1 of 32

ChatGPT : pour un jargon scientifique plus accessible grâce à l’IA

Laboratório em Humanidades Digitais | UMinho

Journée d’études Mots/Machines # 5

Université de Bretagne Occidentale

Brest | 17 mars 2023

2 of 32

Outline

  • Exploring ChatGPT as a learning assistant
  • Exploring terminology with ChatGPT
  • ChatGPT - Building mind maps
  • ChatGPT - Building conceptual maps

3 of 32

Exploring terminology

with ChatGPT

Can help automating repetitive tasks

Organizing and presenting information in various formats

Source of structured information like examples and definitions

Adapting information to different tones and complexity levels

Source of feedback for written text

Source of insights on complex problems and hypotheses validation

4 of 32

ChatGPT Drawbacks

Accuracy

Generates wrong and artificial information, including non-existent references (Hallucination)

Bias

Bias in the training data will cause bias in the model

Understanding

The model doesn’t have a true understanding

Creativity

Creativity is limited

by patterns

in the training data

5 of 32

Workflow

Extract > Define > Visualize

6 of 32

MULTIMODALITY

> use of more than one mode of communication in a text to create meaning

7 of 32

Building mind maps with

ChatGPT + Draw.io

8 of 32

Mind maps for structuring information

Mind map:

Arranges information in a hierarchical structure, showcasing how each piece of information is related to the overall concept.

It presents a clear picture of the relationships among different components of the information, making it easy to comprehend and analyze

Source: https://en.wikipedia.org/wiki/File:Cubital_Fossa2.png

9 of 32

1- Getting structured text

10 of 32

2- Getting PlantUML code

Prompt:

write mind map in PlantUML format

central topic: ChatGPT

first level: list hierarchy first level

second level: split elements list hierarchy second level

must have domains and subdomains branching out from central topic

11 of 32

2- Getting PlantUML code

Output:

@startmindmap

* ChatGPT

** Features

*** Language generation

*** Learning from data

*** Fine-tuning for domains

** Limitations

*** Bias and context sensitivity

*** Limited understanding of non-textual cues

*** High computational requirements

** Applications

*** Customer service

*** Personalized assistants

*** Language learning tools

@endmindmap

12 of 32

2- Getting mind map

13 of 32

2- Getting mind map

14 of 32

2- Getting mind map

15 of 32

Text mining | Outputs

16 of 32

Creating videos with Simpleshow

17 of 32

2- Getting PlantUML code

Prompt:

write the script for a 30 seconds video using the text from the mind map from your last answer

18 of 32

2- Script

Output:

19 of 32

2- Making the video

Prompt:

write the script for a 30 seconds video using the text from the mind map from your last answer

20 of 32

Text mining | Outputs

https://videos.simpleshow.com/JSZ33Ha2fq

21 of 32

Share your work in our Padlet!

https://bityli.com/h6bYyE

22 of 32

23 of 32

Workflow

Extract > Define > Visualize

24 of 32

KEYWORD EXTRACTION

Prompt:

List all technical terms from the text “text”.

25 of 32

KEYWORD EXTRACTION

26 of 32

DEFINITION

Prompt:

write a glossary using all the terms from your last answer. Each term must have a definition, example use in context, synonym list, related terms list, in this order. Example must be different from original text.

27 of 32

DEFINITION

28 of 32

VISUALIZATION

Prompt:

Given a text extrapolate as many relationships as possible from it. Every node has an color (in hex). Every edge has a to and from with node and must have a label. Edges are directed, so the order of the from and to is important. Write Graphviz code do build a knowledge graph diagram. text = "text"

29 of 32

30 of 32

VISUALISATION

31 of 32

Links

32 of 32

Thank you!

Francisco Lopes

franciscolopesdacunha@gmail.com

Sílvia Araújo

saraujo@elach.uminho.pt

Micaela Aguiar

maguiar60@gmail.com