Скачать или смотреть I Built My Own AI That Runs Locally on My CPU

In this video I have demonstrated the process of creating an AI, by the use of Machine Learning in Python. The three learning methods i.e. Supervised Learning, Unsupervised Learning and Reinforcement Learning is explained and demonstrated.

The AI created for determining the category of text input by user is made with the help of ChatGPT. The model created in the video is of few thousand parameters far away from modern day LLMs with hundred's of billions of parameters and huge datasets for training.

The Prompt Used in the Video is:

You are an expert AI researcher and senior Python engineer.

I want you to design and write a SINGLE-FILE Python program that demonstrates how AI works locally on a CPU using PyTorch.

IMPORTANT CONTEXT:
This is an educational AI system, not a ChatGPT clone or large language model.
The goal is to clearly demonstrate core AI learning concepts in a realistic, honest way.

GOAL:
Build a local AI system that:
1. Runs fully on CPU (no GPU, no cloud)
2. Uses PyTorch for model definition and training
3. Demonstrates:
Supervised Learning
Unsupervised Learning
Reinforcement Learning
4. Accepts user feedback (correct / incorrect)
5. Saves learning progress to disk and improves over time
6. Uses Wikipedia as a lightweight text data source

STRICT REQUIREMENTS:
Single Python file only
CPU-only execution (explicitly force CPU usage)
Use PyTorch for learning models
Allowed libraries: torch, numpy, sklearn, wikipedia, pickle/json
No heavy pretrained models
Code must be readable and heavily commented
Explain intuition, not math-heavy theory

DATA REQUIREMENTS:
Fetch small text samples from Wikipedia using the wikipedia library
Use limited text to avoid heavy compute
If Wikipedia fetch fails, fall back to built-in sample text

LEARNING IMPLEMENTATION DETAILS:

1. SUPERVISED LEARNING:
Simple text classification (e.g. topic/category)
Convert text to basic numerical features (e.g. TF-IDF)
Train a small PyTorch model
Allow the user to input text
Ask the user whether the prediction is correct
If incorrect, update the model using the feedback
Save updated model weights to disk

2. UNSUPERVISED LEARNING:
Cluster Wikipedia text into groups
Use clustering (e.g. KMeans)
Print discovered clusters and example texts
Explain what the AI learned without labels

3. REINFORCEMENT LEARNING:
Create a simple environment (e.g. guessing correct topic)
Use rewards (+1) and penalties (-1)
Implement a simple Q-learning loop
Save Q-table to disk
Show improvement across runs

FEEDBACK SYSTEM:
After predictions, ask the user:
“Was this correct? (y/n)”
Use this feedback to:
Retrain supervised model
Update reinforcement rewards
Persist learning so the AI improves over time

INTERACTION REQUIREMENT:
The AI must accept user questions via console input
For each question:
The AI must generate an answer or prediction
Clearly print its answer
Ask the user: “Is this answer correct? (y/n)”
If the user says “yes”:
Reinforce the current model or policy
If the user says “no”:
Ask the user for the correct answer
Use that correction to update supervised learning
Apply negative reward in reinforcement learning
Save all updates so future runs improve answers


CPU USAGE:
Explicitly set PyTorch to use CPU
Optimize for CPU usage (batch size, small models)
Clearly print that the AI is running locally on CPU

STRUCTURE:
Clear separation of sections
main() function
Reusable helper functions
Clear console output showing learning progress

OUTPUT:
Console-based interaction
Show predictions, learning updates, and saved progress
Print limitations honestly

FINAL STEP:
After the code, explain:
What kind of AI this is
What it can and cannot do
How supervised, unsupervised, and reinforcement learning work
Why large AI models need GPUs but this one does not

Now generate the full Python code and explanation.

--- END OF PROMPT -----


Make sure to like, share and subscribe!
Thanks for Watching!

#python #ai #artificialintelligence #chatgpt #machinelearning