My current project has 0$ budget and I have yet to use AI art for music or images.
I was able to squeeze 15$ to buy some UI assets and you can easily get Audio loops for about the same range (If not free for commercial use)
The only AI I use is for coding, mostly to learn (I’m not a coder) and to trouble shoot. And even then, I’m using it less every time.
Money is not a valid reason for using AI generated art while there is PLENTY of accessible assets of all kind.
Resources like open game art are amazing and contains a lot of fantastic resources
I don’t recommend using LLMs to learn Godot and GDScript. they do not understand the Godot editor, and give overly complicated coding solutions, often with fake function names or old code examples.
Also, if you don’t use LLM-generated art for ethical reasons, all those ethics still apply to the use of LLMs for coding. (Including the fact that being legally allowed to scrape people’s answers to programming questions on the internet does not make it necessarily ethical.) Beware of high horses - you can fall.
The same could be said about code.
Just a detail, and an honest question : Is the water used for these data centres not recycled, and eventually put back into the water system..? Where is the ‘consumption’ coming from..? Losses from steam or boiling off, or simple evaporation..?
(Disclaimer : I’m not trying to defend these data centres, simply wondering exactly what the issue here is…)
Um, can a moderator lock this forum thread, this has become ridiculous…
While it is true that water will eventually “return” as it evaporates, it really does not matter where it will return.
The main issue here is that a lot of data centers are in areas where water is already a scarce resource, and by having that water siphoned off, just for that water to maybe come back as rain somewhere else won’t help the local population at all. This will also quickly contribute to overall inbalance, where some areas will have too much water, and others will have barely any.
Save for a few of the comments being a bit light hearted there’s nothing ridiculous about this thread, and you saying people are being so is a bit offensive
What @tibaverus said.
What you have to keep in mind is that in certain areas, water is a finite resource. For example, the Colorado River supports seven states: Arizona, California, Colorado, Nevada, New Mexico, Utah, and Wyoming. When rain doesn’t fall, or snow doesn’t melt into the river at the expected times, if someone upriver uses up too much water, people downriver don’t have enough water. This is a common problem and why many cities in these states have ordinances on how much water can be used. They prevent people from watering lawns, etc.
Each AI data center uses up the same amount of water as an entire city. Which means, if you build up enough data centers in places that don’t have issues with water, you can create them.
The problem isn’t that water is not renewable, but that it cannot be renewed while its in use.
Ridiculous in what way?
Did you get your question answered? You never really said if you did.
I think the length of this thread - which is still on topic - is another answer to your initial question. People have a lot to say on it.
You can always mute the thread if the number of replies is annoying you.
Also, if you don’t use LLM-generated art for ethical reasons, all those ethics still apply to the use of LLMs for coding. (Including the fact that being legally allowed to scrape people’s answers to programming questions on the internet does not make it necessarily ethical.)
In that case, how is it different from simply manually go into those forums. Copy-paste the code that seems to fits (Changing lines as needed for the project) and learn that way.
In the end, both are scouring the internet for the coding solution you need, the main difference is speed.
I’m curious how much bigger the energy impact is, between manual Internet searching that can take hours (Depending on how specific the problem is) and a few seconds to 1-2 minutes of AI searching.
Also, would it be more ethical to Use AI to code quick prototype (regardless of how bad its coding) and recode more eficiently afterward?
I’m not well-versed into the coding community, but from other project I worked on, I often heard “I coded it base on some old code I found on the internet”. Consider those comment was before AI chatGPT and such.
That part is exactly what, you are doing the work and learning from that
Would you say that you learn to write code if you ask your teacher to write the solution for you?
It’s minutes from your machine’s perspective. It’s possibly years for the infrastructure when translated to your machine’s compute/time resources.
An average person can hardly comprehend how many compute-dollars went and is still going into this shit. The sad thing is that most of them were borrowed from the future.
As for the direct answer to the question from the title of the thread, it’s simple, they’re looked down upon because “ai” is unethical while its results are not really aesthetically pleasing. That’s why it’s called - slop.
How much money are you making off someone using your searches?
The main difference to you is speed. I’ve already laid out the many differences in multiple replies in this thread, including the first reply at the top of the thread. But it case you missed it:
30 minutes of searching using Google without AI LLM assistance uses 0.009 to 0.018 kWh of power and NO water.
1 single LLM query uses 0.3 to 3 watt-hours (Wh) of electricity and approximately 0.3 to 5 milliliters (mL) of water. (On the low end, that’s the amount of energy a high-efficiency lightbulb uses in a couple of minutes.)
So you would have to do Google searches for at least 8 hours and 20 minutes to equal the amount of power consumed by 1 LLM query. And that’s assuming each search took the maximum amount of power and the LLM query used the least amount. On the upper end, you’d have to do Google searches for almost 7 days straight, 24 hours a day to equal 1 LLM search using the max power, and assuming the Google searches used minimal power.
Whether it’s 7 hours or 7 days of searching, you can tell that Google searches are much more energy efficient - and LLM queries have improved their efficiency by a factor of 50 in the last year.
Of course, I don’t think you’re actually curious, because you could have found that same information like I did, by doing a Google search.
No. Changing AI output after the fact does not change the ethical implications of using it to begin with.
I am well-versed in the coding community. I’ve been coding professionally for 30 years. The difference between those two examples is legion.
For example, by searching yourself, reading the comments on multiple posts, and selecting the code you need to copy, you are exercising your brain. You will learn more, and retain more information about the problem for next time. Studies show that people’s ability to retain information drastically decreases each time the use an LLM to do work for them. Here’s an MIT study done on the subject.
These answers on forums like Stack Overflow also rely on reputation. When a person answers your question, you can see how many questions they’ve answered before and what their overall reputation is on the site (it’s tracked by a number). This gives you an indication of how good their answer is going to be. It also gives you the ability to give feedback on the answer, and upvote that person’s answer - giving them a reason to keep helping others.
One of the problems with relying on Chatbots is they rely on humans to answer questions. When people stop asking questions on forums like this one (which is already a problem happening now), people stop answering them. That means that in the future, LLMs will not have new information to draw from because experienced programmers like myself solve most problems on our own. And if we aren’t helping newer people to answer questions, then there’s nothing for LLMs to scrape - and they get dumber, because they continue to scrape the answers of other LLMs that don’t know what they’re doing. (That’s why LLM answers are not allowed on this forum.)
This brings us to the other thing that copy some dude’s answer on the Internet causes: When no one has answered your question, you post on a forum and engage with a community. You then also become an expert in things over time and might start contributing answer5s to people newer than you. If you are just talking to an LLM chatbot, you’ll never engage with the community providing the answer, and the community will die - and the answers will dry up.
Godot 4.7 has some awesome new features. There’s no information on the internet for Chatbots to scrape. They cannot tell you how it works. The only people who can tell you how it works is the person who built it, or someone who plays with it or a programmer who reads the code and writes up instructions (or makes a YouTube video) on how to use it.
I get how if you don’t know the history of programming, it seems like LLMs are the future, and they’re just a faster version of the past. But LLMs only work because humans are creating content for them to steal. Without that, they atrophy and die.
When ChatGPT came out on Nov 30th, 2022, it was using GPT-3.5 to power it. It did not know anything that happened before September of 2021, because it was trained on the Internet up to that date. That meant it knew nothing of current events, or of what had happened in programming in the past year. It was great at writing cool-looking stories. It sucked at most technical things, and could not do math at all. Models these days are more up-to-date, but if you experienced that early knowledge gap, it helps you to understand their limitations.
Knowing history can help you predict the future. If you set the ecological and ethical impacts aside for a moment, it’s important to see that the more humans rely on LLMs to give them answers, the less people there are to provide answers for the LLMs to scrape.
I found this article: How Gossip Goblin is becoming the auteur of the AI age
You will probably hate it. I found it very interesting though. Because he talks about making videos and shorts with all the existing filmmaking tools (and workers), but utilizing a few dozen LLM tools to create visuals.
I personally don’t like what the guys is producing, but apparently a million people do. (Though I suppose you could say in the whole population of the world, that’s rounding error.) I find his characters to be unsettling and upsetting. But perhaps that’s what he’s going for.
As a result of people complaining about AI, Sora is being taken down.
Yesterday OpenAI was saying it was for cost cutting reasons, who knows where reality lays?
Cheers !
I hate almost every article you link. They are all Anthropic’s PR 
Yeah the film looks awful, and not in a “shocking” kind of way they were supposedly going for.
There’s this overarching design principle called gestalt. It posits that the unified whole should always be more than the sum of its parts. Every good piece of human art, in any discipline, has this unifying quality. Every piece of slop I’ve ever seen or heard somehow falls short in this regard.
That’s just for direct “inference”. There’s additional massive overhead in data scraping, pre-training and reinforcement training. The latter also includes hordes of humans doing ginormous amounts of soul-sucking work.
There are many ways to be persuasive; it’s coercion that disturbed me, which communicates something that I find disagreeable.
In this post you outlined the practical impact of AI, e.g. ecological and economic impact. I think it’s good to stand against AI from a standpoint of facts of how destructive AI can be.
But some may hold AI to be harmful to our creativity as well, which I completely understand and agree to. The consequence of relegating creativity to AI is that you might produce things no one will be interested in. Or not. Nonetheless, I don’t think it’s the place for fellow game developers to dictate how other devs approach their work.
(I occasionally use LLMs/AI because it’s there, it is helpful at times, but not necessary because I do the art, music, story, pipeline, game logic, etc, of my wip game. I use AI images to learn or get inspired from its results. I don’t want to use the AI images themselves because, I want to make things, not just display things.)
The explanation in the MSN article might be incomplete.
My takeaway is that they sucked at monetising entertainment tech (thank heaven), and prefer the easier route of governement contracts (military or otherwise) hence the pivot to ‘robotics’.