update

update

update

wow i tried it out! great!

P1N3APPL3_TESTS wrote:

wow i tried it out! great!

TY! Hope you enjoyed it!

update

update

update

update

update

I like how you put a whole lot of math equations and their solutions on the list of responses

happywaffle0 wrote:

I like how you put a whole lot of math equations and their solutions on the list of responses

? no i didnt
also how did you look inside unless you downloaded it and put it in turbowarp

austin_intel001jrahs wrote:

happywaffle0 wrote:

I like how you put a whole lot of math equations and their solutions on the list of responses

? no i didnt
also how did you look inside unless you downloaded it and put it in turbowarp

https://scratch.mit.edu/projects/1367999750/editor/

Its in the “Response Bank” list. it starts from index 648 to index 5414.

austin_intel001jrahs wrote:

I BUILT A 14,000+ BLOCK AI CHATBOT. ENTIRELY IN SCRATCH.
NanoChat Ultra Gen 1 V10.00 — FINAL STABLE RELEASE

►►► PLAY NANOCHAT ULTRA V10.00 NOW ◄◄◄

https://scratch.mit.edu/projects/1367999750/

No ChatGPT.
No API.
No internet.
No external AI secretly doing the work.

Everything runs inside Scratch.

NanoChat Ultra V10.00 is built from 14,000+ Scratch blocks working together as one conversational engine.

My goal was ridiculous:

How close can you get to the BEST AI chatbot realistically possible inside Scratch?

After months of rebuilding, testing and pushing Scratch far beyond what it was designed for, V10.00 is my answer.

It may genuinely be one of the most advanced conversational AI systems ever built entirely in Scratch.

Think that's an exaggeration?

Try to break it. :P


WHAT'S INSIDE V10.00?

• 622 structured knowledge entities + 1,870 aliases
  
• 11,323 spelling and normalisation entries — typos and messy wording don't instantly beat it
  
• 129 relation records for causes, connections, comparisons and other reusable reasoning patterns
  
• 54 extra fact units + 72 alternate examples — ask for another example and actually get another one
  
• Multi-fact responses — ask for several facts and NanoChat constructs multiple distinct grounded answers
  
• TEACH / QUIZ / DEBUG / SEQUENCE dialogue modes that remember your progress
  
• Conversation memory — “why?”, “tell me more”, “make it simpler”, corrections and returning to previous topics
  
• Clarification — ambiguous messages can trigger a question instead of a random guess
  
• Fallback recovery — if NanoChat gets stuck, it can explain the problem and suggest how to rephrase
  
• Built-in maths for arithmetic and expressions
  
• Safety filtering for inappropriate and dangerous requests
  
• Full custom chat UI — scrollable history, draggable scrollbar, pen-rendered text and Normal / Fast / Instant typing
  

THIS ISN'T THE USUAL SCRATCH CHATBOT.

A lot of Scratch chatbots basically work like this:

if <(answer) = [hello]> then
say [Hi!]
end

Then:

if <(answer) = [how are you]> then
say [I'm good!]
end

Then another if block.

And another.

And another.

NanoChat doesn't work like that.

Your message passes through systems for normalisation, intent detection, contextual retrieval, candidate competition, confidence scoring, dialogue planning, grounded response composition and episodic conversation memory before NanoChat decides what to say.

It isn't just looking for one exact sentence.

It's trying to work out:

“What is this person actually asking me?”

NanoChat is not an LLM, and I'm not pretending it is.

Instead, it uses a specialised AI-style architecture designed specifically around the limits of Scratch.

No billion-parameter neural network.

No server.

No hidden ChatGPT request.

Just an absurd amount of Scratch engineering.


DON'T JUST TYPE “HELLO”.

Try:

• Compare Earth and Mars.
  
• Teach me photosynthesis.
  
• Quiz me on gravity.
  
• Explain RAM step by step.
  
• Help me debug this.
  
• What causes rust?
  
• (10 + 3) * 2
  
• 2 ^ 8
  

Then follow up with:

“why?”

“tell me more”

“make it simpler”

“give me another example”

“what about Mars?”

That's where NanoChat gets much more interesting.


THE NANOCHAT CHALLENGE

Can you confuse it?

Make a typo.

Phrase something strangely.

Change topics halfway through.

Ask follow-up questions.

Give it something ambiguous.

Try to find an input I didn't think of.

If you break NanoChat, post a screenshot and the exact message you used.

I WANT the failure cases.

Every good failure gives me another weakness to fix.


Tip: Click the green flag and wait for READY.

Shift + click the green flag for Turbo Mode.


►►► TRY NANOCHAT ULTRA V10.00 ◄◄◄

https://scratch.mit.edu/projects/1367999750/

If NanoChat genuinely surprises you, love + favourite the project and follow for more slightly unreasonable Scratch experiments.

Remixes are welcome with credit.

And seriously.

Don't only type “hello”. :P

Chat text powered by Pen Text Engine++ by -Rex-. NanoChat is experimental and can make mistakes, so don't rely on it for medical, legal or financial advice.

cool can you check this out please https://scratch.mit.edu/projects/1396470576/

happywaffle0 wrote:

austin_intel001jrahs wrote:

happywaffle0 wrote:

I like how you put a whole lot of math equations and their solutions on the list of responses

? no i didnt
also how did you look inside unless you downloaded it and put it in turbowarp

https://scratch.mit.edu/projects/1367999750/editor/

Its in the “Response Bank” list. it starts from index 648 to index 5414.

WOW how did you find this thank you so much i just remembered that this is part of a legacy model that is now retired (V1.21), and I have forgotten to remove the unreachable list items! btw, this current V10 model DOES NOT use the list items to calculate math. It is now using an actual math engine (check out my scientific calculator) to do the equations. TYSM! i can now remove like 10000 list items