If Claude Code and the likes really have gotten THAT good at generating code, could it end up being a win for open source?
-
@tofticles
Jeg er ikke sikker på det overbeviser mig. Hvis det er de samme fejl der gentages, må de også hurtigt kunne rettes og stoppes?
Grunden til min hurtige tanke er at det virker til at der indenfor de sidste få uger er sket en væsentlig forbedring, men jeg kan selvfølgelig sagtens tage fejl.@jramskov For at rette den type fejl, skal du rette i kildedata til modellen. Kildedata, som du ikke har (særlig meget) kontrol over.
Det viser sig at når du bruger materiale fra Stackoverflow + Reddit + Github som primære kilder... Så bliver resultatet derefter.
Læg mærke til at den fejl Kevin påpeger er i kategorien: åbenlys (hvis du leder efter den) - ikke subtile race-conditions osv.
-
This week, Godot (open source game engine) had a press release saying that they were drowning i AI-slop code and don't know if they can keep up the project as they don't have time too sort the crap from the gold.
Edit: https://www.gamingonlinux.com/2026/02/godot-engine-suffering-from-lots-of-ai-slop-code-submissions/
@CarstenBoll Yes, I have seen several such mentions. It is certainly an issue and I don't pretend to have any answers. The Godot team could put an AI to respond 🤪
It is a problem that didn't exist before. I guess projects will need some kind of gatekeeper to prevent them getting flooded or a way to scale up to handle that. -
@CarstenBoll Yes, I have seen several such mentions. It is certainly an issue and I don't pretend to have any answers. The Godot team could put an AI to respond 🤪
It is a problem that didn't exist before. I guess projects will need some kind of gatekeeper to prevent them getting flooded or a way to scale up to handle that.@jramskov The problem they have is the problem I have with AI-generated documentation contributions.
I cannot trust the information to be correct, and it's a lot harder and more time consuming to verify contribution than it is to do the work in the first place.
Simply put, it's not very helpful. The programmer likely saves some time and feels productive, but the time saved is then put onto QA and documentation.
I guess it makes sense if you don't have to worry about your colleagues and how they can do their job.
-
@jramskov For at rette den type fejl, skal du rette i kildedata til modellen. Kildedata, som du ikke har (særlig meget) kontrol over.
Det viser sig at når du bruger materiale fra Stackoverflow + Reddit + Github som primære kilder... Så bliver resultatet derefter.
Læg mærke til at den fejl Kevin påpeger er i kategorien: åbenlys (hvis du leder efter den) - ikke subtile race-conditions osv.
@tofticles Men er det ikke hvad niveauet var for et halvt år siden? Som sagt, det virker til at seneste nye versioner af AI er blevet væsentlig bedre (igen, jeg kan sagtens tage fejl)? Hvor god er AI om 5 år? Har vi allerede nået et plateau eller vil de fortsætte med at blive bedre?
-
@tofticles Men er det ikke hvad niveauet var for et halvt år siden? Som sagt, det virker til at seneste nye versioner af AI er blevet væsentlig bedre (igen, jeg kan sagtens tage fejl)? Hvor god er AI om 5 år? Har vi allerede nået et plateau eller vil de fortsætte med at blive bedre?
@jramskov bedre til HVAD? P.t. gør det ting, mennesker godt kan i forvejen - bare med et meget højere energiforbrug og dybt løsagtigt forhold til resourcer.
-
@jramskov The problem they have is the problem I have with AI-generated documentation contributions.
I cannot trust the information to be correct, and it's a lot harder and more time consuming to verify contribution than it is to do the work in the first place.
Simply put, it's not very helpful. The programmer likely saves some time and feels productive, but the time saved is then put onto QA and documentation.
I guess it makes sense if you don't have to worry about your colleagues and how they can do their job.
@CarstenBoll Excellent points! I guess wonder if that will ever change. To me it feels like everthing is rushing past at breakneck speed and it's impossible to keep up.
I see what I believe are experienced developers be pretty impressed with the latest models then results they are able to get. -
@CarstenBoll Excellent points! I guess wonder if that will ever change. To me it feels like everthing is rushing past at breakneck speed and it's impossible to keep up.
I see what I believe are experienced developers be pretty impressed with the latest models then results they are able to get.@jramskov I mean, I don't know what's the worse outcome - if if works or if it doesn't work.
Who'd benefit from AI that actually works? Certainly not the millions - billions? - made redundant.
Overall, how does this benefit humanity?
-
If Claude Code and the likes really have gotten THAT good at generating code, could it end up being a win for open source?
If you rapidly can re-create new features, then the OS platforms could quickly mirror the proprietary platform offerrings in most ways. No, I haven't thought this through
#AI #RandomThoughts@jramskov Those are massive ifs, though.
As the philosopher Ice-T put it in his seminal work "Bodycount" (1992): "unfortunately, shit ain't like that".
-
@jramskov Those are massive ifs, though.
As the philosopher Ice-T put it in his seminal work "Bodycount" (1992): "unfortunately, shit ain't like that".
-
@jramskov bedre til HVAD? P.t. gør det ting, mennesker godt kan i forvejen - bare med et meget højere energiforbrug og dybt løsagtigt forhold til resourcer.
@Kjaerulv Jeg har ikke et godt svar, det var blot nogle strøtanker eller hvad man nu kalder det.
Men som jeg lige har svaret i tråden, ser jeg hvad jeg anser for værende dygtige udviklere være ret imponerede over hvad de seneste AI modeller kan.*HVIS* (stort hvis) vi ender et sted hvor en meget stor del af den kode der tidligere blev skrevet manuelt, kan produceres på "samlebånd", gætter jeg på det vil betyde store omvæltninger - det gør det på en måde allerede.
-
@Kjaerulv Jeg har ikke et godt svar, det var blot nogle strøtanker eller hvad man nu kalder det.
Men som jeg lige har svaret i tråden, ser jeg hvad jeg anser for værende dygtige udviklere være ret imponerede over hvad de seneste AI modeller kan.*HVIS* (stort hvis) vi ender et sted hvor en meget stor del af den kode der tidligere blev skrevet manuelt, kan produceres på "samlebånd", gætter jeg på det vil betyde store omvæltninger - det gør det på en måde allerede.
@Kjaerulv Ps. Jeg er bestemt ikke blind overfor udfordringerne med højt ressourceforbrug osv., men jeg ser det som en separat diskussion.
-
@jramskov Those are massive ifs, though.
As the philosopher Ice-T put it in his seminal work "Bodycount" (1992): "unfortunately, shit ain't like that".
@fanden Massive ifs indeed
But AI already have major impact in various ways (good and bad). -
@fanden Massive ifs indeed
But AI already have major impact in various ways (good and bad).@jramskov ... the "good and bad" part has serious "very fine people on both sides" vibes, IMHO.
-
@Kjaerulv Ps. Jeg er bestemt ikke blind overfor udfordringerne med højt ressourceforbrug osv., men jeg ser det som en separat diskussion.
@jramskov for mig hænger de to ting uløseligt sammen(entropi) - hvis det koster mere energi at få maskiner til at lave det, end mennesker, så er det spørgsmålet om maskiner skal - for BEDRE end os, det kniber det altså stadig med - og de kodere jeg kender, er mildt sagt ikke imponerede - måske kender jeg de forkerte?
https://media.ccc.de/v/39c3-ai-generated-content-in-wikipedia-a-tale-of-caution -
F folfdk@helvede.net shared this topic
