context window.
I only do gamedev, which means a little bit more simple scriots, but it could handle even more involved systems.
If i force it to first document and explain the wholw architecture and data flow.
Did it help? Yes, but it still did take only a slightly faster. I could do it in a day more, probably.
It also cost like 50$ in tokens, in today’s prices - where every AI company is loosing trillions of money, so the costs will get a lot worse. And if I tried to conserve tokens, it’s shit. You have to feed it 10$ of data to be useful.
Add to that the fact that it also causes skill attrition, so once the expensive-cost future arrives, you probsbly won’t be able to afford it, and good luck getting your skills back after that.
Our company wants us to use it, and the average token consumption is like 100$ per day in consumer prices. How is that even considered for such a minor gain?
So, I’ll pass.
Especially once they kill the “old net” and you won’t be able to browse anything to learn skills at all.
kagi is the only way i can stay sane on web 3.0.
There’s enough people who drink the koolaid.
I helped this one as guy use an LLM to migrate his test suite to java21. It did help him incorporate some new language features but I don’t see how it made up for my time sitting with him
…… yet to management he saved 20% time. They trust that number despite no actual measurement, and hold it up as efficiency we all need to find.
But certainly if a 20% efficiency gain were real, that would be well worth $100
Id say you dont know how to use the tool. ‘Write tests here, develop.feature X’ is not a good way to use llms. Using 80k tokens and keep using same a Session is context rotting. There are a lot of boring, everyday tasks in my job that got faster. Many others that meh. Use AI, dont be driven by AI.
I think I’m going to use AI to tell me how to use AI.
I know this community is all about fuck AI, but this is just straight echo chambering.
But honestly your post sounds like you’re just not using it right? You can get pretty good results with it with enough guardrails. Just because you can’t get the results you want doesn’t mean that no one can.
That said, fuck AI. It’s all a bunch of bullshit, but denying real results just means you’re sticking your head in the sand and that’s not how you fix this problem.
I agree… Saying LLMs are good at nothing is just plain ignorance… One can disagree with the philosophy or dislike hallucinations, but they are definitely good at some things.
It’s basically like Google with a bit more detail in my experience. Everytime I’ve tried to use it in a professional context, I’ve come up massively empty. Pages and pages and pages and pages of just absolutely walls of text, but nothing actually useful. I mean I’ve got it to calculate stuff and whatever, but then you examine something and its not coming up for you like the LLM says it should be. Which pretty much immediately means you have to validate everything else, and then it’s like well hey look here I am however many hours later, manually doing something.
Our executives keep telling us to adapt or we’ll be on the losing end. At this point, I’d just like the check please. Because if the company can survive on images of Super Mario committing 9/11, or walls of useless text or just straight up make belief, that’s something I’d like to watch from the sidelines.
pretty good results with it with enough guardrails
examples?
For a research project, I had to convert 20+ projects from a dataset into a new format. The old format was simply a single script for each project that builds it. But I needed a format with a Docker file and a script. It would’ve taken me around a week to do all that one by one.
I got Claude to do it in 2 hours.
I know people hate AI in this community, but to say it doesn’t do anything good or to insult all people who use it is just pure negativity.
Thanks for sharing. I’m not sure where me asking for examples “say it doesn’t do anything good or to insult all people who use it”. Someone makes a claim without any proof, I ask for proof. To me that sounds both simple and legitimate.
Thats good. It has use cases. Is the monetary and earth destroying cost worth it? Not in the slightest
Or that it’s not right for their use case.
Like someone throwing a bunch of data into an LLM and trying to use it to process it into a chart or something. It can work, but it was never designed to be used in that manner.
I’ve got an acquaintance who does that, despite the fact that python would be a better thing to use.
Personally, I sometimes run a few saved images thorough a multi-modal 8 gigaparameter local model on my computer, so I can automate giving them more descriptive names than randomnumbers.png, and that seems to work fine. I could do it by hand, but it would take hours and days, compared to minutes, and since it’s not too important, it doesn’t matter if it’s wrong. The resource usage is also less of an issue, since it’s my own computer.
holy fucking shit man. this community has a clear astroturfing problem.
A lot of people have their livelihood tied to the narrative that LLM deserves every cent of investment. The fact that it’s utility is more limited is an existential threat to their careers.
The truth that it is selectively useful gives them a thread of hope, but the fact it is useless for a lot of stuff drives irritation. We don’t make a distinction between the sort of work that LLM can do and can’t so people end up completely dumbfounded by the other perspective.
I don’t really know what the word means. I was posting my experiences.
not anything to do with your post. it’s the comments here.
fuckai used to be a community where no exceptions were made for AI. it’s quite literally in the name of the community.
so many posts from this community over the last 3-4 weeks has had an increasing amount of users that are “astroturfing” that AI has its uses and can be helpful sometimes. I kind of feel like these comments are made disingenuously as a way to silence the community at large by over commenting in a community that was created literally to hate AI, no exceptions.
anyway, won’t stop me from never using AI. If anything it’ll just make me read more books from before AI was a thing.
OHH I thought the opposite.
I recently used it to install Nvidia l40s drivers on redhat 9 and pass it through to my Frigate instance. Took me a few minutes. Would have been a lot of reading to find the exact answers manually.
Not a bad usage
I also used it to compile the correct Yolo detection models since I wrote this comment. I gave it my specific cameras and settings and it told me how to compile the correct model. I have 200 cameras running now. Almost unheard of on a deployed Frigate instance. Using the correct cuda model now I’m seeing 38 percent usage average on my l40s compared to 80 before.
I cut my LLM usage to almost zero because of environmental and political reasons, but it was helpful enough to wish it could be sustainable and not another tool in the dystopian take on the world.
local models are advanced enough to the point where you can run em as needed without datacenter.
the datacenter craze is basically just an excuse to get the banks (and eventually the american taxpayer, via bailouts when they fail) to fund your local nepitistic infrastructure rollout.
the entire US economy is built around the purposeful boom/bust system, as it’s very effecient at “bagging” people that don’t know the rules.
They’ve still had a huge power investment in creating them.
Oooh buddy, is isn’t even young engineers using these to destroy their designs. I was at a building construction conference recently where one of the presentations was about how AI is going to “give us so much time back” as designers. He then told us about how the AIs hallucinate math still, and that the AI companies are not liable for their output. After the presentation, I and another person asked him a question about who exactly the liability will lie with and how someone could protect themselves from the liability without spending all the time we “save” meticulously checking the outputs. His response was to generate thousands of outputs for the same task and then only check “the best versions.” Okay, so how will we know which are the “best” without meticulously checking thousands of them?
Anyway, afterwards, I asked my colleagues from all around the country who were at the conference for their opinion on AI and the presentation, and most of these 50-60 year old men told me they regularly use it in their work already. So be prepare for things constructed in the past few years to be incredibly dangerous facilities to be in or near.
An idiot intern is better.
Well, 100% because the intern WILL eventually learn. That’s the entire difference. It won’t be about adjusting the prompt, or add yet another layer of “reasoning”, or wait for the next “version” with a different code name an .1% larger dataset. No, you’ll point to the intern they did a mistake, try not calling them an idiot, explain WHY it’s wrong, optionally explain how to do it right, THEN the next time they’ll avoid it or fix it after.
That’s the entire point of having an intern : initially they suck BUT as you train them, they don’t! Meanwhile an LLM, despite technical jargon hijacked by the marketing department, they don’t “learn” (from machine learning) or train (from “training dataset”) or have “neurons” (from “artificial neural networks”) rather it’s just statistics on the next most probable world, sounding right with 0 “reasoning”.
Had a person a few years back who would never ever learn.
In fact, a way I have expressed my opinion of LLM is that it is like working with that useless guy, except at least faster.
Based on my experience, the broader company is chock full of the never learn developers and I suppose I can see why they see value in the LLM, but either way their product sucks and no one likes them.
You’re so right .
And if the person sucks that bad, get rid of them
Yeah, but the same bad management that keeps thinking LLMs are magic are the same bad management that kept that guy around.
Every interaction that guy had where a senior tech ever dared to say he was useless ultimately landed the senior tech in hot water with management, as they claim “he says you aren’t providing what he needs to suceed, that he is very skilled and willing to work, but you never told him how or gave him access or (a million other excuses that were generally lies)”.
After a way too long career with us, he finally overplayed his hand by making the same old claims to the manager about no one giving him what he needed to work. Except he forgot that this time, the manager himself was the one who had been directing him and so he accidentally was accusing the manager of lying to himself.
Finally, the only person with credibility to the manager was on the receiving end of this guys grift.
Its all a grift in the end!
Thats why youll mostly see conservatives/Nazis in love with llms. It fits their propaganda agenda perfectly.
Yeah, don’t generate code with it. Treat it like StackOverflow. It does pretty good at that.
This is the only way I use it, and I do it grudgingly only because AI has ironically also ruined the web and web search. It’s also a last resort for when Kagi isn’t helping.
Unfortunately for me it’s a kpi so I need to figure out how to do something useful with it.
LLM is good for
- temporary scripts like to export data
- boilerplate for new code
- simple or repetitious code like unit tests
But just in time for my performance review, I spent a week ignoring my work to set and tweak rule sets. Now it can be noticeably more useful
- set context so it understands your code better. No more stupid results like switching languages, making up a new test framework, or randomly use a different mocking tool
- create actions. I’m very happy with a code refactoring ruleset I created. It successfully finds refactoring opportunities (matches cyclonatic complexity hotspots) and recommends approaches and is really good at presenting recommendations so I can understand and accept or reject. I tweaked it until it no longer suggests stupid crap, although I really haven’t been able to use much of the code it tries.
- establish workflow. Still in progress but a ruleset to understand how we use our ticketing system, conventions for commit messages , etc. if I can get it to the point of trusting it, it should automate some of the source control actions and work tracking actions
I agree with all that, especially if your performance is being measured by your use of LLMs. Those are cases where I find the code generation to be ok and doesn’t create comprehension debt.
Just literally make something up and get it to lie about something. This is literally the land of make belief at this point, all this KPI shit. Don’t stress about it. Execs want slop, give em slop.
For programming, at least it’s a good way to speed up things that you know how to do but take some time to type, or you don’t remember the syntax of. But relying on AI any more than that usually means you’ll be adding free technical debt and debugging time or becoming dependent on it.
Claude and super powers / planning have changed my mind more on AI feature development. Iterating on the spec and making it as unambiguous as possible gives good results when you clear context and have it implement the plan. Even if it starts to stray you can just do a git reset and start a new session with the spec, adjusting it a bit, because time wise you probably haven’t invested much.
It also depends on the code base, if the code base has very clear separation of concerns, good documentation, and good contracts between layers then claude can handle it pretty well. If the code base is full of spaghetti code with multiple ways to do the same thing then AI will struggle with it. In our large legacy monolith repo it doesn’t do well, in our micro service repos it does great.
Also time wise it may not seem like a benefit if you just set it and wait for it to complete, the productivity advantage comes from running a couple sessions in parallel.
Also context is key, having a good claude.md file in the repo to explain patterns helps it to avoid pitfalls. If it’s only context is the prompt you gave it and you tell it to implement a feature without a plan / spec outlined it will generate shit code.
Making it as unambiguous as possible
If only we had a way to communicate with machines in a reliable, deterministic and unambiguous way.
But natural language will let people without computer programming skills use their business domain knowledge to create compute programs
It’s why COBOL is so popular.
Yeah you can write the code yourself. You can also write in c or even assembly if you really want to make it as unambiguous as possible, it’ll just take more time. Some people like to code in Python though because they can write faster with it even if a lot of implementation details and choices are hidden from them because they don’t care about those details.
Spec driven development in my view is just another step, albeit a big one, on the level of abstraction between assembly and python. Like python it has its places and has places where it should never be used for safety and performance reasons.
They may not care about the implementation details of a Python library, they do care about consistent execution and predictable results. And in some edge cases, they will care about the documentation saying exactly how those edge cases are handled.
Writing Python is abstraction, yes, but it’s still programming. Once that Python code is written and tested and the dependencies are locked down, you can ship it and be certain it always works as designed.
Spec-driven code generation is nothing like that. I can’t ship the specs. I could generate the code in a pipeline and ship that, maybe. But there’s no way I’m getting consistent builds from a code generator. So what do people do? They generate the code and put it in source control for review. When have you ever checked-in a compiled executable or looked at it? There’s machine code in there, shouldn’t you review that the compiler did what you asked of it?
Consistency is dependent on the code base and not the “compiler” in this sense. If the code base has consistent patterns and only has one well documented way to implement something then the AI will follow those patterns, ie. If there is only one way to run a job, AI will use that method. There might be some variation in variable names, formatting, etc. but the core flow should be consistent between “runs”
You can and should still test your code , both manually and with automation to ensure it does what it says it does. Testing should be the way you are certain it always works as designed. IMO understanding your tests and test coverage is more important than understanding the implementation. This is why part of the spec for superpowers is a test plan, and that should be the most reviewed / iterated part.
