Using AI in any capacity is a bad decision even ignoring the horrible environmental effects it has. Not only do you not learn the better practices it “suggests,” you are training yourself to be reliant on something that is going to go away when the bubble bursts.
Putting any ethical concerns aside, this is a terrible take. If you know what you’re doing AI can be an amazing tool to speed up your development processes or help especially with repetitive idioms. You don’t “learn the better practices it suggests” unless you make a deliberate effort to. Your whole argument is applicable specifically to people who rely entirely on it to vibecode without knowing what’s going on (usually don’t know how to prompt or deal with hallucinations either).
Also, if its tedious there’s a good chance there’s a more efficient way to do what you’re doing or its important enough you don’t want something like AI to do it.
Note sure what you’re talking about here? OP said it can be used to speed up tedious processes of development that you do understand yourself, but don’t want to waste time on implementing. For example, you can use it to prototype faster once you have a base framework established already, say applying some logic to other objects based on certain conditions that are way easier to put in words rather than handle this manually. If something is tedious that does not automatically mean a “more efficient” way is likely to exist.
Your entire argument is read as though it’s directed to people who think LLMs are a good substitute to thinking for yourself, not to people who are experienced in using it.
One thing I’m glad for is people like you exist on the forum to show how logical arguments work, logical fallacies are rampant whenever arguments are involved and has been like this since I joined. Probably something to do with a majority of the audience on this forum being kids and that being more likely to occur.
Agreed, that argument should only be made if someone heavily relies on AI. However, I still think we shouldn’t devote alot of resources and time into making all of these tools when they’re only going to be our assistants. Sure, it’s a nice thing for like advanced scripters as they can debug and still be efficient with it, but beginners might rely on it a little too much to the point that they don’t even know any code and try to make a game while heavily relying on AI to do it. As I would say, if a beginner ever tried to even make a gun system with AI, they would totally fail as most of the time it’s small errors and just human mistakes. Alot of people blame AI for their wrong doings during development, which might be making it learn the wrong stuff.
No tyler I am 19 and I do not feel like a clown, in fact I myself use AI to to find ways to achieve things I cannot waste time to ask for advice on the forum, or learn math like CFrames. I don’t hate AI but when people make things like these that do all the homework for them without this is where I draw the line (studio already has an AI assistant so why even use this).
I find it also odd that this is coming from you considering you yourself hate when low quality slop is out there being more popular than the WildWest which had way more effort poured into and I respect that.
And yeah you are right I’ve never released a game so guess you are right there, although I make my projects as a hobby for mostly myself.
Okay my bad I should have dug further, and I apologize for that, finding games with cutscenes on Roblox is like a rare thing nowadays and those games actually look like they had a lot of effort poured into on them.
I’m sorry but since when was it “massive experience” if you use Rojo…?? It seems like you don’t know what you’re doing if that’s what you define as massive experience.
A completely new developer could know about Rojo if they watched the right video that told them about it. But personally, I highly dislike Rojo — it adds unnecessary complexity without bringing any new features — and is by no means a sign of experience or skill.
I can’t be the only one who reads this as “AI-generated, AI-generated, AI-generated…”
I’m aware that you’re advertising AI, but ironically I feel like AI is a bad way to advertise it.
This is the problem that should be focused on being solved instead of trying to advocate for technological progress to stagnate. These tools won’t be valuable assistants if they don’t have resources dedicated to improving them so that they are actually more useful to experts in their respective fields.
You’re absolutely right, just knowing how to run Rojo doesn’t equate to “massive experience.” That’s my mistake in wording.
What I meant is that it used to take serious effort to even get to that point. I am a self-taught programmer. I was 12 years old when I started programming. Back in 2016–2021, there was no AI, barely any documentation, and almost no mentors because the few skilled devs were already making serious money. I spent years figuring things out the hard way, googling undocumented quirks, reverse-engineering limitations in development. Only when I was mentored by a real software engineer I was able to eventually realize Rojo’s value for real version control and team workflows.
It took me 3 years to even consider Rojo useful. Today, SuperbulletAI removes that barrier. Rojo can be auto-setup in 30 seconds. I forked it and re-modified Rojo to make that possible.
So yeah, running Rojo today is easy. But getting there, back then, wasn’t.
My stance on the situation of Artificial Intelligence is that we should develop the assistants for further and more complex operations, and still balance the tide by encouraging in-coming developers to gain knowledge on how to program and give them more resources that can be used to assist them in their path. AI is still a reliable assistant as long as somebody doesn’t rely on it too much and uses it wisely. We should just balance it out to help improve the new programmers and guide them, that’s just what I believe.
Absolutely. I completely agree. Expert feedback is essential and in this case you’ll be the expert now, not just to avoid stagnation, but to ensure these tools actually evolve to serve developers instead of replacing or frustrating them.
That’s exactly why I’ve positioned this tool as both a learning ground for beginners and a productivity booster for intermediate scripters. Our dev team, including developers with 2–4 years of Roblox experience (some of whom I personally mentored), have been genuinely finding value in it. Their feedback has been critical in helping us improve it rapidly. We’re not building this in a vacuum, we want to build it with the community.
And let’s be real, AI will always make mistakes. It’s not perfect, and it’ll take time before it consistently nails everything. By 2030, it’ll be so strong it’s inevitable that all Roblox programmers will use AI to prototype faster. We already use this AI tool in our own workflows, and it still occasionally messes up API calls or logic. But it’s a massive game changer, and even with those flaws, it’s helping non-coders learn Roblox Lua faster than if they were doing everything manually, based on many users we have in Early Access phase.
Why? Because instead of staring at a blank script, they get working examples, instant iterations, and a faster feedback loop.
Just laying down the foundations of Roblox Lua is already hard unless you’re seriously persistent just like me who learnt it by never giving up, and that’s exactly why there are so few Roblox programmers in the world today. We’re trying to change that.
We’ve even seen 30+ year old veteran game designers outside the Roblox industry but been in the game industry who are beta testers, people who never got into Roblox because of the technical barrier, finally breaking through and learning to script. One of them has been making massive progress recently thanks to this tool. That’s the kind of impact we’re aiming for.
I know, I read the documentations to learn Roblox Lua. I’m talking about people being able to understand it clearly unless they’ve read similar documentations in the past. (or have learned programming elsewhere such as schools)
Do take in mind I learned Roblox at 12 years old. Nobody taught me.
The AI we use to assist us and teach us is valuable because it accelerates learning and efficiency. While I agree no one should heavily rely on AI to develop complex programming and such, it’s still would be a nice idea to give AI the power to do that so it can help with lesser situations. My perspective has changed since the past few hours, since now that I think about it. And to your reply below, yeah I also kinda self-taught myself how to code at twelve until a few years into a future I started using AI. I had to steer myself away, but again, it’s still a useful tool for assisting and teaching.
I’ve added a couple of FAQs in the post, just recently. Initially, I built this product for our company internally ONLY. To create more engaging Roblox games faster than almost anyone. Focusing on the gameplay of our Roblox games, we’re huge game designers after all with tons of ideas surely players would love. But over time, I realized I wanted to contribute something meaningful and help position our company as a leading Roblox game studio that taught many Roblox players. By releasing SuperbulletAI publicly, we’re giving everyday players, especially those who love Roblox and dreamt of being able to make their Roblox game, a chance to try game development for themselves.
It’s mainly due to technical barrier for beginners that we built this product. Roblox AI won’t cut it, hence why this product is live.
This feels like more of encouragement to put less effort into your games as opposed to actually gaining the experience and skill to make a good game in the first place.
I’m not a fan of the vagueness of the policies on the site either. All of this seems especially broad and non-specific, which hits harder when it comes to something as data-focused as AI gen-- and ESPECIALLY since you’ve used an example of a young child hypothetically using this tool, even if hyperbole:
Also ignoring that the majority of your replies and written text on the websites dedicated to information about BulletMindV1 are clearly AI generated. (chatgpt voice) It isn’t just sketchy, it’s inauthentic!
Privacy policy, 2.1-2.4, regarding data collection:
This seems… extensive, even for an AI tool.
Would there be a way to opt-out of collection of the assets and information you upload, or even to anonymize any of it? Specifically looking for a more in-depth explanation of what you mean by “all of that”. There’s a lot of data you’re collecting, which is specified in the privacy policy as what is necessary to provide [your] services.
What does “user interactions” entail here? To what magnitude?
Why would you need the user’s browser information?
You say the user has ownership of all assets and content created by the generator, but immediately contradict it by claiming royalty-free usage for yourself, implying for training purposes in the Privacy Policy, but remain broad in the TOS:
In the case of games with IP sensitivity or any kind of workflow involving NDAs, this is an absolute no-go. Again, does the user have the ability to opt-out of training and data collection?
More non-specific information is gathered, “analyze usage patterns” and service improvements, “similar tracking technologies”… what technologies?
No where are these services, save for Google, are actually listed.
I have concerns and questions regarding the model itself and its data.
The claims itself in the doc’s performance page are unrealistic at best, especially as the first model.
All the benchmarks appear to be entirely self-reported and aren’t verifiable with no indication of peer review, or any methodology.
We also don’t know where or how you’ve sourced the data, especially considering the admission of RAG and thus the possibility of generating code that someone else created--------- completely glossing over that despite mentioning fine-tuning and RAG, there’s no specifics regarding the actual training methodology or how these improvements were actually made.
Which brings me to assume that the code you generate and feed into the program is going to be used to train the RAG system, but we don’t know that because everything regarding data collection is as vague as I’ve previously shown.
Also this is. Huh. I don’t know what to say about this other than Huh. Safety and performance aren’t mutually exclusive. Safety should absolutely be taken account of especially in this current usecase. It’s Roblox. This is a kid’s gaming platform at its core. I’d sure hope that Roblox would be prioritizing the safety of it’s userbase as opposed to unregulated technological advancement.
At least to me, it feels like this point is being framed as being able to skip security checks and other moderation systems, despite the wording in the TOS that prohibits the generation of such content:
If anything, this entire project reads more like a marketing announcement while throwing in AI technobabble to sound reputable, while simultaneously making unsubstantiated claims and performance predictions YEARS out. This all feels very unserious and self-bolstering.
This is all just speculation and inference on my part, but none of this is giving a good impression.
If read carefully, these are “estimated internal benchmarks” for the upcoming BulletMindV1 LLM. A formal research paper, including peer-reviewed methodology and reproducible results, will be published to support these claims. Even releasing our own suite of benchmarking for AI engineers to use for themselves.
What do you mean by this? That’s not how RAG works. It’s not possible to feed RAG system. RAG is not an AI.
Also, most of the data collection, like HWID or usage logs, is strictly for preventing multiple account abuse, especially on the free plan where token usage can be exploited. We’re not interested in selling or misusing user data, it’s just to keep things fair and secure.
