Give me a reason why it can't replace a software engineer? > Extremely good at finding bugs> Extremely good at writing code> Extremely good at testingIt has it all.
>>110114253Too expensive for customer, but not expensive enough for Anthropic
>cluadedam op you're black
>>110114253well, because... it might make mistakes along the way and only humans don't make mistakes.
Anthropic is a scam company. Mistral and the Chinese do 80% of what they do at 20% of the cost.
> Extremely good at bloating codebases> Extremely good at introducing bugs> Extremely good at causing psychosis > Extremely good at costing thousands of dollarsGreat for making throwaway shit, terrible if you're making anything big unsupervised. All these so called big impressive projects made 100% with Claude would be in use instead of nobody using them, Anthropic itself wouldn't be hiring programmers, and we would have shot and buried retarded shit like Electron apps.
>>110114253Software engineering, network development and data engineering were always dead marketing terms, it's up to you for not jumping ship in 2019. Fuck marketing with a cactus, if you can't I feel sorry for you.
>>110114253all of those things are incidental, the role of the software engineer is being the computer guy
Explain why we needed yet another shill thread?
>>110114253There are things only human can understand like user experience, what can be sacrificed for a performance gain and what cannot.
>>110114253the software engineer that you pay minimum wage writes the code.the software engineer that makes a six figure salary spends all of their time reviewing other people's code.yes an AI is probably better than you at reviewing code.but unlike a human, it won't get better at reviewing code from it's mistakes. If it makes a certain mistake, it will keep on making those mistakes over and over again. And yes, before you ask, this still applies if you tell the AI to not make mistakes.So basically, it's probably cheaper to ask a programmer on fiverr to use an AI (probably a chinese AI that's 200x cheaper) so that they review the code that the AI wrote for you, and they can prompt the AI better because they should know how to code than you. Because the AI won't always implement important features for you that you didn't ask for.For stuff that has already been done 1000 times in every single language, AI is perfect. Like making your own twitter or website or whatever. But for stuff that you can't easily verify with tests, or no readily available open source examples, the AI sucks.
>>110114253Basically:> Extremely good at finding bugs sometimes> Extremely good at writing code sometimes> Extremely good at testing sometimesAnd if you're a nocoder, you can't see when it fucked up, or even name the mistakes to fix em, or have knowledge about what it could fuck up on other machines.It's pretty much like getting a chinese tattoo.
Just out of curiosity. Has claude ever done anything without ever being prompted?
>>110115142You don't need a lot of people to write a prompt.
>>110114253Using Claude makes you a cuck. It's that simple. Do you want your business to rely on cuckoldry?
>>110115396It's just an example. Codex, Astra, you name it.
>>110115406You sit on the sidelines watching someone else write code for you. Using any ai to write for you makes you a certified cuck.
>>110114296I don't disagree with you but to be fair, those points also apply to a large majority of programmers.
>>110114253every day with this fucking thing>hey no, we need to do it this way>yes, obviously>okay, and? keep going>why are you doing that? i told you to do that? okay we need to rephrase that instruction>okay i see so i told you to do things a certain way last week and then you phrased it both broadly and even more strictly>and yes, please continue
>>110115155You just need 1 person to write a prompt. What is interesting is that claude needs a prompt. Meanwhile, actual humans don't require a prompt. They have an ability to act without one.
>>110114253You need to become the engineer
>>110115435Yeah but he gets money for it.
>>110114296>>110114296>All these so called big impressive projects made 100% with Claude would be in use instead of nobody using them, Anthropic itself wouldn't be hiring programmersThey are using Claude though.
>>110114253because shit in, shit out. if you want good code, you need a good coder that steers claude and has the domain knowledge and experience. this will lead to the top10% of programmers ruling like gods getting paid high 6 figures maybe even 7 and the bottom 90% losing their job.
>>110115639So it's acceptable to let someone else fuck your wife so long as they pay?
>>110114253AI could replace an engineer, but AI + a human engineer is probably almost always going to be better.
>>110114296It's hilarious that people like you suddenly assume human-written code isn't full of bugs and bloated as fuck. Opus writes cleaner shit than 99% of codemonkeys out there.Hell, you can even ask a cheaper model like Sonnet to document the whole codebase in a clean .md and add proper comments.
>>110114253only good at doing things that have already been done 10000 times
>>110115824Elaborate
>>110115821So, Indians can finally write a good code now?
>>110115713>saaar
>>110115897Yes, timmies are useless. A Jeet with $20 Claude subscription can replace a white boi senior SWE.
>>110115570Why would you ever talk to it? Write a proper contract/agent file and let the clanker get to work. It can go file its objections to /dev/null for all I care
>>110115784Except the engineer you hired isn't even reading the code so you're better off alone with your sub
>>110115897No because jeets can't imagine concepts to prompt them in the first place.
>>110115996Sounds like cope. Prompting is way easier. Jeets are winning actually.
>>110114253Because AI is the pinnacle of the Pareto Principle. It will get you 80% of the way there, but that last 20% will cause all the bugs and issues, and running claude on a loop to try and fix those issues only creates new ones and exponentially increases the price, to the point where it's completely unfeasible.80% of the effort will cost you maybe like, 500 to 10k depending on what you're trying to make. That last 20% will cost you a god damn fortune. Upwards of like 50k to 100k depending on what godforsaken thing you're trying to make, and the software will feel like a buggy sloppy mess until you put in the tokens to fix that 20%.So as of right now, the best thing is a codemonkey that actually knows what he wants, how to spot issues, and how to fix things, and to use AI like cruise control for the more mundane, time consuming code.
>>110114253is there an equivalent local model? the idea of being connected and in communication with a giant corpo to use this shit sounds gay and retarded
>>110116001jeets are not needed anymore. claude is the jeet now
>>110114253learn welding
>>110116066I wear glasses
>>110116054Wait for a year, the Chinese will eventually catch up. But you'll never be able to run this locally anyway, unless there's some kind of insane leap in efficiency.Though it's still better to rent just the compute and use an open source model.
>>110114253It's so fucking dumb. Got it reviewing PRs and it says x y z happens, if this is intention pls document. Meanwhile the comment above that line says x y z happens and this is intentional. Push back and it says oopsie woopsie false positive xd FUCKING DUMB BOT STOP WASTING MY TIME and yes this happened 5 mins ago. Please kill yourself if you shill this garbage.
>>110116121Opus 5 on high reasoning never does that. It's extremely good and smarter than you, actually.
>>110116121Are you sure you've selected Opus 5.5? Sounds like Haiku type shit.
>>110116016>It will get you 80% of the way there, but that last 20% will cause all the bugs and issues, and running claude on a loop to try and fix those issues only creates new ones and exponentially increases the price, to the point where it's completely unfeasible.This is no different from pre-AI software except you just threw more nerds at it instead of tokens lol.
>>110114253it needs precedents, incapable of novel deliberation, non-sentient
>>110114253It's really quick at making websites/webapps not good, but very fast. But web shit doesn't matter as you can half ass-it and it still works even if it is only 5% efficientIf you are a CEO type person then sending a prompt to get a web app dashboard or whatever is the greatest thing to ever exist.For all other real work it sucks and can't do it
>>110114660>And yes, before you ask, this still applies if you tell the AI to not make mistakes.
>>110116182
>>110114296Every piece of commercial software sold today is 95+% SI generated, retard. Everything you use is written almost entirely by SI.
>>110114253
>>110115959Then you get fired when the AI fucks up on autopilot kek.
>>110114253I had ChatGPT make a webpage and I still had to comb through it and fix random things and it's still not fully fixed by the end of the work day. LLMs are anything but perfect. Maybe HTML just isn't their strong suit.
>>110114253Try configuring Neovim from scratch without plugins using an LLM and you'll see what a pain it is.LLMs are all fun and games until you actually have to work with what they made.
>>110116286>>110116282>>110116277Cheap or free models. Frontier models like Opus or Astra don't fuck up.
>>110114253Non software engineers are shockingly bad at using Claude Code.That's the only real moat right now
>>110116277>storing important shit in /tmp/sounds like the problem lies between the chair and the monitor
>>110116491The problem is the keyboard.
>>110116491It fucking did that on its own without asking me
>>110116551it probably looked for the only place it could write cause you harnessed it like a tardogive it a separate user and a sandbox where it has proper permissions
>>110114253>Give me a reason why it can't replace a software engineer? >Proceeds to list a subset of SWEI wonder
>>110116572Luddite cope
>>110116572>>110114266>>110114296Cope. AI is extremely good
