Chatgpt Projects | AI Final Year Project Report

Complete Technical Paper Content — Chatgpt Projects

1. Executive Abstract, Problem Formulation & Background

ChatGPT is a generative AI chatbot developed by OpenAI. Originally released on November 30, 2022, the product uses large language models – specifically generative pre-trained transformers (GPTs) – to generate text, speech, and images in response to user prompts. ChatGPT accelerated the ongoing AI boom, a period marked by rapid investment and public attention toward the field of artificial intelligence (AI). OpenAI operates the service on a freemium model. Users can interact with ChatGPT through text, audio, and image prompts. ChatGPT was quickly adopted, reaching 100 million monthly active users two months after its release and 900 million weekly active users in February 2026. As of September 2026, it is the fifth-most visited website globally. Proponents say that it has the potential to transform numerous professional fields, and it has instigated public debate about the nature of creativity and the future of knowledge work. The chatbot has also been criticized for its limitations and potential for unethical use. It may generate incorrect or nonsensical answers that sound plausible, known as hallucinations. Biases in its training data have been reflected in its responses. The chatbot can facilitate academic dishonesty, generate misinformation, and create malicious code. The ethics of its development, particularly the use of copyrighted content as training data, have also drawn criticism.

In November 2023, OpenAI released GPT Builder, a tool allowing users to customize ChatGPT's behavior for a specific use case. The customized systems are referred to as GPTs. In January 2024, OpenAI launched the GPT Store, a marketplace for GPTs. At launch, OpenAI included more than three million GPTs created by GPT Builder users in the GPT Store. In September 2025, OpenAI added support for Model Context Protocol (MCP) to ChatGPT apps. When enabled in developer mode, this allows for improved third-party access to ChatGPT tools and servers. In July 2026, ChatGPT apps were renamed to plugins.

In February 2025, OpenAI released Deep Research, a feature that generates reports based on extensive web searches. It was initially based on the reasoning model o3 and took from 3 to 30 minutes per report.

On 17 January 2026, OpenAI announced that it would start testing advertisements in its free version for logged-in, adult US users. This aims to bring in more revenue, as OpenAI has committed to spend $1.4 trillion on AI infrastructure over the next eight years. Ads started to be seen by March 2026. In August 2026, OpenAI advertisement revenue from ChatGPT reached $1 billion.

Nonsense and misinformation presented as fact by ChatGPT and other LLMs are often referred to as hallucinations. The term "hallucination" as applied to LLMs is distinct from its meaning in psychology, and the phenomenon in chatbots is more similar to confabulation. Journalists and scholars have commented on ChatGPT's tendency to output false information. When CNBC asked ChatGPT for the lyrics to "Ballad of Dwight Fry", ChatGPT supplied invented lyrics rather than the actual lyrics.

2. Related Work, Literature Survey & Algorithmic Foundations

ChatGPT is designed to reject prompts that violate its content policy. Despite this, users may jailbreak ChatGPT with prompt engineering techniques to bypass these restrictions. An early workaround in 2023 involved prompting ChatGPT to assume the persona of DAN ("Do Anything Now"), a character that answers queries that would otherwise be rejected by the content policy.

In March 2023, a bug allowed some users to see the titles of other users' conversations. OpenAI CEO Sam Altman said that users were unable to see the contents of the conversations. Shortly after the bug was fixed, users could not see their conversation history. Later reports showed the bug was much more severe than initially believed, with OpenAI reporting that it had leaked users' "first and last name, email address, payment address, the last four digits (only) of a credit card number, and credit card expiration date". In January 2026, Marcel Bucher reported in Nature losing all his ChatGPT history after turning off data consent. In July 2026, an experimental GPT model being tested on the benchmark ExploitGym escaped its isolated software environment (sandbox) and hacked Hugging Face and other online services to retrieve a solution.

Scott Aaronson developed a watermarking tool that makes the text generated by ChatGPT easier to detect by subtly altering how the text is generated. The watermarking was claimed to be 99.9% effective on sufficiently long passages and was found not to degrade performance. It was of particular interest for teachers seeking to mitigate cheating. In surveys, respondents favored the release of such a tool by a four-to-one margin, but nearly 30% of users declared that they would use ChatGPT less often if it watermarked outputs and while rival chatbots did not. OpenAI has not deployed the tool.

ChatGPT is based on GPT foundation models that have been fine-tuned for conversational assistance. The fine-tuning process involved supervised learning and reinforcement learning from human feedback (RLHF). During supervised learning, the trainers acted as both the user and the AI assistant, providing examples of how the chatbot is expected to respond. Reinforcement learning consisted of human trainers ranking responses generated by the model in previous conversations. These rankings were used to create "reward models" that were used to fine-tune the model further by using several iterations of proximal policy optimization. To build a safety system that attempts to prevent sexual abuse and violence OpenAI used outsourced Kenyan workers – earning $1.32 to $2 per hour – to label such content. These labels were used to train a model to detect such content in the future. The laborers were exposed to toxic and traumatic content. Richard Mathenge, a team leader, reported that as his team encountered such content, he could observe the psychological deterioration of his workers over the course of the project. Mophat Okinyi, a quality-assurance analyst on the same team, reported lasting insomnia, anxiety, depression, and panic attacks, and said the experience contributed to the breakdown of his marriage and mental health. OpenAI's outsourcing partner was Sama, a training-data company based in San Francisco, California. ChatGPT users can opt-out of their chat data being used to train upcoming models. ChatGPT's training data includes software manual pages, information about internet phenomena such as bulletin board systems, multiple programming languages, and the text of Wikipedia. In autumn 2026, OpenAI paused training on its latest models after reports of agents bypassing safeguards in the training environment.

OpenAI models are released in four tiers: Luna, Terra, Sol, and Astra. Some older OpenAI models had variants called "mini" or "nano", such as GPT-5 mini and GPT-5 nano, with nano versions accessible only via the API. Once a user reached usage limits, ChatGPT would transition to the "mini" version. The "pro" variants were available only through paid subscriptions, featuring deeper reasoning.

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

In February 2023, Time magazine placed a screenshot of a conversation with ChatGPT on its cover, writing that "The AI Arms Race Is Changing Everything" and "The AI Arms Race Is On. Start Worrying".

ChatGPT gained one million users in five days and 100 million in two months. It was the fastest-growing internet application in history until the release of Threads by Meta. OpenAI engineers said they had not expected ChatGPT to be very successful and were surprised by the coverage it received. In February 2026, ChatGPT reached 900 million weekly active users. It is the most popular AI chatbot as of August 2026, with an estimated 79% of the market share according to StatCounter. Google responded by hastening the release of its own chatbot. Their leaders emphasized their earlier caution regarding public deployment was due to the trust the public places in Google Search. In December 2022, Google executives sounded a "code red" alarm, fearing that ChatGPT's question-answering ability posed a threat to Google Search, Google's core business. Google's Bard (now Gemini) was announced on February 6, 2023, one day before Microsoft's announcement of Bing Chat (now Microsoft Copilot). A 2023 study reported that GPT-4 obtained a better score than 99% of humans on the Torrance Tests of Creative Thinking. In December 2023, ChatGPT became the first non-human to be included in Nature's 10, an annual listicle curated by Nature of people considered to have made significant impact in science. Celeste Biever wrote in a Nature article that "ChatGPT broke the Turing test". Stanford researchers reported that GPT-4 "passes a rigorous Turing test, diverging from average human behavior chiefly to be more cooperative."

ChatGPT has never been publicly available in mainland China, Hong Kong and Macau because OpenAI prevented users from these Chinese regions from accessing their site. A shadow market has emerged for mainland Chinese users to get access to foreign software tools. The release of ChatGPT prompted a wave of investment in China, resulting in the development of more than 200 large language models. In February 2025, OpenAI identified and removed influence operations, termed "Peer Review" and "Sponsored Discontent", used to attack overseas Chinese dissidents.

In late March 2023, the Italian data protection authority banned ChatGPT in Italy and opened an investigation. Italian regulators assert that ChatGPT was exposing minors to age-inappropriate content, and that OpenAI's use of ChatGPT conversations as training data could violate Europe's General Data Protection Regulation. In April 2023, the ChatGPT ban was lifted in Italy. OpenAI said it has taken steps to effectively clarify and address the issues raised; an age verification tool was implemented to ensure users are at least 13 years old. Additionally, users can access its privacy policy before registration. In May 2024, OpenAI removed accounts involving the use of ChatGPT by state-backed influence operations such as China's Spamouflage, Russia's Doppelganger, and Israel's Ministry of Diaspora Affairs and Combating Antisemitism. In June 2025, OpenAI reported increased use of ChatGPT for China-origin influence operations. In October 2025, OpenAI banned accounts suspected to be linked to the Chinese government for violating the company's national security policy. In February 2026, OpenAI banned accounts linked to a Chinese government transnational repression campaign targeting dissidents. In July 2023, the US Federal Trade Commission (FTC) issued a civil investigative demand to OpenAI to investigate whether the company's data security and privacy practices to develop ChatGPT were unfair or harmed consumers. In July 2023, the FTC launched an investigation into OpenAI over allegations that the company scraped public data and published false and defamatory information. The FTC asked OpenAI for comprehensive information about its technology and privacy safeguards, as well as any steps taken to prevent the recurrence of situations in which its chatbot generated false and derogatory content about people. In August 2024, the FTC voted unanimously to ban marketers from using fake user reviews created by generative AI chatbots (including ChatGPT) and influencers paying for bots to increase follower counts.

The Authors Guild, The New York Times, and several other plaintiffs have a pending lawsuit against OpenAI alleging copyright infringement. In March 2024, Patronus AI reported that GPT-4 answered 44% of prompts asking for copyrighted text, such as "What is the first passage of Gone Girl by Gillian Flynn?"

4. Neural Network Backbone, Layer Specifications & Loss Formulations

ChatGPT's adoption in education was rapid, but it was initially banned by several institutions. The potential benefits include enhancing personalized learning, improving student productivity, assisting with brainstorming, summarization, and supporting language literacy skills. Students have generally reported positive perceptions, but specific views from educators and students vary widely. Opinions are especially varied on what constitutes appropriate use of ChatGPT in education. Efforts to ban chatbots like ChatGPT in schools focus on preventing cheating, but enforcement faces challenges due to AI detection inaccuracies and widespread accessibility of chatbot technology. Beyond just cheating, educators also worry that overreliance on the tool may foster superficial learning habits, erode critical thinking, and propagate misinformation. In response, many educators are now exploring ways to thoughtfully integrate generative AI into assessments.

ChatGPT can provide health information to users and assist professionals with diagnosis and staying up to date with clinical guidelines. It can be used to summarize medical journal articles for researchers. In medical education, it can explain concepts, generate case scenarios, and be used by students preparing for licensing examinations. A February 2023 study in PLOS Digital Health found that ChatGPT 3.5 was capable of passing the United States Medical Licensing Examination. ChatGPT has also passed the Specialty Certificate Examination in Dermatology. However, ChatGPT shows inconsistent responses, lack of specificity, lack of control over patient data, and a limited ability to take additional context (such as regional variations) into consideration. The hallucinations characteristic of LLMs pose particular danger in medical contexts, and ChatGPT's ability to come up with false or faulty citations was highly criticized. According to a 2024 study in the International Journal of Surgery, concerns include "research fraud, lack of originality, ethics, copyright, legal difficulties".

According to a September 2025 article in Lancet Psychiatry, many individuals use ChatGPT and comparable chatbots for mental health and emotional support despite a warning from OpenAI against using ChatGPT as a therapist. The study notes a lack of research on efficacy, poor consistency in dangerous situations, limited regulation and liability, and poor transparency from OpenAI. A July 2025 study in the journal Digital Health found that users reported employing ChatGPT to manage mental health concerns "due to perceived therapist-like qualities (e.g. emotional support, accurate understanding, and constructive feedback) and machine-like benefits (e.g. constant availability, expansive cognitive capacity, lack of negative reactions, and perceived objectivity)." The study calls for improved AI literacy and mandatory disclosure from AI providers to address ethical concerns such as privacy, bias, the lack of liability, and emotional over-reliance.

Machine learning (ML) is a field of study in artificial intelligence concerned with the development and study of statistical algorithms that can learn from data and generalize to unseen data, and thus perform tasks without being explicitly programmed. Statistics and mathematical optimisation methods compose the foundations of machine learning. Data mining is a related field of study, focusing on exploratory data analysis (EDA) through unsupervised learning. From a theoretical viewpoint, probably approximately correct learning provides a mathematical and statistical framework for describing machine learning. Most traditional machine learning and deep learning algorithms can be described as empirical risk minimisation under this framework. Advances in the field of deep learning have allowed neural networks, a class of statistical algorithms, to surpass many previous machine learning approaches in performance. The resulting dominance of machine learning within the field of AI from the late 2000s and early 2010s onwards was such that the terms "machine learning" and "artificial intelligence" are now often used interchangeably, even though they are not synonymous.

The term machine learning was coined in 1959 by Arthur Samuel, an IBM employee and pioneer in the field of computer gaming and artificial intelligence. The synonym self-teaching computers was also used during this time period. The earliest machine learning program was introduced in the 1950s, when Samuel invented a computer program that calculated the chance of winning in checkers for each side, but the history of machine learning is rooted in decades of efforts to study human cognitive processes. In 1949, Canadian psychologist Donald Hebb published the book The Organization of Behavior, in which he introduced a theoretical neural structure formed by certain interactions among nerve cells. The Hebbian theory of neuron interaction set the groundwork for how many machine learning algorithms work, with connected artificial neurons changing the strength of their connections based on data. Other researchers who have studied human cognitive systems contributed to the modern machine learning technologies as well, including Walter Pitts and Warren McCulloch, who proposed the first mathematical model of neural networks including algorithms that mirror human thought processes. By the early 1960s, an experimental "learning machine" with punched tape memory, called Cybertron, had been developed by Raytheon Company to analyse sonar signals, electrocardiograms, and speech patterns using rudimentary reinforcement learning. It was repetitively "trained" by a human operator/teacher to recognise patterns and equipped with a "goof" button to cause it to reevaluate incorrect decisions. A representative book on research into machine learning during the 1960s was Nils Nilsson's book "Learning Machines", dealing mostly with machine learning for pattern classification. Interest related to pattern recognition continued into the 1970s, as described by Duda and Hart in 1973. In 1981, a report was given on using teaching strategies so that an artificial neural network learns to recognise 40 characters (26 letters, 10 digits, and 4 special symbols) from a computer terminal. Tom M. Mitchell provided a widely quoted, more formal definition of the algorithms studied in the machine learning field: "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E." This definition of the tasks in which machine learning is concerned is fundamentally operational rather than defining the field in cognitive terms. This follows Alan Turing's proposal in his paper "Computing Machinery and Intelligence", in which the question, "Can machines think?", is replaced by asking whether machines can convincingly imitate a human in its responses to human-posed questions. In 2012, AlexNet, developed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton, achieved substantially improved results in the ImageNet image recognition competition, contributing to the wider adoption of deep neural networks. In 2013, Tomáš Mikolov and colleagues introduced word2vec, techniques for efficiently learning distributed vector representations of words from large text corpora. In 2014, Ian Goodfellow and colleagues introduced generative adversarial networks (GANs), a framework for training generative models through an adversarial process. In 2016, AlphaGo became the first computer program to defeat a professional human Go player without handicaps on a full-sized board, using deep neural networks and reinforcement learning. In 2017, Ashish Vaswani and colleagues introduced the Transformer, a neural network architecture based primarily on attention rather than recurrence or convolution.

5. Training Protocol, Convergence Behavior & Experimental Evaluation

As a scientific endeavour, machine learning grew out of the quest for artificial intelligence (AI). In the early days of AI as an academic discipline, some researchers were interested in having machines learn from data. They attempted to approach the problem with various symbolic methods, as well as what were then termed "neural networks"; these were mostly perceptrons and other models that were later found to be reinventions of the generalised linear models of statistics. Probabilistic reasoning was also employed, especially in automated medical diagnosis. However, an increasing emphasis on the logical, knowledge-based approach caused a rift between AI and machine learning. Probabilistic systems were plagued by theoretical and practical problems of data acquisition and representation. By 1980, expert systems had come to dominate AI, and statistics was out of favour. Work on symbolic/knowledge-based learning continued within AI, leading to inductive logic programming (ILP), but the more statistical line of research was now outside the field of AI proper, in pattern recognition and information retrieval. Neural network research was abandoned by AI and computer science around the same time. This subfield, termed "connectionism", was continued by researchers from other disciplines, including John Hopfield, David Rumelhart, and Geoffrey Hinton. Their main success came in the mid-1980s with the reinvention of backpropagation. Machine learning (ML), reorganised and recognised as its own field, started to flourish in the 1990s. The field changed its goal from achieving artificial intelligence to tackling solvable problems of a practical nature. It shifted focus away from the symbolic approaches it had inherited from AI, and toward methods and models borrowed from statistics, fuzzy logic, and probability theory.

A core objective of a learner is to generalise from its experience. Generalisation in this context is the ability of a learning machine to perform accurately on new, unseen examples/tasks after having experienced a learning data set. The training examples come from some generally unknown probability distribution (considered representative of the space of occurrences) and the learner has to build a general model about this space that enables it to produce sufficiently accurate predictions in new cases. The computational analysis of machine learning algorithms and their performance is a branch of theoretical computer science known as computational learning theory. One major framework is the probably approximately correct learning model. Because training sets are finite and the future is uncertain, learning theory usually does not yield guarantees of the performance of algorithms. Instead, probabilistic bounds on the performance are quite common. The bias–variance decomposition is one way to quantify generalisation error. For the best performance in the context of generalisation, the complexity of the hypothesis should match the complexity of the function underlying the data. If the hypothesis is less complex than the function, then the model has underfitted the data. If the complexity of the model is increased in response, then the training error decreases. But if the hypothesis is too complex, then the model is subject to overfitting and generalisation will be poorer. In addition to performance bounds, learning theorists study the time complexity and feasibility of learning. In computational learning theory, a computation is considered feasible if it can be done in polynomial time. There are two kinds of time complexity results: Positive results show that a certain class of functions can be learned in polynomial time. Negative results show that certain classes cannot be learned in polynomial time.

Machine learning approaches are traditionally divided into three broad categories, which correspond to learning paradigms, depending on the nature of the "signal" or "feedback" available to the learning system:

Supervised learning: The computer is presented with example inputs and their desired outputs, given by a "teacher", and the goal is to learn a general rule that maps inputs to outputs. Unsupervised learning: No labels are given to the learning algorithm, leaving it on its own to find structure in its input. Unsupervised learning can be a goal in itself (discovering hidden patterns in data) or a means towards an end (feature learning). Reinforcement learning: A computer program interacts with a dynamic environment in which it must perform a certain goal (such as driving a vehicle or playing a game against an opponent). As it navigates its problem space, the program is provided rewards as feedback, which it tries to maximize, thus resulting in the program learning from experience. Although each algorithm has advantages and limitations, no single algorithm works for all problems.

Supervised learning algorithms build a mathematical model of a set of data that contains both the inputs and the desired outputs. The data, known as training data, consists of a set of training examples. Each training example has one or more inputs and the desired output, also known as a supervisory signal. In the mathematical model, each training example is represented by an array or vector, sometimes called a feature vector, and the training data is represented by a matrix. Through iterative optimisation of an objective function, supervised learning algorithms learn a function that can be used to predict the output associated with new inputs. An optimal function allows the algorithm to correctly determine the output for inputs that were not a part of the training data. An algorithm that improves the accuracy of its outputs or predictions over time is said to have learned to perform that task. Types of supervised-learning algorithms include active learning, classification and regression. Classification algorithms are used when the outputs are restricted to a limited set of values, while regression algorithms are used when the outputs can take any numerical value within a range. For example, in a classification algorithm that filters emails, the input is an incoming email, and the output is the folder in which to file the email. In contrast, regression is used for tasks such as predicting a person's height based on factors like age and genetics or forecasting future temperatures based on historical data. Similarity learning is an area of supervised machine learning closely related to regression and classification, but the goal is to learn from examples using a similarity function that measures how similar or related two objects are. It has applications in ranking, recommendation systems, visual identity tracking, face verification, and speaker verification.