LLMs currently are a system for learning* large complex functions from example data**. Currently those functions do not represent very long temporal sequences.*** They are also extremely energy inefficient.****
* approximating
** The data should really be cleaned, curated, and presented in the right order.
*** i.e., there is limited short term memory. This somewhat limits the tasks these LLMs are good for. But, again, different architectures are good for different tasks/specialties.
**** this is a rather brute force approach
No comments:
Post a Comment