Скачать или смотреть 26 Hacks to Optimize LLM Results in 2024

In this video, I reviewed and tested 26 principled instructions provided by researchers that can significantly improve the output results of Large Language Models (LLMs) such as ChatGPT, Bard, LLAMA, Perplexity, Claude and so on.

Link to the paper: https://doi.org/10.48550/arXiv.2312.1...
Github: https://github.com/VILA-Lab/ATLAS

00:00 - Intro
02:22 - Don't be polite with LLMs
02:47 - Integrate audience, break down tasks and be affirmative
03:30 - Simplify
05:00 - Claim that you tip $xxx for a better solution!
06:00 - Give instructions
06:20 - Use "You MUST"
06:35 - Use "You will be penalized"
06:51 - Human-like answer
08:04 - Think step by step
08:14 - Get unbiased answer
08:28 - Tell the model to ask you questions
09:00 - Tell the model to test you
09:36 - Assign a role
09:46 - Use delimiters
10:25 - Combine chain-of-thought with few-shot-prompting
11:36 - Use output primers
12:39 - How to continue a text
14:27 - Experiments