#80 Solving NLP (Natural Language Processing) Tasks Using LLMs (Large Language Models)

[samvmdev]

Pablo Duboue: Solving NLP (Natural Language Processing) Tasks Using LLMs (Large Language Models) 

## Timestamps 
00:00 Data Umbrella Introduction
02:43 Speaker Introduction + Land Acknowledgment
04:52 Agenda
06:15 NLP history (rule-based, statistical, deep learning)
09:00 What is a Language Model?
10:35 Large Language Models
11:55 Training LLMs - more than just language
12:48 Speaker background
13:35 About this talk - more background
14:26 Section 1: NLP / LLM Tasks - Part-of-Speech tagging
15:48 POS tagging example
16:50 NLP Tasks - Named Entity Recognition (NER), example
17:50 NLP Tasks - Information Extraction (IE), example
19:08 NLP Tasks - Sentiment Analysis, example
20:32 Q&A - data tagging
22:41 Section 2: Prompting 101
22:51: OpenAI API - intro, CLI, Python
25:44 Zero shot - no examples, temperature, output/hallucinations
28:35 Few shot - training data, output, GPT-4
30:17 Handling priors in exemplars
30:40 Chain-of-thought (CoT)
31:13 LLM role
31:43 Recursing
32:23 Learning more - additional resources
33:14 Section 3: Solving NLP Tasks with OpenAI API
33:27 OpenAI POS tagging
34:11 Output is unstable
34:21 Better prompt
34:40 Annotation Manual
36:03 NER prompt, unstable output, MUC-6 locations
38:28 ChatGPT output
38:41 GPT4 output
39:00 Q&A - AGI
40:28 IE prompt - relation extraction, stable output
42:31 Sentiment Analysis prompt
43:26 Additional discourse
44:22 Section 4: Using open source LLMs
44:39 Why open source LLMs
45:44 Issues with open source models
46:33 Examples of open source LLMs
49:29 Conclusions
51:34 Q&A - contributing to new models v. expanding on older ones, LLMs in cell phones, communication changes and abstraction, etc.

## Resources
- https://tellandshow.org/ (community-owned machine learning)
- http://textualization.com/gptwhitepaper/ 
- http://artoffeatureengineering.com/ 
- http://wiki.duboue.net/A_Dollar_Worth_Of_Ideas (project ideas)

## Connecting
- LinkedIn: https://www.linkedin.com/in/pabloduboue/
- GitHub: https://github.com/drdub 
- Twitter: @pabloduboue

[welcome[bot]]

:tada: Welcome to Data Umbrella! :tada: We're really excited to have your input into the project! :sparkling_heart:
If you haven't done so already, please make sure you check out our Contributing Guidelines and Code of Conduct.

[reshamas]

Thank you @greenhacks. I've added the timestamps to: a) file here: https://github.com/data-umbrella/event-transcripts/blob/main/2023/80-pablo-llm.md b) the video