LLMAIX2001a

LLMAIX2001a is the flagship ABC4RD Academy course and product path: a practical route from language-model foundations to a Storyteller AI prototype, AI agents, AGI/ASI literacy, and blockchain-backed trust.

The repository is built around one principle: learners should be able to run something real before reading a grand strategy.

Current Public Artifact

Status	Artifact	What it proves
active	Module 01 Bigram Lab	Runnable Python, toy corpus, expected output
active	Module 01 tests	Helpers are testable and deterministic
draft	Module map	The course can grow one reviewable unit at a time
draft	Innovation agenda	Agents, AGI/ASI literacy, and blockchain trust stay scoped

Run The First Lab

From the repository root:

python labs/module-01-bigram/bigram.py

Try a second deterministic sample:

python labs/module-01-bigram/bigram.py --start t --length 80 --seed 11

Run tests:

python -m unittest discover -s tests

Learning Path

Stage	Topic	Outcome
01	Bigram language model	Understand next-token prediction
02	Micrograd	Build intuition for gradients and backpropagation
03	N-gram and MLP	Move from counts to neural networks
04	Attention	Learn context, softmax, and positional encoding
05	Transformer	Implement a small GPT-style model
06	Tokenization	Build a minimal BPE tokenizer
07	Optimization	Learn initialization, AdamW, and training stability
08	Device optimization	Understand CPU, GPU, and accelerator tradeoffs
09	Precision optimization	Learn fp16, bf16, and mixed precision concepts
10	Distributed training	Study DDP and ZeRO-style scaling ideas
11	Datasets	Prepare, load, inspect, and document datasets
12	Efficient inference	Add key-value cache concepts
13	Quantization	Reduce model size for inference
14	Fine-tuning I	Explore SFT, PEFT, LoRA, and chatbot development
15	Fine-tuning II	Study RLHF, PPO, and DPO at a literacy level
16	Deployment	Build APIs and a small web application
17	Multimodal AI	Explore image-text workflows and diffusion ideas

Storyteller AI Direction

The Storyteller AI prototype should grow from small, inspectable pieces:

• toy language models before neural networks;
• small neural labs before large model claims;
• reproducible datasets and expected outputs;
• safety and limitation notes before demos;
• agent workflows with dry-run and approval gates;
• blockchain-backed credentials and provenance as a research track, not a marketing claim.

Frontier Tracks

Track	Scope
AI agents	Tools, retrieval, sandboxing, observability, approval gates
AGI literacy	Definitions, evaluations, preparedness, governance
ASI safety literacy	Alignment, oversight, red-team practice, staged deployment
Blockchain trust	Credentials, provenance, attestations, contributor incentives
Open compute	Efficient training, inference, hardware literacy, energy context

ABC4RD Academy does not claim that this repository has built AGI or ASI. These topics are taught as research literacy, safety framing, and governance context.

Contributor Path

Good first contributions should be small and reviewable:

1. Run Module 01 and confirm the expected output.
2. Improve one explanation in labs/module-01-bigram/.
3. Add one glossary term.
4. Add one learner exercise.
5. Add one unit test for a helper function.
6. Convert one draft module into a runnable lab plan.
7. Open one precise issue with a source link and expected output.

Start with:

• Module 01 lab
• Exercises
• Glossary
• Contributing
• Flagship backlog

Public Activity

• Roadmap
• Module map
• Innovation agenda
• Publishing safety
• GitHub discussions
• Community content calendar

Claim Discipline

Public wording should be source-backed and should not imply endorsement, partnership, safety-critical deployment, investment advice, medical advice, or AGI/ASI capability unless there is explicit public evidence and approval.

Community

• Website: abc4rd.org
• Main hub: ABC4RD
• Discussions: ABC4RD Discussions
• X: @academy_abc4rd
• Telegram channel: abc4rdchannel
• Telegram chat: abc4rdchat
• Telegram bot: abc4rd_bot
• Discord: ABC4RD invite