Then you, and Linus, miss the entire fucking point. LLM code submissions look fine at first, but come in huge numbers and eventually lead to massive reverts because they’re pretty fucking useless. Code autocomplete? Maybe that’s a use for LLMs, but they are barely better than non-llm autocompletes. Actual coding LLMs? Universally shit, but are getting better and better at hiding their shit. Like how most LLM agents when asked health questions get 99% of shit right in an answer, and then fuck up so royally on the 1% you will die if you just follow their response.
And it can’t get better. LLMs don’t know know anything, they can’t. They only hallucinate. That is their sole capability.
So to sum up the danger is two fold:
LLM hallucinations get better at hiding behind plausible code.
LLM code submissions drastically increase the number of submissions of code, incentivizing using LLMs to review the code.
Fortune 500 companies are currently in this trap. Microsoft famously is having to roll back so many Windows 11 changes that they admit were primarily coded with Copilot.
FOSS is supposed to be smarter than this, not give into stupid trends that have hard fucking evidence they are temporarily, terrible trends and not a useful tool for the future.
You are using your emotions as statements of fact your terminology is buzzwords you are useless and annoying. Yea we get it you don’t like generated code that is not hard evidence of anything.
You have some surreal sentiment towards code but you seem to not understand that natural language programming, self generated code and artificial intelligence are holy grails that have been dreamed of since the start that have existed in many forms before chatbots got good enough to make waves.
They aren’t ‘making waves’ though. That’s the problem. Companies are quickly finding the permanent inherent limitations of LLMs in regards to coding and several have started desperately looking for Jr Devs again after having a policy of not hiring them for more than a year.
LLMs are objectively a dead end, all current evidence points to this. It would be great if it really were this miracle technology the people selling you the technology claims that it is, but it’s not. It’s not even competent. Bottom barrel slop devs from exploited countries like India are infinitely more capable than claude or copilot. And unfortunately they always will be.
You are just as dumb as the people that over invested you have no idea what you are talking about… Like seriously the extent of what you say is “I don’t like chatbots because they cannot solve every problem ever all at once immediately and I don’t understand code” those are the facts that can be extracted. You are a whiny bullshit artist. If you want people to take you seriously how about you take the subject seriously.
Hilarious you chose this hill to exercise your insecurity on. You are just regurgitating someone elses words but you don’t actually understand what they actually mean so you chuck a tantrum when challenged. The point stands yes, it is still that you don’t know what you are talking about.
Kinda funny you are trying to cite a 50 year old rant as fact but mostly it is just sad.
Then you, and Linus, miss the entire fucking point. LLM code submissions look fine at first, but come in huge numbers and eventually lead to massive reverts because they’re pretty fucking useless. Code autocomplete? Maybe that’s a use for LLMs, but they are barely better than non-llm autocompletes. Actual coding LLMs? Universally shit, but are getting better and better at hiding their shit. Like how most LLM agents when asked health questions get 99% of shit right in an answer, and then fuck up so royally on the 1% you will die if you just follow their response.
And it can’t get better. LLMs don’t know know anything, they can’t. They only hallucinate. That is their sole capability.
So to sum up the danger is two fold:
LLM hallucinations get better at hiding behind plausible code.
LLM code submissions drastically increase the number of submissions of code, incentivizing using LLMs to review the code.
Fortune 500 companies are currently in this trap. Microsoft famously is having to roll back so many Windows 11 changes that they admit were primarily coded with Copilot.
FOSS is supposed to be smarter than this, not give into stupid trends that have hard fucking evidence they are temporarily, terrible trends and not a useful tool for the future.
You are using your emotions as statements of fact your terminology is buzzwords you are useless and annoying. Yea we get it you don’t like generated code that is not hard evidence of anything.
You have some surreal sentiment towards code but you seem to not understand that natural language programming, self generated code and artificial intelligence are holy grails that have been dreamed of since the start that have existed in many forms before chatbots got good enough to make waves.
They aren’t ‘making waves’ though. That’s the problem. Companies are quickly finding the permanent inherent limitations of LLMs in regards to coding and several have started desperately looking for Jr Devs again after having a policy of not hiring them for more than a year.
LLMs are objectively a dead end, all current evidence points to this. It would be great if it really were this miracle technology the people selling you the technology claims that it is, but it’s not. It’s not even competent. Bottom barrel slop devs from exploited countries like India are infinitely more capable than claude or copilot. And unfortunately they always will be.
You are just as dumb as the people that over invested you have no idea what you are talking about… Like seriously the extent of what you say is “I don’t like chatbots because they cannot solve every problem ever all at once immediately and I don’t understand code” those are the facts that can be extracted. You are a whiny bullshit artist. If you want people to take you seriously how about you take the subject seriously.
Disjkstra wrote a nice paper explaining why natural language is terrible to write code… It’s a dream for idiots, actual engineers dread the ambiguity.
Lol you are not knowledgable you think you are using words that have impact but you are just an idiot with nothing to say.
For anyone interested: https://www.cs.utexas.edu/~EWD/transcriptions/EWD06xx/EWD667.html
It’s not because you don’t understand them that they’re meaningless.
Classic LLM-fueled Dunning-Krueger nutcase.
Perfection
deleted by creator
Ah, touché. That was a bit… grandiloquent 😄
The original point still stands though, there was nothing actually mystical about what I was saying before, it was fairly simple to understand.
I honestly cannot figure out why it would fly above your head like that.
Hilarious you chose this hill to exercise your insecurity on. You are just regurgitating someone elses words but you don’t actually understand what they actually mean so you chuck a tantrum when challenged. The point stands yes, it is still that you don’t know what you are talking about.
Kinda funny you are trying to cite a 50 year old rant as fact but mostly it is just sad.
I am… linking to a relevant article? Are you OK?