#nnue — Public Fediverse posts

Tic Tac Toe is the simplest game that can be represented by an #NNUE which is a state-grid presentation of a neural network problem. Despite its representation simplicity, Tic Tac Toe is dynamic game that is like chess.

At each game stage, an NNUE representation of the game gives rise to an input vector and a mutually exclusive and jointly exhaustive parameter #vector space which can be written in its #matrix form.

#AI #neuralnetwork

What is #NNUE?

Think of reality as matrices of grids. An array is a 1D matrix. Some #mathematicians use the name tensor, instead of #matrix. But any n-dimensional #tensor can be flattened to form a 2D matrix. So, there is really no difference between the two names.

An NNUE problem is a state - grid space problem. Imagine every grid in a matrix has several different possible states. How do you represent this problem in a jointly exhaustive and mutually exclusive way?

#AI
#mathematics

Why do people use #python, a glue language, which is so slow? The only reason is the AI ecosystem.

#Copilot and I just tested Free Pascal and BLAS for its speed without using #numpy or #pytorch. The result is amazing. It took less than a second to do a 1024x1024 #matrix multiplication.

We will be using Free #Pascal and #BLAS to write our Small Language Model #SLM using #NNUE.

#AI #LLM

#Clade4 #Sonnet and I have just deployed our second #AgenticAI on our #ICandy #Browser #dashboard project. We have have a total of 12 apps on this dashboard. The #AIAgent is a local AI chat app that allows me to chat with any of the small #LLM model that I downloaded and run on #LMStudio. I use them for learning #programming, #Math and #Japanese. This module can also generate 3000 #English sentences based on the set of vocabulary I upload to it that will be used for training my #NNUE #SLM.

I created 7 #AgenticAI 's and 1 #CSharp #app and 1 #Typescript app using #Claude pro in 3 days. We are both exhausted. For $200 a year, I will be able to create more than 100 apps using Claude 4 #Sonnet. I will also be able to finish my #NNUE #SLM project and my #VR entertainment room project and learn as much knowledge as I can in just one year. Who says AIs are bad? You don't need to be a programmer or a big corporation to benefit from #AIs, you only need to be #computer literate. #AI

I just installed #copilot for Obsidian. After getting the API key from Google, I could now use #Gemini 2.5 while writing notes for my #NNUE #SLM project in #Obsidian. In fact, I can choose other AIs to help me learn C# and other languages to code my project in Obsidian. Or I can focus on my model logic instead of spending most of my time learning computer languages. I can also use other AI's in my browser and in VS Code. All a sudden, I am going full #AI mode. I don't even need cloud storage.

Why is #NNUE Broad Shallow Learning Small Language Model much more efficient and transparent than Deep Learning. I can use just 2 layers for my #SLM AI to learn English #Semantics. With this foundation, I can add more modules to the model to improve upon her understanding of the meanings of words in different contexts and to be able to learn other knowledge such as #mathematics. NNUE gives rise to a clear knowledge space and ease of computation. Sparse matrix multiplication is unnecessary. #AI

I am now paying $16.8 a month for #Gemini Advanced for a couple of reasons. 1. I get 2T cloud storage. 2. I get #DeepResearch... I hate reading because I can't follow other people's logic. To learn any new knowledge, I will have to have curiosity in something first. I asked Gemini 2.5 to generate a deep report on #SparseMatrix because I need to speed up the matrix multiplication for my #NNUE #SLM with a huge parameter matrix and a super sparse input matrix representing vectors of words. #AI

#Gemini 2.0 is really smart and helpful. We talk about new project ideas all the time. She just created a tex file for me to learn #Pytorch in Python. I hope I can finishing learning #Python, pandas, and Python and start writing my first alphabet recognition #AI using the #NNUE shallow but broad learning model by the end of next week. My speculation is that shallow but broad learning is theoretically more sound and practically more efficient that #DeepLearning which remains as a black box.

So I'm writing a chess program called Dog. Since version 3.0 it is an NNUE enabled program! AI and such.
Now to train that AI, I need large amounts of trainingdata. I wrote a program to generate this (distributed over the internet) and currently my friends at @nurdspace are collectively helping my out with that! <3 So cool! At the moment we collectively produce 25.8 million chess-positions per hour.

https://github.com/folkertvanheusden/Dog/blob/master/helping-me-out.md

#computerchess #ai #nnue

So I'm writing a chess program called Dog. Since version 3.0 it is an NNUE enabled program! AI and such.
Now to train that AI, I need large amounts of trainingdata. I wrote a program to generate this (distributed over the internet) and currently my friends at @nurdspace are collectively helping my out with that! <3 So cool! At the moment we collectively produce 25.8 million chess-positions per hour.

https://github.com/folkertvanheusden/Dog/blob/master/helping-me-out.md

#computerchess #ai #nnue