opt-in for existing assets, for new assets you either have to a. both 1. know about the existence of a checkbox and 2. click it every time, or b. both 1. know about the existence of the opt-out buttons (several) and 2. find them deep in the jungle of the new develop page
(this is what in the ui design world you might call a dark pattern 
)
To ensure the quality of our AI models, we employ various quality control measures around data filtering to only train on what we believe is high quality data, and then verify our results with evaluation frameworks.
For example, both Code Assist and Assistant have evaluation suites where we run the output through a series of unit tests to verify correctness.
Some of the data quality checks include safety checks, deduplication, and both model and heuristic-based methods to filter out low-quality code. Additionally, we conduct rigorous A/B testing when launching new model versions to ensure we are continually making the tools better for you.
As others mentioned, I don’t believe it could work in any other way if you like publishing free assets. From what I understood, if you published a model, even if they gave you the option to opt out on free models, as soon as someone uses that model in their project, if they don’t opt out too your model would be shared anyways, so it doesn’t look like a realistic approach to hope for Roblox not using free models and assets to train AI.
Just wanted to reply because I love your work and wouldn’t want you to stop contributing. they way you have for the past years!
Why not improve existing AI tech, like your horrendous moderation? Or maybe improve on the algorithm used to show experiences? Or maybe comb through the endless issues that plague the engine on the daily instead of leaving reports to rot.
You can claim this assists people all you want, but all it’s doing is enforcing this idea that artists aren’t necessary anymore. Who needs a texture artist when you can click a button to generate a messy rock texture that doesn’t look correct? Why hire a coder when you can wait for the code assistant to generate a function that doesn’t even have proper variables listed?
This company genuinely makes no sense on what their priorities are, and it seems like it’ll continue to double down. A shame.
Thank you for your kind words.
I’d assume that this would be the case for paid assets too though? If I create paid assets for customers to use in their games, and that is opted-in, then wouldn’t my work (that I opted-out) also be fed into the training data?
This reasoning makes it sound like nothing will be safe from becoming training data, regardless of the original creator’s sharing settings, because someone else can opt your work in without your consent?
Paid plugins currently available on the Creator Store aren’t part of training. Beyond that, paid models, decals and meshes don’t exist yet, and I suspect this is part of why. I imagine paid assets would necessitate more work being done on package permissions and read-only packages, especially if they want to uphold licences in other scenarios anyway such as licences that don’t grant derivative works rights.
It is disingenuous to pretend what this user is saying is that argument. You’re both saying different things.
What you are talking about is the definition of “disruptive technology”. Its been a long time since humanity had one of these, they’re uncomfortable and by definition will disturb the status quo, but we’ve gone through several of these through history (see: electricity, cellphones, blockchain, uber, home food delivery, even the wheel). There is no way to put this genie back in the bottle. It’s better to embrace it as an inevitability and try to steer it, rather than panic every time someone new adopts it.
This. We don’t have the power to turn back time, but we do have the power to shape the future into an image that ensures better outcomes for others. That’s exactly the philosophy I take with this.
We have technology that’s now automating away a lot of tasks. Some of these tasks were jobs for some people. That’s the new reality we’re in. What we can do now, though, is hold those with the technology accountable and ensure that they’re paying attention to the law and to moral and practical issues that affect everyone impacted by it. We need to be productive and we need to actually do things, rather than wishing it were different.
Don’t lament we’re here. Always push for a positive next step.
Yeah, I doubt it would be of any use to third parties. All of that data is Roblox-specific, isn’t it?
I swear I just had a post up here and it’s gone??? Let me retype it.
EDIT: Original post was removed for profanity.
Free Creator Store Assets (Meshes, Images, Models) Included in generative AI training, no ability to set data preference
My games community relies on making their models temporarily set to ‘Distribute on Creator Store’ in order to be able to whitelist their content to be loaded in my game, before they switch it back off.
This is required in order for the game to be able to load the community’s content.
Are you telling me that even if they only distribute their model for 60 seconds in order to get their content whitelisted into my game, that they are all FORCED into this program even though that is not what they want to do with their content???
CommunitySift, the tool Roblox uses to check every single chat message, asset, description update, etc is improving every day and has been for over a decade at this stage.
In the last 3 months of 2023, Roblox dealt with 204.431 billion assets, across 98 days that’s on average 2.086m assets. Across specifically “Threats, bullying, and harassment”, “discrimination, slurs, and hate speech”, and “terrorism and violent extremism” (which is what Roblox is legally to report on) they got 12.8m flags and 98.714% was actioned on and 2.22% of appeals on review was accepted
Roblox needs to work on moderation but please don’t kid yourself of the problems that working on the scale of Roblox, the models used for moderation, recommendations, or code are very different and can be handled. For any bug reports, AI is not the bottleneck, it’s having the humans to identify, verify, and fix any issues reported. GenAI will only ever produce the most mid of mid code, I don’t want mid code.
Take a look at all of the reports, don’t expect much better Terms of Service Reports | State of California - Department of Justice - Office of the Attorney General
I say all that while having not opted into the AI data feature because I don’t want my private work to be part of that system, but I want to be realistic about what’s actually going on and what we can do to steer the issue to solve this massive genie out of the bottle
For those who are concerned, Roblox has already been using free assets to train their AI, this is nothing new. Now they’re just being more transparent about how your data can be used.
If you distribute a free asset on the store, anyone is allowed to do whatever they want with it, this has always been the case. That includes Roblox.
Kinda starting to feel less comfortable sharing my assets publicly now…
I will NEVER give consent for my work to be used in AI training.
This will be a pretty serious problem for UGC creators who don’t want to provide their assets to train AI. Although it is appreciated that they give the possibility of deactivating this function, it will be something tedious to do for each new experience that we create from now on. But what can we do
I have a question about this new function when bringing a model into the studio and having the AI option activated, will the model be automatically provided to the AI if we do not deactivate this function? And if this is true, once the AI has the model, if we disable this feature, will the AI no longer have access to the model?
To give an example, if from now on I carry a model like this of the image of a shark, if I do not deactivate the function and bring it into the studio, will the AI automatically have the model?
Don’t worry, my friend! According to the lawyers, you already have.
That’s amazing, but let’s consider how I was called a slur and that super-amazing AI model that got to 98% actioned flags didn’t even attempt to ban this user. Or maybe we can take a stroll through the marketplace and look at the lengthening amount of flaggable assets still up and being purchased today.
Anyway, I strongly believe that I am within my right to continue to be agitated at this company’s enforcement of useless features and updates such as this when the other 99% of their engine, website, and various features are either half-baked, unimplemented, or just plain not working. Whether or not there’s different teams does not change my opinion in the slightest.
I did not work my butt off for years just to be replaced by a half-baked AI model.
If I hated Roblox and everything it did and did not trust them at all, I would simply not use the platform. It feels foolish to use a platform if you don’t trust them to not literally steal from you.
Just a general announcement in response to reading through what’s been said on this thread. Not aimed at anyone in particular.
I feel like this is a weird take when you contrast it with companies that have done far, far less in the name of transparency around data usage and who have buried their data sharing preferences in dark corners of the app - if they’re even there at all.
Roblox have been more upfront than 98% of companies I’ve seen. They’re even willing to sit one of their historically biggest critics down in meetings and talk freely. They’ve more than earned my trust.
Dear roblox, NOBODY WANTS TO BE AUTOMATICALLY OPTED IN
Is it just me or have Roblox’s latest decisions been a bit… odd
Edit: I can just smell the lawsuits brewing. Seriously though, they put the buttons for this in a very obscure place that i’m sure a ridiculous amount of people will never look, which will probably result in Roblox being sued
“Just stop using ROBLOX if you don’t like it!”
Why can’t the billion dollar company be responsible for making sane choices like not enforcing data collection for AI generation unless you’re chronically online. No, surely it’s the user’s fault for joining in the first place!
