Ekaterina Seikkinen

How to Use ChatGPT for Program-

ming

Metropolia University of Applied Sciences

Bachelor of Engineering
Information and Communications Technology
Bachelor’s Thesis
26 May 2023
Abstract

Author: Ekaterina Seikkinen

Title: How to Use ChatGPT for Programming
Number of Pages: 35 pages
Date: 26 May 2023

Degree: Bachelor of Engineering

Degree Programme: Information and Communications Technology
Professional Major: Mobile Solutions
Supervisor: Miikka Mäki-Uuro, Senior Lecturer

The main goal of this thesis was to learn how to use ChatGPT for programming and
create a mobile application project using an AI language assistant as a help tool.
Proper use of ChatGPT in the programmer’s workflow can reduce the time required to
solve certain tasks. The thesis was aimed not only at developing application but at
exploring the potential benefits and limitations of using intelligent chatbot as a devel-
opment tool and figuring out the best ways to use it effectively.

Also, this thesis goes through the basic concepts of ChatGPT and explores different
language models, which are used to build neural conversational models and their ap-
plications.

As a result of this thesis, a basic version of the Task Management application was
developed, which included all the functionalities that were desired. The idea of the
application is to help people to manage daily tasks to save time. Frontend was imple-
mented with the help of React Native library and Authentication was implemented with
the help of Firebase.

The process of creating the prototype went smoothly without any big issues. ChatGPT
was especially helpful for generating code snippets, although the intelligent chatbot
could not always solve problems or errors that came up. But overall, ChatGPT helped
to build mobile application with basic functionality in few days. With experience in Re-
act programming and Firebase Authentication, understanding errors and challenges
became easier. To take full advantage from AI language assistant some knowledge
around programming is required.

Keywords: React Native, mobile development, ChatGPT, AI

___________________________________________________________

The originality of this thesis has been checked using Turnitin Originality Check ser-
vice.
Tiivistelmä

Tekijä: Ekaterina Seikkinen

Otsikko: ChatGPT:n käyttö ohjelmoinnissa
Sivumäärä: 35 sivua
Aika: 26.5.2023

Tutkinto: Insinööri (AMK)

Tutkinto-ohjelma: Tieto- ja viestintätekniikka
Ammatillinen pääaine: Mobile Solutions
Ohjaaja: Lehtori Miikka Mäki-Uuro

Opinnäytetyön tarkoituksena oli luoda mobiilisovellus käyttämällä tekoälyn kieliassis-

tenttia apuvälineenä. ChatGPT:n oikea käyttö ohjelmoijan työskentelyssä voi lyhen-
tää tiettyjen tehtävien ratkaisemiseen kuluvaa aikaa. Opinnäytetyön tarkoituksena oli
sovelluksen kehittämisen lisäksi selvittää älykkään chatbotin käytön mahdollisia etuja
ja rajoituksia kehitystyökaluna ja selvittää parhaat tavat käyttää sitä tehokkaasti.

Yksi tekoälyn alueista ovat kielimallit. Kielimalli on todennäköisyysjakauma sanoihin

tai sanasarjoihin. Kielimalleja käytetään monissa palveluissa, esimerkiksi ne auttavat
puheassistentteja yläpitämään dialogeja, antavat nopeita vastauksia haussa, paran-
tavat käännösten laatua ja paljon muuta. Yksi edistyneimmistä kielimalleista on Ope-
nAI:n kehittämä GPT-3. GPT-3:ssa on 175 miljardia parametria, ja sen harjoittami-
seen käytettiin valtava määrä tekstiä. ChatGPT on erittäin suosittu tekoälypohjainen
ohjelma, jota käytetään dialogien luomiseen. Älykkäässä chatbotissa on kielipohjai-
nen malli, jota kehittäjä hienosäätää ihmisten välistä vuorovaikutusta varten.

Opinnäytetyön tuloksena kehitettiin Task Management-sovelluksen perusversio, joka

sisälsi kaikki halutut toiminnot. Sovelluksen idea on auttaa ihmisiä hallitsemaan päi-
vittäisiä tehtäviä ajan säästämiseksi. Käyttöliittymä toteutettiin React Native -kirjaston
avulla, ja todentaminen toteutettiin Firebase-kehitysalustan avulla.

Prototyypin kehityksessä ei ilmennyt suurempia ongelmia. ChatGPT oli erityisen hyö-

dyllinen koodinpätkien luomisessa, vaikka älykäs chatbot ei aina pystynyt ratkaise-
maan esiin tulevia ongelmia tai virheitä. React-ohjelmointiin ja Firebase-todennuk-
seen perehtyminen auttoi virheiden ja haasteiden ymmärtämisessä. Tekoälyn kielias-
sistentin täyden tehon hyödyntäminen edellyttää ohjelmoinnin tuntemusta.

Avainsanat: React Native, mobiilikehittäminen, ChatGPT, tekoäly

___________________________________________________________
Contents
List of Abbrevations

1 Introduction 1

2 Introduction to ChatGPT 2

2.1 Artificial Intelligence 2

2.2 Language Models 3
2.3 OpenAI 5
2.4 GPT-3 Language Model 6
2.5 Instruct GPT 8
2.6 ChatGPT 10

3 Task Management Application 11

3.1 Purpose and Goals of the Application 11

3.2 Development Process: Technologies and Tools 12
3.3 Utilizing ChatGPT for Application Development 13
3.4 Components to be Implemented in the Application 16
3.4.1 User Authentication and Authorization 16
3.4.2 Task Creation and Management 21
3.4.3 Task List and Filtering 24
3.4.4 Task Details 25

4 Results 27

4.1 Advantages of Using ChatGPT for Development 27

4.2 Limitations of Using ChatGPT for Development 28
4.3 Best Practices for Utilizing ChatGPT 29

5 Conclusion 32

References 34
List of Abbrevations

AI: Artificial Intelligence

GPT: Generative Pre-Trained Transformer

IDE: Integrated Development Environment

NLP: Natural Language Processing

RLHF: Reinforcement Learning from Human Feedback

RM: Reward Models

SFT: Supervised Fine-Tuning

 1

1 Introduction

Everything changes and modern technology is in demand in all areas of life.

Neural networks and artificial intelligence make human life easier by solving
many problems several times faster than humans. ChatGPT launched by the
company OpenAI on November 30, 2022, impressed the IT community with the
accuracy of its answers to certain questions, becoming the most discussed pro-
ject in the field of artificial intelligence. ChatGPT is a chatbot that uses artificial
intelligence, and its main features include answering any question, inventing
stories, asking additional questions, translating texts, giving advice, and helping
with programming.

The advent of ChatGPT raises important questions about how many people's
lives will change and whether professionals will lose their jobs. The main pur-
pose of the thesis project is to study the use of ChatGPT in programming and to
check the accuracy and correctness of its answers, as well as to understand
how it can simplify the programmer's workflow. The engineering work also ex-
amines the language models used to create neural conversational models and
explores the basic concepts of ChatGPT.

The work first introduces the technologies related to ChatGPT, including its his-
tory of creation and prototypes, and then demonstrates its use in programming
by providing an example of creating a mobile application.

The idea for the engineering job came from Suomen Sijoitutkimus Oy, the com-
pany where I work. The correct use of ChatGPT in a programmer's work can
shorten the time required to solve certain problems and become an indispensa-
ble assistant at work.

Chapter 2 provides an overview of artificial intelligence and its basic concepts,

including the history of the creation of ChatGPT and key elements of intelligent
 2

chatbot. In chapter 3, the thesis goes through the details of case project, explor-
ing the development process of a mobile application using a conversational AI.
Chapter 4 aims to provide practical insights and best practices for utilizing Chat
GPT for application development.

2 Introduction to ChatGPT

The chapter begins with an overview of artificial intelligence, its definition, and
key technique such as natural language processing. This is followed by an ex-
planation of language models, including their purpose, structure, and types. Af-
ter that, chapter introduces the history of ChatGPT, its creators, the process of
development and fundamental principles of artificial language model.

2.1 Artificial Intelligence

Artificial intelligence has permeated every aspect of modern life and continues
to gain interest. AI refers to the simulation of human intelligence in machines
and is based on the principle that human intelligence can be defined in a way
that allows machines to imitate it and perform tasks. (Frankenfield 2022). It can
be a manifestation of some kind of creative ability, a tendency to reason, to gen-
eralize, to learn from early experience.

John McCarthy (the founder of functional programming and the inventor of the
Lisp programming language) first mentioned the term "artificial intelligence" in
1956. However, Alan Turing formed the idea of such a system in 1935. The sci-
entist created a description of an abstract computer consisting of unlimited
memory and a scanner that moves back and forth through the memory. The
earliest successful artificial intelligence program was created by Christopher
Strachey in 1951, and already in 1952 it played checkers with a human and sur-
prised the public with its ability to predict movements (History of Artificial Intelli-
gence 2020). Nowadays, such programs are developing very quickly due to the
digitization of information and the growth of its turnover and volume.
 3

AI algorithms are at the core of artificial intelligence, which exhibit some of the
behaviours associated with human intelligence. These include computer-en-
hanced learning, reasoning, perception, and imitating human cognitive activity
(History of Artificial Intelligence 2020). AI systems work by analysing large
amounts of labeled training data for correlations and patterns, which are used to
predict future states. For example, an AI language assistant that is fed text chat
examples can learn to produce a realistic conversation with people through Nat-
ural Language Processing (NLP) (Roldós 2020).

There are two types of artificial intelligence: strong AI and narrow AI. Strong AI
refers to programming that can replicate the cognitive abilities of the human
brain, such as cyber security. Narrow AI, on the other hand, is an AI system de-
signed and trained to perform a specific task, such as Apple's Siri.

AI is widely used in various fields from finance to healthcare, and it has become
integrated into everyday devices, such as vacuum cleaners. Examples of AI in-
clude chatbots, intelligent assistants, robo-advisors, email spam filters, and Net-
flix recommendations.

Nowadays artificial intelligence is used in various fields from finance to

healthcare, and it is also integrated into the most common facilities, such as the
vacuum cleaner. The inner workers of artificial intelligence are not fully under-
stood, but it is predicted by experts that the development of artificial intelligence
will come even closer to the development of the human brain within the upcom-
ing years. (Frankenfield 2022).

2.2 Language Models

A language model is a core component of modern Natural Language Pro-

cessing (NLP). It is a statistical tool that analyses the pattern of human lan-
guage to predict words. NLP is a branch of artificial intelligence that allows ma-
chines to understand human language. Its goal is to build systems that
 4

understand text and automatically perform tasks such as translation, spelling, or

topic classification. (Roldós 2020).

The language model uses machine learning to perform a probability distribution

on the words used to predict the most likely next word in the sentence based on
the previous entry. Simply put, a language model is an algorithm that, based on
n words of the proposed text, can guess what the next word is and what the
probability of this word's occurrence is. The language model always receives
text as input and produces text data as output. Since many tasks can be formu-
lated in the text, language models are used extensively. (Kapronczay 2022).

Working with language models is based on probability theory: algorithms calcu-

late the probability that a certain word appears in the text. For this, context, style
of speech and meaning of words must be considered, but the machine can only
work with numbers. Therefore, the text is converted into its numerical represen-
tation - this process is called encoding, and the result is a numerical sector.
Each word has a numerical representation, and the neural network looks at
which word combinations and in which order are most often found together in
the language. The larger and more diverse text set is taught to it, the more com-
plex dependencies the model captures and repeats them with new information.
(Kapronczay 2022).

There are two types of language models:

1. Probability-based methods

The language model examines the word's probability of occurrence based on

text examples. Simpler models can consider the context of a short string of
words, while larger models can work at the sentence or paragraph level.

2. Neural network-based modern language models

Neural network-based language models ease the sparsity problem by encoding

inputs. Word embedding layers create a vector of arbitrary size from each word,
 5

which also contains semantic relations. These continuous vectors create the de-
sired precision in the probability distribution of the next word. (Kapronczay
2022).

The most famous language models are GPT-3 (developers: OpenAI, parame-
ters: 175 billion), Bloom (developers: Hugging Face, BigScience; parameters:
176 billion), ESMFold (developers: Meta AI, parameters: 15 billion), Gato (de-
velopers: DeepMind, parameters: 79 million, 364 million and 1.18 billion),
LamDa (developers: Google, parameters: 137 billion). The number of parame-
ters in a language model is a measure of its ability to recognize complex pat-
terns and relationships in data and develop new emergent capabilities.
(Wodeski 2022).

2.3 OpenAI

OpenAI is an artificial intelligence research laboratory and development com-

pany. It has created a variety of AI-powered programs and machine learning al-
gorithms that allow computers to do all kinds of things, such as create images
from text or make a robotic hand that solves Rubik's cubes (Ot 2022). This altru-
istic AI company was founded by Elon Musk and Sam Altman in 2015. It is offi-
cially a non-profit organization and works with funds invested by investors. The
company is directly trying to build safe and useful artificial intelligence (AGI), but
also considers its mission accomplished if the work helps others achieve that
result. (A Guide to OpenAI and its Products and Services 2023).

OpenAI conducts research in many fields related to artificial intelligence, includ-

ing machine learning, robotics, economics, and computing. In addition to re-
search work, OpenAI also aims to educate the public about artificial intelligence
and its possible effects and to promote the responsible development and use of
artificial intelligence.

During its seven years of existence, OpenAI has developed many artificial intel-
ligence tools that perform specific tasks. They are as follows:
 6

1. OpenAI Gym is a toolkit for developing and comparing reinforcement learning

algorithms.

2. OpenAI RoboSumo is a simulated environment for developing and testing ro-

bot control algorithms.

3. The OpenAI Debate Game simulates a debate between two or more partici-
pants using artificial intelligence.

4. OpenAI Dactyl is a robotic hand designed to learn and adapt to new tasks us-
ing machine learning algorithms.

5. OpenAI Generative models (including DALL-E and ChatGPT) are artificial in-
telligence systems trained to generate new, original content based on a set of
examples or a specific model. (A Guide to OpenAI and its Products and Ser-
vices 2023).

2.4 GPT-3 Language Model

GPT-3 is Generative Pre-trained Transformer 3, and it is the third version of the

language model released by OpenAI in May 2020. It is generative because
GPT-3 can produce long sentences from unique text as output. Most neural net-
works can only produce yes or no answers or simple sentences. Pretrained
means that the language model is not built with domain-specific knowledge but
is able to perform domain-specific tasks such as translation. Transformer is a
neural network architecture developed by Google researchers in 2017. (What is
GPT-3, How Does It Work, and What Does It Actually Do? 2021).

Deep learning can be considered as a specialized branch of machine learning.

It is a field that relies on learning and improving on its own by examining com-
puter algorithms. While machine learning uses simpler concepts, deep learning
works with artificial neural networks, which are designed to imitate how humans
think and learn. (Reyes 2023).
 7

GPT-3 is a deep learning-based language model that can generate text that
closely resembles human-written content. In addition to text, it can also produce
code, stories, poems, and other forms of creative content. As a result of these
capabilities, it has gained significant attention and popularity in the field of natu-
ral language processing. The full version of OpenAI GPT-3 is, with approxi-
mately 175 billion parameters to be trained, the largest model trained so far
compared to other language models. (Cooper 2021).

The GPT-3 model differs from other language models in that it has a huge num-
ber of parameters to train, 10 times more than any previous model. In general,
the more parameters a model has, the more data is needed to train the model.
According to the authors, the OpenAI GPT-3 model has been trained with
around 45 terabytes of text data from various sources such as Wikipedia and
books. (Cooper 2021).

GPT-3 is not a single model but a family of models. Each model in the family
has a different number of parameters to be trained. The table in Figure 1 shows
each model, architecture, and corresponding parameters.

Figure 1. An example of the GPT-3 models family (Cooper 2021).

The OpenAI GPT-3 model family is based on a transformer-based architecture

(Figure 2).
 8

Figure 2. Transformer-based architecture (Cooper 2021).

Encoder: The encoder receives the input and constructs a representation (its
features) from it. This means that the model is optimized to understand the in-
put.

Decoder: The decoder uses the encoder representation (features) along with
other inputs to generate the target sequence. This means that the model is opti-
mized for output.

A key part of transformers is the attention process, which allows the model to
understand which words are most important to consider when judging what the
sequence produces. In each iteration of the encoder, each word is given numer-
ical weights, which are then analysed by the decoder to determine key aspects
of the input text on which the output should be based. (Bard 2021).

2.5 Instruct GPT

Instruct GPT, or simply Instruct, is a powerful tool that allows users to fine-tune
the generation features of the GPT model language. Developed by OpenAI, In-
struct GPT allows users to train a model for specific tasks and generate text
personalized to their specific needs. This tool is designed to facilitate develop-
ers and researchers fine-tuning GPT for a wide range of natural language pro-
cessing (NLP) tasks such as language translation, text summarization, and
 9

question answering. Instruct GPT allows users to train the model with their own
data, fine-tune the model for specific use cases, and create text that is relevant
and accurate. The goal of Instruct GPT is to facilitate and improve the automa-
tion of repetitive and time-consuming tasks of companies and organizations.
(Dongre 2023).

The key difference between GPT and Instruct GPT is that GPT can be fine-
tuned for any task, while Instruct GPT needs to be specifically fine-tuned for a
specific set of tasks. While Instruct GPT is highly proficient at certain tasks, its
narrower focus makes it less adaptable than the original GPT.

Compared to GPT-3, the new Instruct GPT models adhere better to English in-
structions, are less likely to produce false information, and at least slightly less
likely to produce harmful results.

One issue with GPT-3 is that it was trained to predict the next word from a large
data set and not safely perform the task the user wanted. To solve the problem,
OpenAI used a method called reinforcement learning with human feedback
(RLHF). Reinforcement learning is a technique that involves an AI agent making
decisions based on the actions it performs in its environment and the feedback
it receives in the form of rewards or punishments. (Cuofano 2023).

To convert GPT-3 models to Instruct GPT models, OpenAI designed a three-

step procedure. The first is model fine-tuning. The second is building a reward
model (RM). Third, the SFT (Supervised Fine Tuning) model must be taken and
further fine-tuned using reinforcement learning. Instruct GPT is more attuned to
human preferences, which ultimately better predicts actual performance. The
reason is that Instruct GPT is better aligned with human intentions through a re-
inforcement learning paradigm that makes it learn from human feedback. (Lowe
& Wainwright 2022).
 10

2.6 ChatGPT

OpenAI's ChatGPT is a very popular AI-based program used to create dialogs.

An intelligent chatbot has a language-based model that the developer fine-tunes
for human interaction (Pocock 2023).

Instruct GPT and ChatGPT are both models developed by OpenAI. However,
the biggest difference between them is in their intended use. ChatGPT is a con-
versational AI model trained to generate human-like responses in dialogue. On
the other hand, Instruct GPT is a task-oriented AI model that is trained to per-
form specific tasks such as code generation, text SQL queries, and summariza-
tion by following instructions given to it.

Both models are built on top of transformer architecture but Instruct GPT is
trained on more versatile tasks and is fine-tuned to perform specific tasks, while
ChatGPT is trained on more general conversational data. Instruct GPT is a spe-
cialized version of ChatGPT designed for a task-oriented context. (Dongre
2023).

ChatGPT works by collecting human-written data from the Internet and using
calculated predictions to answer questions and queries entered by the user.
The responses it generates are based on text queries and information from
which the AI language assistant "learns" more about various topics and how to
discuss them. (Pocock 2023).

It is well trained on biased and unbiased data in the form of texts from books,
articles, and websites. ChatGPT can reproduce output and reliability, which is
critical for many sensitive applications and other valuable AI systems. However,
it is still prone to error and relies on training data provided in 2021. (Pocock
2023).
 11

3 Task Management Application

Chapter introduces the idea and goals of the case project, technologies, and
tools that are used to create the application. After that, details of development
process are provided through this chapter such as key components of the appli-
cation, which are developed with React Native (Task List and Task Details), im-
plementation of user authentication and authorization through Firebase as well
as storing task details in Firestore database.

3.1 Purpose and Goals of the Application

During the search for the best application idea to implement for an IT project,
ChatGPT provided 10 different options. The idea of creating a Task Manage-
ment application was found to be the most appealing, as it has the potential to
help individuals who often struggles with managing multiple tasks.

The purpose of the task management application is to provide platform to sim-

plify the task management process and help users achieve their goals with
greater ease and efficiency. With an increasing number of tasks to complete in
our daily lives, it is becoming more challenging to stay organized and prioritize
our tasks. This application solves this problem by providing users with an intui-
tive and user-friendly interface to manage their tasks.

The application will allow users to create new tasks and filter them based on
their completion status. This will help users understand which tasks have been
completed and which ones still need to be done.

The basic functionality of the Task Management Application has been success-
fully implemented with the help of ChatGPT. The application consists of five
screens, including Authentication Screens for user login and registration, a Task
List Screen that displays all the tasks in a list, a Task Form Screen for creating
new tasks, and a Task Details Screen that allows users to view additional
 12

details about a task, such as its description. The Task Details Screen also fea-
tures a Delete button, which enables users to remove tasks from their task list.

3.2 Development Process: Technologies and Tools

The application was developed using React Native, TypeScript, Expo, Firebase
Authentication, and Firebase Firestore. Both web and mobile applications are
divided into frontend and backend. Frontend refers to the part visible to the user
and its functions, while backend refers to the basis of the application, which the
frontend is designed to utilize.

In a React Native application, the line between frontend and backend can be a
bit blurry since much of the application logic and state management is handled
on the client-side. However, Firebase Authentication and Firestore can be con-
sidered part of the backend since they are cloud-based services that provide
authentication and data storage functionality.

Frontend:

• React Native: a JavaScript framework for writing real, natively rendering

mobile applications for iOS and Android. It is based on React, Face-
book's JavaScript library for building user interfaces, but instead of tar-
geting to the browser, it targets at mobile platforms. In other words: web
developers can now write mobile apps that look and feel truly native, all
with the JavaScript library we already know and love. (Eisenman, 2015).

• TypeScript: a superset of features that enable us to use types in JavaS-

cript. It gives us a module system, classes, interfaces, and a rich type of
system. By using TypeScript write better code, with fewer errors, it helps
to catch mistakes statically that can be committed along the way while
applications are build. The code quality and understandability of our ap-
plication can be enhanced just by starting to use types in our code. (Mal-
donado 2019).
 13

• Expo: an open-source platform to build React Native apps. It is a set with

tools and services built for React Native. Expo can help developers get
started with building React Native apps easily. It offers a range of tools
that simplify the creation and testing of React Native apps. Additionally,
Expo provides a more robust and convenient development workflow with
the flexibility to customize and configure project settings to suit individual
needs. (Borozenets 2022).

Backend:

• Firebase Authentication: a service provided by Firebase that enables de-

velopers to authenticate users using email and password, phone num-
ber, or social media accounts such as Google, Facebook, and Twitter. A
secure and easy way to authenticate users in the application is provided
by it. (Firebase Authentication 2023).

• Firebase Firestore: a NoSQL cloud database provided by Google that al-

lows developers to store and sync data between client-side and server-
side applications in real-time. Store, synchronization, and query applica-
tion data at global scale can be enabled by Cloud Firestore. (Cloud Fire-
store 2023).

3.3 Utilizing ChatGPT for Application Development

ChatGPT was used as a helpful tool throughout the mobile application develop-
ment process. Various application ideas were received from ChatGPT initially,
and then the AI language assistant was asked to determine the most suitable
elements to include, taking into consideration that it was a study project (Figure
3). Valuable advice was provided by chat bot throughout the entire process.
 14

Figure 3. Appropriate components to include in study case.

This list can be utilized to create a product design map. It has been decided that
first three components would be implemented in study case, only limited num-
ber of features in the first iteration of the product.

Chat bot also helped with the application folder structure (Figure 4) and with a
choice of colour scheme (Figure 5).
 15

Figure 4. Possible Task Management application folder structure, suggested by

ChatGPT.

The folder structure seems reasonable and logically organized.

Figure 5. Possible colour schemes for Task Management application, sug-

gested by ChatGPT (represented by Figma).

The colour schemes are created primarily from neutral colours, these schemes
are helpful and seems suitable for the Task Management application.

In addition to that, the AI language assistant provided helpful guidance on how

to initiate a new React Native project with TypeScript and Expo, including step-
by-step instructions for setting up the development environment, configuring the
project, and creating the necessary files and directories.

When the decision about the components to be used in the application was
made, the ChatGPT was asked for assistance in creating and styling individual
components (Figure 6).
 16

Figure 6. Example of creating individual component with ChatGPT on the exam-

ple of Task List.

3.4 Components to be Implemented in the Application

The Task Management application was developed by creating a simple version

with a few screens and basic functionality, and all screens, elements, styles and
functions were created with the help of ChatGPT. In general, there are five
screens in the study project: Sign In, Sign Up, Task List, Task Form, Task De-
tails.

3.4.1 User Authentication and Authorization

The User Authentication and Authentication components are implemented

through two screens, the Sign In and Sign Up screens (Figure 7). These
screens are integrated with Firebase Authentication to provide a secure and ef-
ficient user authentication process.
 17

Figure 7: Sign In and Sign Up Screens of the Task Management application.

The implementation of the Authorization and Authentication of the application

started with the choice of technologies, where it was decided to use third-party
authentication service: Firebase Authentication. ChatGPT suggested a few
more options such as: AuthO, AWS Cognito and Okta, but since I have an ex-
perience working with Firebase Authentication, the decision was clear.

ChatGPT helped to create new Firebase project for the React Native application
with the comprehensive and detailed answer. Firebase Authentication offers a
set of backend services, SDKs, and pre-built UI libraries to authenticate users to
the application, and the email/password sign-in method was chosen for this pro-
ject. Firebase Authentication has intuitive and user-friendly interface (Figure 8).
 18

Figure 8: Firebase Authentication user interface

After Firebase project with the Authentication was created, ChatGPT was re-
quested to provide guidance on implementing Sign In and Sign Up screens in
React Native using Type Script with the Firebase Authentication. The main step
to initialize a Firebase project in a React Native mobile application is to create a
configuration object that contains the following fields: apiKey, authDomain, pro-
jectId, storageBucket, messagingSenderId, and appId. This information can be
easily obtained from Firebase Console.

The chatbot generated code for Sign In and Sign Up screens can be easily inte-
grated into the React Native project by copying and pasting the code (Figure 9).
 19

Figure 9. Sign In and Sign Up screen implementation in React Native generated

by ChatGPT.

The solution generated by ChatGPT for Sign In and Sign Up screens implemen-
tation in React Native has some disadvantages. One of them is that it does not
include import statements for the useState() React hook, which need to be
added manually in order for the code to work properly. Second one is the auth()
method, imported from react-native-firebase, is not compatible with Expo, which
I used for a smoother and faster development process. As a result, I had to use
the "firebase" library instead of "react-native-firebase". To avoid such issues in
the future, it is important to provide ChatGPT with more detailed information
about the project, such as how to implement Sign In and Sign Up screens in
React Native with TypeScript and Expo.

I encountered another problem while trying to initialize the Firebase project in

my React Native application. The solution provided by ChatGPT was to use the
command "firebase.initializeApp(firebaseConfig)" to initialize the Firebase pro-
ject in the React Native application. But there was an error: “TypeError: Cannot
read property 'initializeApp' of undefined”. Unfortunately, the Chatbot was una-
ble to provide a solution for this issue, so I had to search for answers online
 20

through Google. The decision was to refactor the code (Figure 10), when use
Firebase Modular API (v9), which is the newest version available.

Figure 10. Firebase initialization in React Native project (Firebase Modular API
(v9)).

After requesting assistance from chatbot to implement Sign In and Sign Out
screens, it provided recommendations for using Navigation in React Native to
improve the user experience of the application. This suggestion was very use-
ful.

The only issue with the advice provided by the chatbot on how to use Naviga-
tion in React Native was that it didn't include all the necessary code parts for the
implementation. As a result, when the code was copied and pasted into the ap-
plication project, an error occurred (“Cannot find name RootStackParamList”). It
was necessary to seek further information from the AI language model on how
to resolve this error (Figure 11).
 21

Figure 11. Prompt example.

The Chatbot also assisted with styling the Sign In and Sign Up screens. It of-
fered useful advice, such as using a simple colour scheme, ensuring that form
fields are easy to read, using contrasting colours for buttons, making sure but-
tons are large enough, and implementing responsive design.

3.4.2 Task Creation and Management

Task creation is implemented through the Task Form screen (Figure 12). The
Task Form screen is a key component of the Task Management application that
allows users to add new tasks to their task list. Task Form consists of two Text
Input components for entering task item and task description, as well as an Add
Task button for submitting the task. When the user submits a new task using
the Add Task button, the task information is sent to Firestore and stored as a
 22

new document in a 'tasks' collection. Each task document contains fields for the
task item and description, as well as metadata such as the creation date. This
data can then be retrieved and displayed in the Task List screen, allowing users
to see all their tasks.

Figure 12. Create Task screen.

During the development of the Task Form screen, assistance was requested
from ChatGPT to help with writing the code to add items to a Firestore collec-
tion.

Code example to add task to Firestore with React Native and TypeScript (Fig-
ure 13). In this example we are using db (const db = getFirestore(app)) with get-
Firestore() function from “firebase/firestore” library. Then, we're calling the add-
Doc() method on the “task”collection to add a new document with the task
 23

name, task description and creation timestamp. If the operation is successful,

we log a success message to the console. If there's an error, we log an error
message instead.

Figure 13. Code snippet to add task to Firestore.

Cloud-based NoSQL document database Firestore provides real-time updates

and synchronization between different devices and platforms, allowing for
seamless collaboration and data sharing. The chatbot provided clear instruc-
tions on how to create a Firestore collection within a Firebase application. The
Firestore interface is designed to be intuitive and user-friendly (Figure 14).

Figure 14. Firestore database user interface.

 24

3.4.3 Task List and Filtering

The Task List (Figure 15) screen is a one of the most important components of
the Task Management application that displays all the user's tasks in an orga-
nized and easily accessible manner. This screen includes a list of all the user's
tasks and simple filter that allows the user to easily view their completed and
uncompleted tasks. It also allows users to navigate to the Task Details screen
by clicking on a specific task. Completed tasks are displayed with a blue back-
ground while uncompleted tasks are displayed with a white background on the
screen. This helps users easily identify the status of their tasks. By clicking the
plus button in the right bottom corner on the Task List screen, users can navi-
gate to the Task Form screen to create a new task. Task List screen provides
an efficient and user-friendly way for users to manage their tasks.

Figure 15. Task List screen.

 25

The task items displayed in the Task List screen are fetched from Firestore,
providing real-time updates to the user's task list.

There is a React Hook within filter function (Figure 16). This Hook is triggered
whenever the status of the filter is changed and is responsible for returning a fil-
tered list of tasks based on the selected filter status. The “useEffect” Hook is a
built-in React Hook that enables developers to perform side effects after a com-
ponent has been rendered. In this case, the Hook is used to apply the appropri-
ate filter to the list of tasks based on the value of the “filterStatus” variable.

By utilizing the “useEffect” Hook, the filter function is automatically triggered

whenever the value of the “filterStatus” variable is updated. This ensures that
the filtered list of tasks is always up-to-date and reflects the current filter status.

Figure 16. React Hook

3.4.4 Task Details

The Task Details screen (Figure 17) provides users with detailed information
about a selected task, such as the task name and description. Users can also
mark the task as completed and delete the task from the list. The screen in-
cludes button for deleting the task.
 26

Figure 17. Task Details screen with examples of completed and uncompleted
task.

The task deletion feature has been implemented with Firestore, allowing users
to delete tasks from their task list with ease.

There is a Switch component in Task Detail screen, which is used to toggle the
completion status of a task (Figure 18). The Switch component is a built-in Re-
act Native component that provides a visual toggle control. The “trackColor”
prop is used to define the colours of the switch when it is in the 'off' and 'on' po-
sitions. The “thumbColor” prop is used to define the colour of the switch toggle.
In this case, the “thumbColor” changes based on whether the task is completed
or not.
 27

Figure 18. Switch component.

4 Results

4.1 Advantages of Using ChatGPT for Development

AI language models such as ChatGPT are becoming increasingly popular in the

world of software and mobile app development as they can provide developers
with a wide range of benefits.

ChatGPT can help mobile app developers improve their workflows. With an AI
language model, you can automate tasks, optimize coding practices, and iden-
tify potential areas for improvement. AI language models can also provide lan-
guage-specific recommendations and fixes to ensure code is accurate and func-
tional.

One of the main advantages of using ChatGPT in mobile application develop-

ment is its ability to understand and respond to natural language input. This
means that users can interact with the application in a more natural and intuitive
 28

way, which can lead to a more engaging and enjoyable user experience. With
ChatGPT, developers can create chatbots, voice assistants, or other NLP fea-
tures for their mobile applications that can understand user requests in natural
language and respond in a useful way. (Top benefits of ChatGPT in Mobile ap-
plication development, 2023).

Another important benefit is the ability to automate repetitive tasks. This can
help developers save time and focus on more complex and creative tasks such
as developing new features and improving the user experience. For example, AI
language models can automatically generate snippets of code based on user in-
put, reducing the time and effort required to write new code from scratch.

The chatbot can identify potential areas for optimization in the code. By analys-
ing large amounts of code, AI language models can help identify inefficiencies
or areas where performance can be improved. This can help developers opti-
mize their code and ensure it runs as efficiently as possible.

Additionally, AI language models can provide language-specific recommenda-

tions and corrections to ensure code accuracy and functionality. This can help
developers catch errors and improve code quality, leading to more stable and
reliable software applications.

ChatGPT has had a significant impact on mobile app development. Developers

are using this technology to build better and smarter chatbots that can under-
stand and respond to complex queries, provide personalized recommendations,
improve user engagement, and enhance customer support. (Saleem 2023).

4.2 Limitations of Using ChatGPT for Development

Despite the many improvements that Chat-GPT has made, it still has certain
limitations. One major limitation is that it cannot yet adapt to changes in conver-
sation or context. Although it understands very well what the user is saying and
generates an appropriate response, it does not always help to understand the
 29

changing context of the conversation. This may limit the success of the machine
conversation and may mean that the conversation may get stuck or cannot pro-
ceed any further. (Bansal 2023).

The chatbot can only provide answers from some time ago. ChatGPT data ends
in 2021. This can be a problem for those who don't know and can result in incor-
rect information being added when writing an essay or posting on a blog. The
fact is that AI can be useful when the user continues to use it with the utmost
caution. (The Limitations of ChatGPT 2023).

Sometimes ChatGPT gives incorrect answers, which may be due to insufficient

training. Stackoverflow, a coding website, has officially banned the output gen-
erated by chatbot as most of it was incorrect. ChatGPT often provides answers
that appear to be correct, even though they may be incorrect. Anyone who is
not an expert can take them for the truth. Even if the question is not particularly
difficult, ChatGPT can still give a very wrong answer. (The Limitations of
ChatGPT 2023).

Limited domain expertise. Domain expertise refers to the knowledge and experi-
ence in a particular field or industry. ChatGPT is a general-purpose language
model and may not have the same level of expertise as a human expert in a
specific domain. Developers can fine-tune ChatGPT on domain-specific data to
improve its domain expertise, but it requires a significant amount of data and
expertise.

4.3 Best Practices for Utilizing ChatGPT

The interesting thing about the ChatGPT is that you can have an ongoing con-
versation with it, and it will remember your previous interactions.

As an example, when I requested ChatGPT to assist me with integrating Fire-

base Authentication to my project without providing any additional details (Fig-
ure 19), it was able to recall from previous prompts that I was working on a
 30

React Native application using TypeScript, and as a result, it provided me with

code that was adapted to the correct language and framework.

Figure 19. Prompt Example.

To ensure that ChatGPT responds accurately and effectively, it is important to

provide it with clear and detailed prompts. For example, unclear prompt: "Help
me with my app”, in contrast clear prompt: "I'm developing a mobile task man-
agement application using React Native and Firebase. I need assistance with
implementing the Sign In and Sign Out screens, as well as creating a Firestore
collection for storing task data. Can you provide guidance on these tasks and
suggest best practices for improving the user experience of the application?"

If you want ChatGPT to assist you with adding a new feature to your applica-
tion, it is important to provide specific details such as the name of the feature,
 31

the functionality it should have, and any relevant constraints or limitations. Addi-
tionally, you could provide examples of similar features in other applications to
give ChatGPT a better understanding of your requirements. This will help en-
sure that the suggestions and advice provided by ChatGPT are relevant and
useful to your specific needs.

Another way to use ChatGPT as a developer helper when building a mobile ap-
plication is to integrate it into your Integrated Development Environment (IDE).
Many popular IDEs, such as Visual Studio Code, offer extensions that allow you
to use ChatGPT directly within your development environment.

Here is an example, how Code GPT extension can be used within Visual Studio
Code (Figure 18). It is possible to generate code using ChatGPT, simply add a
comment (here is “create typescript function to get completed tasks”) for the
task you wish to accomplish and press the “Ctrl + Shift + I” keys together.

Figure 20. Comment for the task you want accomplish.

This will generate the code on the right side (Figure 21).
 32

Figure 21. Code generated by ChatGPT.

Apart from that, it is possible to select a code, right-click on it, and ask Code
GPT to explain the code, refactor it, find problems, debug, and more (Figure
22).

Figure 20. How to utilize Code GPT within Visual Studio Code.

5 Conclusion

The purpose of the engineering work was to develop a mobile application for
the task management using React Native technology, Firebase Authentication
and Firestore with the ChatGPT assistance and at the same time to study the
 33

use of AI language model in the programmer's work, to examine the language

models used to create chatbots and to study the basic concepts of ChatGPT.

In the research phase, various materials, and articles about ChatGPT were ex-
plored to gain a comprehensive understanding of language models and artificial
intelligence. The history of ChatGPT and its creators were also studied, provid-
ing valuable insights into its capabilities and potential applications.

Artificial intelligence is developing more and more every day, and ChatGPT is a
clear proof of this. This versatile chatbot can already compete with humans in
intellectual and communicative tasks. The use of ChatGPT is expanding, and
this trend will continue.

The process of Task Management application development went relatively

smoothly due to the previous familiarity with React Native technologies and
Firebase fundamentals. ChatGPT was a helpful tool during development,
providing ready-made code snippets. However, there were also various errors
encountered that ChatGPT was unable to assist with resolving. In general,
ChatGPT has the potential to improve developer’s workflow and increase effi-
ciency but requires some field knowledge to be fully effective.

ChatGPT is a good programmer but not a good software designer or developer.

This AI cannot design software from an engineer's perspective with business
needs in mind or fix errors other than those detected in its training data. At least
not for now. This tool cannot replace those whose work goes beyond routine
coding. On the contrary, this tool can open a wide range of possibilities for
those who work with the code. Just need to give the right instructions to this AI
language assistant and it can make the workflow easier. This leads us to the fol-
lowing conclusion: ChatGPT is the beginning of more powerful developers.
 34

References

A Guide to OpenAI and its Products and Services. 2023. Online. The Windows
Club. <https://www.thewindowsclub.com/a-guide-to-openai-and-its-products-
and-services>. 9.1.2023. Accessed 7.2.2023.

Bansal, Yash. 2023. Limitations of Chat-GPT. Online. Medium. <https://me-

dium.com/@yashbansal42/the-limitations-of-chat-gpt-and-the-need-for-further-
research-and-development-91bc6caae533>.
29.3.2023. Accessed 25.4.2023.

Bard, Jonah. 2021. How GPT-3 Actually Works, From the Ground Up. Online.
Medium. <https://medium.com/nerd-for-tech/how-gpt-3-actually-works-from-the-
ground-up-5714ae7f3355>. 9.3.2021. Accessed 25.2.2023.

Borozenets, Michael. 2022. React Native Init vs Expo 2022: What Are the Dif-
ferences? Online. Fulkrum. < https://fulcrum.rocks/blog/react-native-init-vs-
expo#what-is-expo>. 22.4.2022. Accessed 14.2.2023.

Cooper, Kindra. 2021. OpenAI GPT-3: Everything You Need to Know. Online.
Springboard. <https://www.springboard.com/blog/data-science/machine-learn-
ing-gpt-3-open-ai/>. 1.11.2021. Accessed 8.2.2023.

Cloud Firestore. 2023. Online. Firebase Documentation. <https://fire-

base.google.com/products/firestore>. 13.4.2023. Accessed 17.4.2023.

Cuofano, Gennaro. 2023. Instruct GPT And Why It Matters For The Success
Of ChatGPT. Online. FourWeekMBA. <https://fourweekmba.com/in-
structgpt/>. 13.2.2023. Accessed 25.2.2023.

Dongre, Anay. 2023. Discover the Revolutionary Instruct GPT. Online. OpenGe-
nius IQ. <https://iq.opengenus.org/instruct-gpt/>. 2023. Accessed
25.2.2023.

Eisenman, Bennie. 2015. Learning React Native. E-book. O’Reilly.

Firebase Authentication. 2023. Online. Firebase Documentation. <https://fire-

base.google.com/docs/auth>. 13.4.2023. Accessed 17.4.2023.

Frankenfield, Jake. 2022. Artificial Intelligence: What It Is and How It Is

Used. Online. Investopedia. <https://www.investopedia.com/terms/a/artificial-in-
telligence-ai.asp>. 6.7.2022. Accessed 7.2.2023.

History of Artificial Intelligence. 2020. Online. Council of Europe.

<https://www.coe.int/en/web/artificial-intelligence/history-of-ai>. 6.7.2020. Ac-
cessed 7.2.2023.
 35

Kapronczay, Mór. 2022. A Beginner’s Guide to Language Models. Online.

BuiltIn. <https://builtin.com/data-science/beginners-guide-language-models>.
13.12.2022. Accessed 7.2.2023.

Lowe, Ryan & Wainwright, Caroll. 2022. Instruct GPT Model Card. Online.
GitHub. <https://github.com/openai/following-instructions-human-feed-
back/blob/main/model-card.md>. 1.2022. Accessed 25.2.2023.

Maldonado, Leonardo. 2019. A Short Introduction to TypeScript. Online. Pro-

gress Telerik. <https://www.telerik.com/blogs/a-short-introduction-to-typescript>.
26.8.2019. Accessed 17.4.2023.

Ot, Annina. 2022. What Is OpenAI and Does It Really Make Coding Easier?
Online. MUO. <https://www.makeuseof.com/openai-coding/>. 19.1.2022. Ac-
cessed 7.2.2023.

Pocock, Gennaro. 2023. What Is ChatGPT? – what is it used for? Online.

PCGuide. <https://www.pcguide.com/apps/what-is-chat-gpt/>. 13.2.2023. Ac-
cessed 25.2.2023.

Reyes, Kate. 2023. What is Deep Learning and How Does It Works. Online.
SimpliLearn. <https://www.simplilearn.com/tutorials/deep-learning-tutorial/what-
is-deep-learning/>. 17.2.2023. Accessed 7.5.2023.

Roldós, Inés. 2020. NLP, Machine Learning & AI, Explained. Online. Mon-
keyLearn. <https://monkeylearn.com/blog/nlp-ai/>. 9.6.2020. Accessed
25.2.2023.

Saleem, Aaliya. How ChatGPT Impact On Mobile App Development. 2023.

Online. Medium. < https://mobileappcircular.com/how-chat-gpt-impact-on-mo-
bile-app-development-d5bdc37b0513>. 15.3.2023 Accessed 25.4.2023.

The Limitations of ChatGPT. 2023. Online. Nandbox. <https://nandbox.com/the-

limitations-of-chatgpt/>. 23.1.2023 Accessed 25.4.2023.

Top benefits of ChatGPT in Mobile application development. 2023. Online. Sieg

Partners. <https://www.siegpartners.com/top-benefits-of-chatgpt-in-mobile-ap-
plication-development/>. Accessed 25.4.2023.

Wodeski, Ben. 2022. 7 language models you need to know. Online. AI Busi-
ness. <https://aibusiness.com/nlp/7-language-models-you-need-to-know>.
27.7.2022. Accessed 25.2.2023.

What is GPT-3, How Does It Work, and What Does It Actually Do? 2021.
Online. Medium. <https://medium.com/sciforce/what-is-gpt-3-how-does-it-work-
and-what-does-it-actually-do-9f721d69e5c1>. 17.9.2021 Accessed 7.2.2023.