Training versus inference, clearly
Learn what happens when a model is trained, what happens when you use it, and why it matters
A taste of a lesson
I corrected the chatbot and it said thanks, I'll remember that. Has it learned?
Within that conversation, yes in a sense: your correction is now in the context, so it can use it in its next replies. That is in context learning. But its weights did not change. Start a new chat and the correction is gone, unless the app has a memory feature that saved a note, which is a stored text snippet, not training. Your chat might be used to train a future version, depending on the product and your settings. True or false: correcting it today improves it for everyone tomorrow?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Explain the difference between training and inference with correct vocabulary
- Describe the stages of training from pretraining to preference tuning
- Explain in context learning and why it is not the same as training
- Read a product's data policy and explain what it means for future training
Lesson plan
- 1 Weights: what a trained model is Understand parameters as the learned numbers that define a model's behaviour. Start
- 2 Pretraining: learning from vast text See how next token prediction on large data builds general ability. Start
- 3 Post training: becoming an assistant Learn how fine tuning and preference tuning shape behaviour. Start
- 4 Inference: what happens when you ask Understand inference as a forward pass with frozen weights. Start
- 5 In context learning versus training Distinguish learning within a chat from changing the model. Start
- 6 Your data and future models Interpret data use policies accurately. Start
Try asking
About this tutor
For learners who use AI and want a firm grip on the two phases in a model's life. You follow a model from raw data through pretraining, fine tuning and evaluation, then into deployment where every chat is inference with fixed weights. You learn what 'weights' and 'parameters' are, why training is so expensive and inference cheaper per request but huge in total, what a knowledge cutoff is, and whether your chats are used for training. The lessons clear up widespread misunderstandings, such as the idea that a chatbot learns from you live, and give you the vocabulary to read technical news and product policies accurately.
Reviews
4.5
2 ratingsSample
- Elif Y.Sample
The true or false drills cleared up three wrong beliefs I had. I finally understand what our vendor's 'we never train on your data' actually covers.
- Marcus B.Sample
Rigorous and clear. The post training lesson was dense, I had to go back over preference tuning twice, but it got there.
About the teacher
I explain how language models really work, from tokens to attention, without hand waving
9 tutors 525 lessons taught Sample
I like taking the mystery out of language models. I teach what happens between typing a question and reading an answer: tokens, context windows, embeddings, attention, training and fine tuning. I start every topic with a picture or a small worked example and only add maths when it earns its place. My work background is in software and teaching workshops...
See Bastian's profile and tutorsMore like this
Other tutors on the same or nearby topics.