![[Pasted image 20250913225254.png]]
# what do you mean when you say
# a language model is worth 96 excel tables?
# this can't be true, can it?
---
<< embed a 30 sec video of
<< going to hugging face for a model
<< downloading one file
<< opening python script to load the tensor
<< showing the visualization
---
# when we say "language model",
# we mean it!
# this technology models our language, which in practice involves estimating a probability distribution over a vocabulary
---
# wtf did you just say?
## probability distribution
ever seen this?
<< insert a gaussian curve
<< preferably not a screenshot but a js script
<< embeddable in obsidian?!
## vocabulary
ever heard of ai needed huge datasets to learn from?
in practice, ai researchers ...
- batches
- self supervised learning
- 100.000 iterations
---
hence,
# each ai model is a table with numbers
# those tables represent a probability distribution
# which generates new words on the basis of the words that come before in a sentence
---