Come with a question. Leave with a better one.
AI Dojo exists to help people understand AI and discover what they can make possible.
Type what you are wondering about. The typewriter looks through what AI Dojo has written and hands you the closest papers.
New here? Start with what happens when you talk to a chatbot 8 min read
Or try a first line:
This searches what AI Dojo has written, here in your browser. Nothing is generated, and what you type is not sent anywhere.
Start here if you are new
Five papers for a first visit. Type a question and the typewriter swaps them for the closest matches.
Everything in the drawer
Every paper the typewriter can find, laid out in full. Rather browse than type? Start anywhere.
You have read 0 of 52 pieces here. Saved in this browser only.
Long reads (5)
Explainers (8)
Questions people ask (12)
- Does it actually understand anything?
- Will AI take my job?
- Is it safe to paste work documents into a chatbot?
- Why do I get a different answer when I ask the same question twice?
- What's the difference between ChatGPT, Claude, Gemini and the others?
- Do I need to learn to code to use AI well?
- Why do chatbots sometimes miscount the letters in a word?
- How do I know if an answer is right?
- Is it conscious?
- What should I learn first?
- How much energy does using it take?
- When a model shows its reasoning, is that how it reached the answer?
Fine-tuning (21 papers and a reference list)
- Fine-tuning, every concept
- Should I fine-tune a model?
- Supervised fine-tuning (SFT)
- LoRA (low-rank adaptation)
- QLoRA
- Adapters, and merging them
- Distillation
- Chat template
- Loss masking
- Held-out test set
- Overfitting
- Catastrophic forgetting
- Learning rate
- Epochs and steps
- RLHF, step by step
- DPO (direct preference optimization)
- GRPO and verifiable rewards
- Reward hacking
- LLM-as-a-judge
- Quantization
- Continued pretraining
- Synthetic data
Exercises to try with a chatbot (6)
History
Glossary (24 words)
- Agent
- Alignment
- Attention
- Benchmark
- Context window
- Embedding
- Evaluation
- Fine-tuning
- Hallucination
- Inference
- Large language model (LLM)
- Multimodal
- Neural network
- Parameter
- Pretraining
- Prompt
- Reinforcement learning from human feedback (RLHF)
- Retrieval-augmented generation (RAG)
- Sampling
- System prompt
- Temperature
- Token
- Tokenizer
- Transformer
Not sure what to type? Five ways in
Brand new and curious
For people who keep hearing about AI and want to understand it without jargon.
- What should I learn first?
- What happens when you talk to a chatbot
- Token
- Does it actually understand anything?
- Same question, two answers
Planned a guided beginner path through the explainers, and beginner sessions. Neither is available yet.
Already using AI tools
For people who use a chatbot or AI assistant at work or at home and want better, more reliable results.
- Why do I get a different answer when I ask the same question twice?
- How do I know if an answer is right?
- Why models make things up
- Is it safe to paste work documents into a chatbot?
- Test it on what you know
- Spell it out first
Planned more practical methods pieces and hands-on workshops. Neither is available yet.
Building with AI
For developers, analysts and makers who connect models to data, tools or products.
- Prompting vs retrieval (RAG) vs fine-tuning
- Should I fine-tune a model?
- What happens when you fine-tune a model
- Will this model fit on my GPU?
- Why is my model slow?
- How do we serve a model?
- What people mean by "AI agent"
- Context window
- Answer from this page only
- Test your prompt on five cases
Planned build notes and small published experiments. Not available yet.
Research or deep expertise
For researchers, engineers and anyone who reads papers and wants the sources and the open questions.
- The "Going deeper" notes in the long read
- What happens when you talk to a chatbot
- When a model shows its reasoning, is that how it reached the answer?
- Does it actually understand anything?
- Is it conscious?
- Fine-tuning, every concept
Planned research explainers and conversations. Not available yet.
Deciding, teaching or leading for others
For managers, teachers, parents, policymakers and anyone choosing how others will use AI.
- How to judge an AI claim in the news
- Will AI take my job?
- Is it safe to paste work documents into a chatbot?
- How much energy does using it take?
- What's the difference between ChatGPT, Claude, Gemini and the others?
- Test your prompt on five cases
Planned workshops for teams and groups. Not available yet; no dates, prices or formats exist.
Why this desk exists
A lot of what gets written about AI either sells it or warns against it, and both tend to skip the part where you find out how it works. AI Dojo exists so that anyone, whether a first-time user or a researcher, has somewhere to understand what is happening and decide what to do with it.
AI Dojo is an independent place to understand artificial intelligence. It explains how AI systems work, where they help, where they fall short and what people can make with them. Explainers start in plain language and go further for readers who want more. You can come to read, to try an exercise or to think a question through, and you are just as welcome whichever you choose. What you read here is free, and stays free. Courses may come later; they will never sit in front of it.
Read why AI Dojo exists, its standards and where things stand
Planned, not here yet
AI Dojo plans to offer courses, live sessions and workshops for people who want guided practice. None exist yet. When they do, they will be there to support the free material on this site, which stands on its own and will stay free to read.
- Planned Courses
- Planned Live sessions
- Planned Workshops