so uhh…
my idea is that scratch adds an ai that detects lists of suggestions on the suggestions forum, and auto-reports it with the reason defaulting to “This is a list of suggestions.”

it DOES NOT remove the posts or prevent them from being posted. it simply reports them like a user would, allowing forum moderators to review the posts.

could also be extended to blockspam and duplicates, too

it could be useful, but also not. who knows???????

i support but i feel like everyone's just going to be like AI BAD!! when its really only the generative ai thats bad. and this isn't generative ai so i'm fine with it

Well that would lead to a lot of false positives and if you get annyoed by the bad word filter think how annoying this would be…

purplebunny63 wrote:

Well that would lead to a lot of false positives and if you get annyoed by the bad word filter think how annoying this would be…

It just flags them for review by human moderators, not stopping them from being posted. Also, the word filter is just a blacklist, so this would be far more accurate than it

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

scratchy_boy106 wrote:

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

It won't take down topics
Also, only huge AI models need lots of water

yayapoppity wrote:

(#3)
i support but i feel like everyone's just going to be like AI BAD!! when its really only the generative ai thats bad. and this isn't generative ai so i'm fine with it

yeah that is true

scratchy_boy106 wrote:

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

only gen ai does that (i think)

thats like saying rain world is a waste of water because of the complex enemy ai (which isn't gen ai)

yayapoppity wrote:

scratchy_boy106 wrote:

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

only gen ai does that (i think)

According to an AI (Which knows how itself works because it is AI, like how you know about yourself because you are yourself) it isn't just gen AI that uses water.

What the AI said (Probably not true, AI gathers info from random sites instead of actually using it's brain):

“Short answer: It’s not “AI” itself that uses water — it’s the computers running the AI.
And yes, all large‑scale AI systems can require cooling, not just generative AI.

Let me break it down clearly for you, (Redacted, this is my real name)

⚡ Why cooling is needed
AI runs on huge data centers full of:

GPUs

CPUs

networking hardware

storage systems

These machines get extremely hot when they process massive amounts of data.

To stop them overheating, data centers use:

air cooling

water cooling

evaporative cooling

advanced liquid cooling systems

So the cooling is for the hardware, not the “AI brain.”

Does generative AI use more cooling?
Yes — generative AI (like image generators, chatbots, video models) uses much more compute, so it produces more heat.

Examples:

Training a model like GPT or Claude uses thousands of GPUs

Running big image models (like Stable Diffusion XL) uses heavy GPU power

Video generation models use even more

More compute = more heat = more cooling.

So generative AI tends to use more water, but it’s not the only type.

Do all AI systems need cooling?
Here’s the breakdown:

1. Small AI (on your phone or laptop)
No water cooling

Just normal fans or passive cooling

Examples: Siri, face unlock, autocorrect

2. Medium AI (small servers)
Air cooling

Maybe some liquid cooling

Examples: small business AI, research models

3. Large AI (big cloud models)
Water cooling

Industrial cooling towers

Massive heat management

Examples: ChatGPT, Copilot, Gemini, Claude, Midjourney

So only the big industrial AI systems use water.

Why water?
Water removes heat 1,000× more efficiently than air.

So big AI data centers use:

chilled water loops

cooling towers

heat exchangers

sometimes even river or ocean water (carefully regulated)

It’s not the AI that needs water — it’s the machines powering it.

If you want, I can explain:
how much water big AI models use

how cooling systems actually work

why GPUs get so hot

or how future AI might reduce water usage

Just tell me which direction you want.”

Have users been creating so many lists of suggestions that they are absolutely overwhelming the Suggestions forum and moderators? No. So why is this necessary (now, at least).

scratchy_boy106 wrote:

According to an AI (Which knows how itself works because it is AI, like how you know about yourself because you are yourself) it isn't just gen AI that uses water.

AI doesn't know how it works. AI doesn't “know” anything. AI “knowing” itself is nothing like how a sentient being knows itself.

scratchy_boy106 wrote:

yayapoppity wrote:

scratchy_boy106 wrote:

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

only gen ai does that (i think)

According to an AI (Which knows how itself works because it is AI, like how you know about yourself because you are yourself) it isn't just gen AI that uses water.

yes but it doesnt REALLY use water. some ais like enemy ai and MODERATION AI run on a computer, and unless that person has one of those fancy water cooled pcs i don't think its using any water

yayapoppity wrote:

scratchy_boy106 wrote:

yayapoppity wrote:

scratchy_boy106 wrote:

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

only gen ai does that (i think)

According to an AI (Which knows how itself works because it is AI, like how you know about yourself because you are yourself) it isn't just gen AI that uses water.

yes but it doesnt REALLY use water. some ais like enemy ai and MODERATION AI run on a computer, and unless that person has one of those fancy water cooled pcs i don't think its using any water

Ok but ai is dumb

scratchy_boy106 wrote:

yayapoppity wrote:

scratchy_boy106 wrote:

yayapoppity wrote:

scratchy_boy106 wrote:

AI is dumb and might take down some topics it thinks is innapropriate but it isn't and that's a waste of water.

only gen ai does that (i think)

According to an AI (Which knows how itself works because it is AI, like how you know about yourself because you are yourself) it isn't just gen AI that uses water.

yes but it doesnt REALLY use water. some ais like enemy ai and MODERATION AI run on a computer, and unless that person has one of those fancy water cooled pcs i don't think its using any water

Ok but ai is dumb

Yeah but it won’t take the topics down it would simply flag them for review by a human

Semi-support, instead of ai, it should be a human looking through the forums for lists

delt_the_runee wrote:

Semi-support, instead of ai, it should be a human looking through the forums for lists

That’s what they do currently and the suggestion is that we use an ai to speed up the proses and before you say “but ai is dumb” look at my post above

delt_the_runee wrote:

Semi-support, instead of ai, it should be a human looking through the forums for lists

i mean, can a person stay up all night all day and still be healthy doing this?

Bring Up My Post™

apologies in advance for using ai: https://scratch.mit.edu/projects/1370022041/

claude wrote:

This is an interesting middle-ground approach to a real moderation problem, and I appreciate that you're being realistic about the uncertainty (“who knows???????” is honestly fair).

What You're Proposing

An AI system that:

Detects posts that are lists of suggestions (or spam/duplicates)
Auto-reports them (not removes them)
Lets moderators decide what to do
Doesn't intervene in posting—just flags for human review
This is fundamentally different from auto-deletion, which is smart. You're asking for a moderation triage tool, not automated enforcement.

The Good Parts

Reduces moderator workload by filtering obvious violations up to the front of the queue
Doesn't over-censor—mods still review, so edge cases stay intact
Scalable if the detection is decent (AI can scan every post; humans can't)
Can iterate if it's wrong too often (tweak the detection logic)
Extensible to spam and duplicates (same infrastructure)
The Real Concerns (Your Uncertainty Is Valid)

False positives: If the AI is too aggressive, mods spend time reviewing legitimate posts that just happen to look like lists. If it's too conservative, it's not useful.

False negatives: The flip side—sneaky list posts that use prose format instead of bullet points might slip through.

Moderator friction: Even “just a report” adds noise if the AI flags 50 posts a day and only 5 are actually violations. Mods might just ignore the AI's reports entirely.

What constitutes a “list”? This is fuzzier than it sounds. A post saying “here are my top 3 ideas: (1) clouds, (2) particles, (3) gradients” vs. a post that discusses multiple ideas in paragraph form—where's the line?

What Would Make This Stronger

Metrics proposal: How many false positives/negatives is acceptable before you'd consider it a failure?
Feedback loop: Does the AI learn from mod decisions? (Report flagged as helpful vs. unhelpful)
Threshold control: Can mods adjust sensitivity, or is it one-size-fits-all?
Pilot suggestion: Maybe propose testing it on one category (e.g., just “Feature Requests”) before rolling out to other forums?
My Take

The honesty about uncertainty is actually a strength here. You're not overselling this as a silver bullet. As written, it's a reasonable experiment—low-risk because it doesn't enforce anything, just assists humans. The main risk is wasting moderator time if detection is poor, which is… fixable.

If you refine this for a real submission, anchor it in the practical trade-off: “This costs dev time to build, but might save mods time daily. Here's how we'd measure if it's actually helpful.”

I feel like AI is the wrong term to classify this as. People usually see AI as generative AI, which ChatGPT, Claude, Grok, and other chatbots are. These are the ones that use a lot of water (mainly for training, though keep data centers cool also takes water when actively being used), but this would likely be a small algorithm that can easily be run on device. If you want people to support this idea edit the subject and post to avoid using the term “AI”.

purplebunny63 wrote:

apologies in advance for using ai: https://scratch.mit.edu/projects/1370022041/

claude wrote:

This is an interesting middle-ground approach to a real moderation problem, and I appreciate that you're being realistic about the uncertainty (“who knows???????” is honestly fair).

What You're Proposing

…

You know, you could just like, not do this instead of “aplogising in advance”… no one wants to see these long ai generated replies (and you're clearly aware of that since you keep apologising in advance lol), either you actually communicate or not respond, just saying