Pushing the limits of AI in Scratch, Arduino and more. Proving big AI doesn't always need big supercomputers.

AI where you least expect it.

<not <limits of AI>>

Welcome to the StructureBot AI discussion forum. This forum is about pushing the limits of AI in Scratch and beyond.

Stay tuned!

Bye for now. Thank you!

What can you do in this forum?

• Feel free to discuss topics about pushing the limits of AI.
  
• Talk about the shortcomings, expected improvements and future ideas of SB.
  
• Share with everyone how you want the future of SBai to be like.
  
• Enjoy your time here!
  

Some links I'd like to share:
StructureBot AI studio: https://scratch.mit.edu/studios/37018650/
StructureBot Home (get the latest updates!): https://scratch.mit.edu/projects/1213855640/

Thank you!
Have a great day!

Nith04 wrote:

Pushing the limits of AI in Scratch, Arduino and more. Proving big AI doesn't always need big supercomputers.

StructureBot AI

AI where you least expect it.

<not <limits of AI>>

big AI sort of does need big supercomputers because bigger AI means a lot more computing is required which means you need bigger, stronger, computers to compute all of that needed computing… Not saying AI in scratch isn't possible, it just won't be very fast.

Cool!

MineTurte wrote:

Nith04 wrote:

Pushing the limits of AI in Scratch, Arduino and more. Proving big AI doesn't always need big supercomputers.

StructureBot AI

AI where you least expect it.

<not <limits of AI>>

big AI sort of does need big supercomputers because bigger AI means a lot more computing is required which means you need bigger, stronger, computers to compute all of that needed computing… Not saying AI in scratch isn't possible, it just won't be very fast.

I agree. But the mission of SBai is to get the most out of AI, in not just Scratch, but any limited platform, like with Arduino. It's a DIY electronic hobbyist company that creates these cool boards you can build projects with. They're not really known for being able to run any kind of AI locally, but that's what SBai is trying to achieve. Thank you for your activity in this thread!

Have a great day!

Nith04 wrote:

MineTurte wrote:

Nith04 wrote:

Pushing the limits of AI in Scratch, Arduino and more. Proving big AI doesn't always need big supercomputers.

StructureBot AI

AI where you least expect it.

<not <limits of AI>>

big AI sort of does need big supercomputers because bigger AI means a lot more computing is required which means you need bigger, stronger, computers to compute all of that needed computing… Not saying AI in scratch isn't possible, it just won't be very fast.

I agree. But the mission of SBai is to get the most out of AI, in not just Scratch, but any limited platform, like with Arduino. It's a DIY electronic hobbyist company that creates these cool boards you can build projects with. They're not really known for being able to run any kind of AI locally, but that's what SBai is trying to achieve. Thank you for your activity in this thread!

Have a great day!

I see. Well good luck! And yeah no problem.

What are you planning to do next??

That’s a really creative idea — using AI to help structure or guide projects instead of just generating them directly.
In Pishi.ai Scratch, we’ve been exploring something similar: extensions that let students learn AI concepts, logic, and data handling inside Scratch itself.
The goal isn’t for AI to build projects automatically, but to help learners understand how AI thinks — so they can design, test, and improve their own logic.
It’s like turning AI into a classroom assistant rather than a project builder.

Scracthviva wrote:

What are you planning to do next??

I believe I'll be working on the SBast (StructureBot Assistant) models area for a bit now, which are meant to be lightweight but good quality AI models for OS projects in Scratch.

Hi guys! Hope you're having a good day. I'm making this post to update the forum on the latest announcement for StructureBot AI: StructureBot Bits.

StructureBot Bits is basically an array of many mini AI models, specialized for specific, true purposes that may not belong as its own separate flagship model in the main studio. These special AI models, would be called “Bits” after the naming scheme. These bits will serve specific, true purposes. For example, models which are made from the latest StructureBot Pattern model, but is optimized for maybe… domains like study, creative writing, research, etc. Another noteworthy example is let's say suppose there was a mainline StructureBot AI model for gaming (which might happen! ), StructureBot Bits can contain “bits” for specific games like Robot Destructor, Taco Burp, and Diep.io (Scratch Port), (these are all great games btw, make sure to check them out!)
and etc. If my explanation wasn't very clear, no worries: you can definitely ask doubts following this post.

StructureBot Bits Studio: https://scratch.mit.edu/studios/51075029
StructureBot Bits Introduction: https://scratch.mit.edu/projects/1246039696/

Thank you for taking your time to read this post!
Hope you have a great day!

You might want to check out Adacraft for inspiration.

A question: and its answer.

Why does the recently released StructureBot Puppet3 look like it works very differently in code compared to other models like Pattern6?

Ans: StructureBot Puppet3 was made from Puppet2 which is an older model which works completely differently compared to the new complete-markov systems seen in SBptn6 and such. It uses a combination of words, rules, associations and a translation filter to smoothen a rough output.

Words
There is a big list of words in a legacy model, which is called its list of words. They are added manually.

Rules
“Rules” basically prohibit certain word combos and replace them with valid ones. For example: “I is” could be recognized and replaced by “I am” because “I is” is gramatically incorrect. They are also added manually.

Associations
Associations basically categorize the list of words into certain manually set groups like “nouns”, “verbs”, etc and using certain sentence structures like “noun verb noun” it picks random words that fit within these groups in the structure described.

Translation Filter
The Translation Filter uses the Translate extension to translate the output many times to regain a nicer output that removes rough effects and link words with each other to form a pseudo-sentence.

This system is quite manual, which is why they have been recently replaced by the Markov systems with the Pattern series. But this system was kept in Puppet3 along with the introduction of a Markov system without removing this manual system; because they contributed to the unique sentence generation provided by the StructureBot Puppet series. I hope this post has provided understanding about why Puppet3's code looks different from other models that have been released recently like Pattern6, MCQ, etc.