Mastering Prompt Engineering
The webinar covered how to correctly formulate queries to a language model to maximize business benefits. Viewers learned about prompt engineering methods, saw effective query examples, and were introduced to tools for creating applications based on YandexGPT.
The webinar discussed:
how prompt engineering helps maximize the value of YandexGPT for solving business problems
how to formulate queries to achieve the best results
what LLM features and limitations are important to know when creating prompts
The webinar was useful for anyone developing their own digital services and wanting to understand the basics and advanced prompt engineering methods when working with the YandexGPT API.
00:00 Introduction
01:24 What are large language models
02:49 How a model works internally
11:14 Few-Shot and Zero-Shot queries
15:29 How a model is trained
18:11 How to experiment with prompts in Yandex Cloud
19:20 How a dialogue is built
21:00 Roles in a prompt
22:45 How to approach GPT
22:55 Why prompt engineering is needed
30:10 The first rule of prompt engineering
31:10 Prompt engineering techniques
33:10 Example: generating a description for a toy airplane
37:10 Tips for writing prompts
37:59 What LLM can do well and badly
40:45 Demo: Extracting structured data from text
43:15 Extracting diagrams from Text
47:28 The difference between GPT and humans
48:44 Chain of Thought
53:00 What to do if prompts don't work
56:10 Retraining scenarios
56:40 How creative are models
01:00:00 About neural network hallucinations
01:02:10 Conclusions
01:06:00 Answers to questions