How do LLM’s work
- Areas of intrest
- Eliza: created in 1966 acts as a psychotherapist
-
ML Model
-
Here the model will have a mathematical function
-
Generally to LLMS we pass text and models understand numbers
- in 2013 word2vec was created
- word2vec converts word into a point in a larger dimensional space which is referred as vector.
-
With this we can do semantic search.
-
Approach
TEXT => Break this text into tokens => each token has a vector
-
LLM is a neural network (ML Model) which is trained on almost all publically available information.
-
LLMs predict next token (word) repeatedly
- So for LLMs to generate text we need to give input (prompt) and llms generate text till limits are hit or model predicts
<EOS> - The core system which made the difference in predicting the next token is called as transformer.
