ChatGPT: AI tools, applications and implications

Ali Abbas

Agenda

  • Intro to ChatGPT
  • State of the art
  • Examples
  • Tools using OpenAI models like ChatGPT (or similar) models
  • Issues
  • Summary

ChatGPT

Overview

  • ChatGPT: Optimizing Language Models for Dialogue
  • Created by OpenAI
  • Debut in November 2022
  • Viral due to usability and accessibility
  • Built on GPT-3.5 series with Reinforcement Learning from Human Feedback
  • Trained on a dataset of over 40GB of text data
  • With 175 billion parameters

(Some of) Applications

  • Text generation
    • Coherent, creative on any topic and in specific style
  • Language translation
    • Translation, idiomatic expressions and remembers context
  • Dialogue systems
    • Conversational agents such as chatbots
  • Summarization
    • Long to short summary and extract key information

State of the Art

Taxonomy AI models

State of the art - popular generative models Gozalo-Brizuela and Garrido-Merchan (2023)

Companies

AI companies Gozalo-Brizuela and Garrido-Merchan (2023)

Timeline

Timeline Gozalo-Brizuela and Garrido-Merchan (2023)

Examples

Note to neighbour

Poem

Poem on benefits of Walking and Cycling

Perplexity

GPT based search engine

Children’s exposure to PM2.5 affecting their health

Elicit

GPT-3 powered AI tool for Literature Review

TutorAI

OpenAI powered AI tool for creating educational content

TutorAI: Educational content creator

TutorAI example - Physical Activity Epidemiology 1

TutorAI example - Physical Activity Epidemiology 2

RTutor

OpenAI’s powerful text-davinci-003 language model to translate natural language into R/Python code

OpenAI's powerful text-davinci-003 language model to translate natural language into R code

GPT Tools

Comment code

Convert a Script into Functions

GPT Studio

Add-on for RStudio for incorportaing OpenAI into project workflows

Code

OpenAI R library

Example

Specify a string to create an image

library(openai)
library(magick)

img <- openai::create_image("simplicity")

print(image_read(img$data$url))
  format width height colorspace matte filesize density
1    PNG  1024   1024       sRGB FALSE  3147977   72x72

Directory of AI tools

Futurepedia

An AI tools library

Issues and Summary

Issues

  • Easier access and universal applicability
  • Black box model
    • Limited knowledge on how they work

    • When they fail

    • Capabilities due to emergent properties - anxiety about unanticipated consequences

    • Context is not known either

    • Downstream bias: intrinsic bias, representational bias, underrepresented and over-represented (Anglocentric population) Bommasani et al. (2021)

    • Compounds existing inequities by producing unfair outcomes, entrenching systems of power, and disproportionately distributing negative consequences of technology to those already marginalized Bommasani et al. (2021)

  • (Almost) exclusively build by industry (Google, Facebook, Microsoft, or Huawei, or startups such as OpenAI) Bommasani et al. (2021)
  • Nefarious purpose: fake news/propagandas, misinformation or any other harmful purpose

Summary

References / further reading

  • Futurepedia: AI Tools directory
  • Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H. P. D. O., Kaplan, J., ... & Zaremba, W. (2021). Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374.
  • Guo, B., Zhang, X., Wang, Z., Jiang, M., Nie, J., Ding, Y., ... & Wu, Y. (2023). How Close is ChatGPT to Human Experts? Comparison Corpus, Evaluation, and Detection. arXiv preprint arXiv:2301.07597.
  • Sobania, D., Briesch, M., Hanna, C., & Petke, J. (2023). An Analysis of the Automatic Bug Fixing Performance of ChatGPT. arXiv preprint arXiv:2301.08653.
Bommasani, Rishi, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S. Bernstein, et al. 2021. “On the Opportunities and Risks of Foundation Models.” https://doi.org/10.48550/ARXIV.2108.07258.
Gozalo-Brizuela, Roberto, and Eduardo C. Garrido-Merchan. 2023. “ChatGPT Is Not All You Need. A State of the Art Review of Large Generative AI Models.” https://doi.org/10.48550/ARXIV.2301.04655.