LLM Q&A #1: Prompting vs Fine-Tuning, More vs Fewer Sources for RAG, Prompting vs LLMs as a System
9I have been receiving some questions lately on how to use LLMs. Here are some of my brief thoughts on the matter:
Q1: Should I prompt or fine-tune an LLM?
Prompt before fine-tuning. It is possible to use both actually if needed. Prompting is about 1000 examples of fine-tuning.
Q2: Should I use more or fewer sources for Retrieval Augmented Generation?
Use as few as possible. A few small models with short context is better than a longer context one.
Q3: Should I focus my efforts on prompting techniques?
Not really. Focus on LLMs as a system so you get to augment the capabilities with tools.
~~~~~~~~~~~~~~~~~~~~~~
References:
How many data points is a prompt worth?
https://arxiv.org/pdf/2103.08493.pdf
Retrieval meets long context large language models
https://browse.arxiv.org/pdf/2310.03025.pdf
RECOMP: Improving retrieval-augmented LMs with compression and selective augmentation
https://arxiv.org/pdf/2310.04408.pdf
LLMs as Optimisers (DeepMind)
https://arxiv.org/pdf/2309.03409.pdf
PromptBreeder (DeepMind)
https://arxiv.org/pdf/2309.16797.pdf
LLMs as a System to solve the ARC Challenge
https://github.com/tanchongmin/TensorFlow-Implementations/blob/main/Paper_Reviews/LLMs%20as%20a%20System%20for%20the%20ARC%20Challenge.pdf
~~~~~~~~~~~~~~~~~~~~~~~
0:00 Q1: Prompting vs Fine-tune an LLM
2:05 Q2: More or Fewer Sources for RAG
4:27 Q3: Prompting vs LLMs as a System
~~~~~~~~~~~~~~~~~~~~~~~
AI and ML enthusiast. Likes to think about the essences behind breakthroughs of AI and explain it in a simple and relatable way. Also, I am an avid game creator.
Discord: https://discord.gg/bzp87AHJy5
LinkedIn: https://www.linkedin.com/in/chong-min-tan-94652288/
Online AI blog: https://delvingintotech.wordpress.com/
Twitter: https://twitter.com/johntanchongmin
Try out my games here: https://simmer.io/@chongmin