James Gosling has laughed off claims that AI can replace SWE. Gosling called AI “mostly a scam” driven by grifters, and generative output “fundamentally horseshit.”https://ecweb.ecer.com/topic/en/detail-286705-java_creator_james_gosling_calls_ai_overhyped_warns_of_big_tech_exploitation.html
>>542231686It'll replace them eventually but the current systems are fundamentally flawed.
>>542231686All these old tech superheroes do not really understand LLMs or transformers or any of that stuffDo not listen to themListen only to what Geoffrey Hinton, the Godfather of AI is sayingYeah Hinton too make prediction mistakes but he is a genius like the other tech superheroesAnd his field of work is ACTUALLY AIDO NOT sit around and wait until you get replace. Do something while you still canSpecifically, save the money you still have in a savings accountDo not spend unnecessarily. Have at least 500 K in reservesYou will need it pretty soon in the future
>>542231686At present, AI cannot replace a SWE fully, but a technically skilled software engineer with the ability to draw up good technical requirements can create great things using AI. An unskilled engineer would probably create horseshit, however.
Because these gigantic giggahomos won’t tell you. SWE means sweatlord programmers.
>>542231686Based.
>>542232204ContdWatch these two Geoffrey Hinton interviewshttps://www.youtube.com/watch?v=l6ZcFa8pybEhttps://www.youtube.com/watch?v=Fj4Dk4wrovEOn this topic Geoffrey Hinton is better
>>542231686> Java creator> AI badmixed messaging
>>542231686The code quality is indeed not as the one of good human software engineers. The reason for that is that, besides the pre-training on open source code, AI is trained to solve programming problems on its own. And for that, the reward function is mostly determined by whether the code works or not. Astra for example produces absolutely remarkable 3D games, but the code (I hear, I have not seen it myself) that it produces is absolutely unreadable.If you have any critical software, this unreadable code is a problem. If you just have a small game or whatever, it may be less of a problem.There is also a good chance that the AI companies will improve on this in the future by rewarding readable code more.
>>542231686He's correct. Same shell game, different tool but always other people's money.
>>542232721Maintaining of the code quality still requires human supervision, and by "human" I don't mean subhuman.Nowadays it's more about being able to use the minimal amount of tokens to enhance the productivity as much as possible i.e., stability and extensibility, instead of running an infinite shitty feature creep.
>>542231686I really do not care, I use claude to do my programming work I'd never do because I am not programmer and I just needed fast excel functions to be implemented or python sorting machine for my data which would work faster than excels pivot tables.It works, it sped up my work, so it is not a scam.
>>542233310> It workshow do you know?you have done basic and edge cases tests, right right?
>>542231812it won't replace shit. AI is a jew scam.
>>542233459What the fuck are you talking about?
>>542232204>jeet talking about others misunderstanding technology lol, lmao. Your people are a plague on all of technology.
>>542233590so the answer is "no"I'm starting to think SWEs aren't going anywhere Kek, this will be worse than hindupoocalipse
>>542232721>The code quality is indeed not as the one of good human software engineersYou don’t know what the fuck you’re talking about. The code is better than 80% of humans. Programmers (especially the ones infesting corporations and academia) are lazy entitled midwits. Many of them can’t even speak English well enough to write a proper code comment. Most let obvious, critical bugs go on for months out of spite for their entitled welfare visa.I can’t wait for AI to sweep them all away.
>>542234024if the 80% of code is shitand if LLMs were trained on that codewouldn't it mean that 80% of LLM generated code will be shit?or will something EMERGE to save the day?
>>542233735I do not even know what SWE is. I just use this shit to speed up my work which is completely unrelated to programming. And it works. I can make FEA mesh models with this shit. You know how long it normally takes to make fucking mesh model in FEA?Yeah it is not perfect model, but considering I'd normally spend 2 weeks doing one, and here I make it in 4 hours and then just tweak it for another 4 it is 14 times faster than doing it by hand. And it works, it gives proper results when compared with real data.
>>542234188Coding is one area where AI can be trained to get better and betterBecause Coding has several measure failure testsSuch as- the code does not compile- the code compiles but the program crashes- the code compiles, the program does not crash but does not work as intended- the program does the above but is too slowAll of the above are measurableAnd since they are measurable the AI can be improved to get better and better at it
>>542231686>mostly a scam” driven by grifters, and generative output “fundamentally horseshit.”So.... just like software "engineering"
>>542234194That's okay, anon. SoftWare Engineer.I'm not asking about tasks which you could expertly judge. If LLMs could do your job, I'm happy for you.However, >>542233310> my PROGRAMMING workThis suggests that programming isn't your forte. That's okay, too. But, how do you know that your LLM generated scripts are producing correct output?You can a) review the code (you can't), b) write tests running the program on predefined inputs and comparing the actual results with the expected ones (you didn't).See the problem already?>>542234374> better and better You don't know what you're talking about.Please start memeflagging.
>>542234374ContdRule of thumbIn those fields, where AI is used, where success or failure, as well as degree of success can be measuredAI can be trained to get better and better during the TRAINING phaseNot the inference phase(Where regular users such as you and I come into the picture)
>>542231686>cobbler laughs off the idea that machines will ever be able to produce shoes as fine as his hand-made goods
>>542234572>You don't know what you're talking about.You don't know much about how LLMs learn codingLooks like you do not know what recursive self learning is
>>542234572>But, how do you know that your LLM generated scripts are producing correct output?Because I check it in software that reads the output and correct it as I stated. Duh. I even wrote it. You might be a programmer, but your understanding of how these things work is apparently quite shitty.
>>542234674For those actually interested in learning about AIAs opposed to just showboatingHere is the relevant material from Google AIYour decision whether you want to listen to what is below or to some Yahoo from 4chanRecursive self-improvement in AI and coding occurs when an artificial intelligence system autonomously writes, tests, and refines its own code or training algorithms to enhance its capabilities without human intervention. How the Coding Loop Works • Generate: The AI agent writes new code, optimizes training loops, or alters its underlying architecture. • Verify: The system executes the code in a sandbox or test environment to measure performance against specific benchmarks or reward functions. • Keep or Discard: Beneficial modifications are retained and added to the persistent memory or meta-library, while failed updates are discarded. • Repeat: The process compounds continuously, with each improved generation accelerating the development of the next. Current Real-World Applications • Self-Editing Agents: Frontier models like Claude handle thousands of repetitive bug fixes, refactor massive codebases, and optimize training efficiency at speeds far exceeding human developers. • Automated Research: Systems are increasingly tasked with designing their own machine learning experiments, generating reward code, and searching for novel algorithmic solutions.
>>542235004you're a joke.
>>542235104So is there a refutation of what was posted thereOr some pretensions of superiority and half witted snark?
>>542231686It's not just him and the arguments are compelling:https://youtu.be/Lf5oqGOCRCM
>>542231686AI is for pajeets.
>>542235004>VerifyHilarity issues
>>542235192That is ED Zitron, famous AI doomerHe hails from the world of 'finance'
>>542231686>retarded dot com boomer yells at cloudI don't care.
>>542231686AI built me this fuckshit in forty minutes. Pretty sweet if you ask me.
>>542231686the scam is they are using data centers to mine crypto currency the cia created. why is the hash rate so high right now when the energy costs more than the crypto it can mine? who is mining the crypto when it is not profitable?
>>542235273He's pretty based. A razor sharp intellect and powerfully eloquent. He may err in many details but in general he has it right.
>>542234188Good code is amplified in the model, retard.
>>542232575Good find Visajeet. I trade you another from Roman Yampolskiyhttps://youtu.be/SnKM6IFzxwM?is=R7hW7-wKhWiiWF8gIt's long but worth it
>>542236703have already watched that as wellPretty good
>>542236797Ah fair. I feel like we need to make a weekly daily of some of the best compilations to wake the people up. This is a major topic that will effect all of our lives
>>542231686KEK, there is not a single application developed in Java in the last 30 years that has not already been surpassed by LLM's from a year ago. Thats why sales force and oracle are firing their entire staff.
>>542232306>can create great thingsCan we see them?
>guy who invented the biggest piece of shit in modern computing thinks SWE wont be replaced by AIKekaroo
>>542237008That’s not impressive when the LLM is a giant database of those very same java programs.
>>542231686>11/26/25The people that poopoo AI at this point are Luddites or people that simply have no idea how to use it.
>>542231686It already did and everyone in the industry is in denial.
>>542231686He sounds like he formed an opinion of it two years ago and hasn't looked at it since. You can't seriously engage with these systems now and hold this position
>>542232204>don't listen to people who know shit>listen to boomer doomer that sells books about doom
>>542232204>save the money you still have in a savings accountSatire.500k today will be worth 200k or less in a year. Buy canned food and extra shoes and a wood stove.
>>542231812>two more weeks
>>542231686>https://ecweb.ecer.com/topic/en/detail-286705-java_creator_james_gosling_calls_ai_overhyped_warns_of_big_tech_exploitation.htmlhe's not wrong about these AI services being run by absolute grifters. It is partially a scam. There is a lot of shady shit going in in the world of AI.
>>542238774how is that any different than people that work on it, except being more efficient
>>542242331>james goslingnobody cares about butthurt canadian faggots say. if canadians knew anything they would actually have an economy. you will never be a real country
