Loading slide

What a Trained Model Actually Is

  1. 01What a Trained Model Actually Is
  2. 02Memory card
  3. 03A model is a file
  4. 04Two files
  5. 05How the two fit together
  6. 06A Piano
  7. 07The valuable file
  8. 08You connect to it
  9. 09Where does training data come from?
  10. 10Cleaning the data
  11. 11The network can't read
  12. 12Guess the next word
  13. 13Right or wrong
  14. 14One nudge, a billion times
  15. 15Take a breath
  16. 16Where your memories live
  17. 17A base model is not an assistant
  18. 18Why it's called pre-training
  19. 19The weights are the model
  20. 20The scale of it
  21. 21Why so many?
  22. 22What a weight actually is
  23. 23Everywhere and nowhere
  24. 24Using the model: inference
  25. 25The whole arrangement
  26. 26Reinforce your understanding
  27. 27Question: What is a model, really?
  28. 28Question: Learning from raw text
  29. 29Question: The model doesn't know what it doesn't know
  30. 30Quiz: answer
  31. 31The machine learned something real.
  32. 32Want to go deeper?
1 / 31
BackNext

Module IV

Learning from Data

Chapter III

What a Trained Model Actually Is

In this chapter

  • The weights are the modelwhat remains after training
  • Distributed knowledgewhy no single weight means anything alone
  • Inferencehow a trained model produces an answer
  • What it can't dothe limits that come with this kind of knowledge
  • A strange kind of knowledgewhat it means to know without understanding

After training, something remains. But it is not what most people picture.

Not a list of rules. Not a database of facts. Not a set of instructions someone wrote down. What remains is a very large collection of numbers, arranged in a configuration that the training process arrived at through billions of small adjustments.

Those numbers encode, in a form no human can directly read, whatever patterns the model extracted from its training data. The knowledge is real. It is also strange. It cannot be opened up and inspected the way a filing cabinet can. It does not live in any one place. It is distributed across the whole.

This is the thing about learned systems: they work, often impressively, but the reason they work is buried inside something that resists explanation. That gap between capability and legibility is not a temporary problem waiting to be solved. It is, in some deep sense, what learning produces.

Citations