01
Tokens
Text isn't read by words. It's chopped into tokens: syllable-sized fragments the model treats as its atomic units of meaning.
Vocabulary
Neural.Literacy
Neural.Literacy
A field guide · from zero to building
Neural.Literacy is an interactive field guide that turns the black box of modern AI into something you can actually see, name, and steer. No hype. No jargon wall. Just clarity, by design.
Each concept is a door. Open them in order and the whole machine starts to make sense, not as magic, but as a sequence of surprisingly human ideas.
Text isn't read by words. It's chopped into tokens: syllable-sized fragments the model treats as its atomic units of meaning.
Every token becomes a point in high-dimensional space. Similar meanings sit close together, so geometry becomes language.
For each word, the model looks back at everything before and decides what to pay attention to. It's reading with a moving spotlight.
A prompt is a contract. The clearer your context, constraints, and examples, the more reliably the model honors it.
The model's working memory. Anything outside the window is forgotten, so what you include is the whole world it can reason about.
The model doesn't know, it predicts. When it's unsure, it can sound perfectly confident and be perfectly wrong.
Start at Zero Knowledge. Each level unlocks the next. Go all the way and you'll understand the full stack, from tokens to agents.
AI vs ML vs LLMs, how AI "learns," hallucination, the models, and tokens. The un-confuser.
01How AI generates an answer, the 80/20 of prompts, temperature & top-p, context windows, system prompts.
02The CRIC framework, few-shot & chain-of-thought, tool calling, memory, RAG, multi-model strategy.
03API vs web vs CLI, inference, the provider landscape, quantization, open vs closed source, the tool loop.
04Embeddings, vector databases, multi-agent systems, and fine-tuning. The deep end.
⚡Applied AI: research, coding, content, and business automation. Theory is cheap; execution is everything.
There are hundreds of AI guides on the internet. Most of them follow the same shape: definition, then example, then done. You read it, you forget it, and you go back to ChatGPT doing the exact same thing. Neural.Literacy is different for three reasons.
Level 0 to 1 to 2 to 3 to 4. Each level builds on the one before. You don't skip around. You level up. By the time you reach multi-agent systems, the foundation is already there.
Other guides teach you how to use AI. We teach you how to use AI without making embarrassing mistakes. From "don't trust a confident answer blindly" all the way to "never hardcode an API key into code you push to GitHub."
You don't just learn AI. You learn how to make AI work for you. Persistent memory, tools, skills, multi-agent orchestration. This is the part other guides don't have, because most of them stop at "type a good prompt."
A toy tokenizer. Type a sentence and watch it split into the fragments a model actually sees.
Hermes is the messenger, and your interface to the model. These twelve moves turn vague requests into reliable results. Internalize them and prompting stops being luck. Read the full Playbook →
"You are a senior editor." Anchoring a persona narrows the output space dramatically.
One sentence. What does success look like? If you can't say it, the model can't either.
Audience, constraints, prior decisions. The model only knows what you tell it.
Two or three input→output pairs teach patterns faster than any instruction.
Bullet list, JSON, table, 200 words. Specify the container before the content.
Direct, playful, technical. Tone is a lever, so pull it deliberately.
"Don't mention pricing." Telling it what not to do is as powerful as the reverse.
"Reason step by step before answering." Thinking out loud improves the answer.
Refine in-thread. "Make it shorter" beats rewriting from scratch every time.
Treat outputs as drafts. Check facts, citations, and code before you ship.
Long threads drift. Summarize and restart fresh when context gets noisy.
The field moves weekly. Keep a finger on what's real versus what's marketed.
You are a patient technical writer. Audience: designers new to AI. We're shipping a 3-button feature. Explain what a context window is in one paragraph. Use an everyday analogy. End with a 5-word summary. No jargon. No more than 80 words.
The tools will change every quarter. The mental models won't. Learn the model beneath the models, and you'll never be at the mercy of a release note again.