Like this week's NextFest. I'm seeing complaints from devs that they can't get noticed because of AI slop, and complaints from content creators that many games look the same.
Exactly. The sloppiness won't ever stop. If this has any chance of stemming the price creep then I'm all for it (but let's be real that won't stop either)
I was messing with dx12 again, i actually managed to do the first couple nvidia raytracing tutorials and was doing the one for instancing.
Spent prolly three days trying to get this to work with rasterization. Asked google ai to help me through it, it explained what i was doing wrong and gave me strong words of encouragement.
When I finally got it working which still felt good because I wasnt letting the machine do all the work for me I had to ask it. Is this what people consider "vibe coding"
It tells me that a majority of people just use the llm to try and do everything for them instead of using it as a learning tool/aide.
People who vibe code just put in a prompt and expect the machine to do it all. To me its like the teacher I felt bad for when I would ask a million questions.
I think it can be a powerful thing in the hands of people who want to learn.
Oh I thought this was gonna be Unreal Engine generating NPC dialog on the fly, not just a fancy wrapper for making API calls to Claude. This is less impressive than the shit my friend built in his spare time with local models.
It's great for small scope games, but AI slop gets exponentially unwieldy the bigger the project gets. Without an knowledgeable engineer giving it technical instructions, AI projects are doomed to fail.
AI isn't good for anything at the project level. It does a decent job at generating a single class with a lot of hand holding, but cannot be trusted whatsoever at any higher scope than that.
Any integration must be done by an actual programmer that understands the entire process in order to make it run correctly and efficiently. Otherwise, the AI will spend 95% of the runtime digging holes and filling them back in, or just outright choking to death.
If you thought today's UE5 slop is buggy and poorly optimized, the vibe coded shit on the horizon is about to say "hold my beer."
this was true 4 months ago, but it's not remotely true anymore. the latest models can generate entirely working small-scale projects with interlocking systems just fine, especially with the right skills files to guide design patterns.
the problem with relying on AI is knowledge of what needs to be built, as it needs to be coached how to build it at the technical level, not just the user requirement level.
And it can be handy for occasional shortcuts to small, quick (temporary) solutions and implementations that might require a second pass anyway, even if it had done by hand.
Personally I prefer to just use some separate standalone AI to point me in the right direction if I really need to figure something out and don't want to spend extra time researching it. I usually know enough to identify and decipher anything wrong with the AI's results, and can figure out where it maybe didn't go far enough to do an optimal job.
Sadly yes, it means a lot of idiots will use it to try to put out slop and market it. Not exactly a new problem though.
Get ready for a tidal wave of slop games.
You best start believin' in tidal waves of slop games Ms. Pancake. Yer already in one!
Like this week's NextFest. I'm seeing complaints from devs that they can't get noticed because of AI slop, and complaints from content creators that many games look the same.
Exactly. The sloppiness won't ever stop. If this has any chance of stemming the price creep then I'm all for it (but let's be real that won't stop either)
So in short nothing will change. Got it.
Yada yada yada ai is terrible.
I was messing with dx12 again, i actually managed to do the first couple nvidia raytracing tutorials and was doing the one for instancing.
Spent prolly three days trying to get this to work with rasterization. Asked google ai to help me through it, it explained what i was doing wrong and gave me strong words of encouragement.
When I finally got it working which still felt good because I wasnt letting the machine do all the work for me I had to ask it. Is this what people consider "vibe coding"
It tells me that a majority of people just use the llm to try and do everything for them instead of using it as a learning tool/aide.
People who vibe code just put in a prompt and expect the machine to do it all. To me its like the teacher I felt bad for when I would ask a million questions.
I think it can be a powerful thing in the hands of people who want to learn.
The larger the gaming company, the more propaganda. So might be good?
Plot twist, slop games will be better and more based than what mainstream developers put out.
Oh I thought this was gonna be Unreal Engine generating NPC dialog on the fly, not just a fancy wrapper for making API calls to Claude. This is less impressive than the shit my friend built in his spare time with local models.
We could have had chatbots in games for that 20 years ago!
It's great for small scope games, but AI slop gets exponentially unwieldy the bigger the project gets. Without an knowledgeable engineer giving it technical instructions, AI projects are doomed to fail.
AI isn't good for anything at the project level. It does a decent job at generating a single class with a lot of hand holding, but cannot be trusted whatsoever at any higher scope than that.
Any integration must be done by an actual programmer that understands the entire process in order to make it run correctly and efficiently. Otherwise, the AI will spend 95% of the runtime digging holes and filling them back in, or just outright choking to death.
If you thought today's UE5 slop is buggy and poorly optimized, the vibe coded shit on the horizon is about to say "hold my beer."
this was true 4 months ago, but it's not remotely true anymore. the latest models can generate entirely working small-scale projects with interlocking systems just fine, especially with the right skills files to guide design patterns.
the problem with relying on AI is knowledge of what needs to be built, as it needs to be coached how to build it at the technical level, not just the user requirement level.
And it can be handy for occasional shortcuts to small, quick (temporary) solutions and implementations that might require a second pass anyway, even if it had done by hand.
Personally I prefer to just use some separate standalone AI to point me in the right direction if I really need to figure something out and don't want to spend extra time researching it. I usually know enough to identify and decipher anything wrong with the AI's results, and can figure out where it maybe didn't go far enough to do an optimal job.
Sadly yes, it means a lot of idiots will use it to try to put out slop and market it. Not exactly a new problem though.
Great, more scammy jeetslop with cryptominers