My wishlist for "LLM alchemy"
Working with LLMs is pretty much like doing a bit of alchemy, especially prompt engineering.
What I want to learn to do:
-
Communicate with LLMs via API and for batch processing of tasks.
-
Text Embedding + clustering using LLM.
-
Making a function call to an external agent/program using LLMs. For example, make it call python symbolic processor.
-
Retrieval Augmented Generation (RAG) ()
Applications and interesting experiments
-
Have LLM judge the reliability of its own outcome and raise the most problematic ones.
-
Get LLMs to draw/modify diagrams in SVG/Tikz
Tips and Tricks
- Getting “dill” to work when an LLM object is present in python notebook.