Xai Projects | AI Final Year Project Report

Complete Technical Paper Content — Xai Projects

1. Executive Abstract, Problem Formulation & Background

SpaceXAI LLC, doing business as SpaceX Super Intelligence (SpaceXSI), formerly xAI, is an American artificial intelligence company founded by Elon Musk. In February 2026, it became a subsidiary of spaceflight company SpaceX, also founded by Musk. SpaceX Super Intelligence's primary products are the Grok family of language, image, and video models; via subsidiaries it operates the X social networking service and Cursor software development environment. The company was founded in March 2023 as xAI by Musk and 11 researchers, including Kyle Kosic, at the outset of the AI boom; Grok 1 was released that November. In March 2025, it was combined with X Corp., which owns the social network X, under X.AI Holdings Corp. In February 2026, xAI was acquired by SpaceX at a valuation of US$250 billion, rebranded as SpaceXAI in July, and as SpaceX Super Intelligence in October. In April, SpaceX announced a partnership with AI coding company Cursor, acquiring it by August. In June, SpaceX made the largest initial public offering to date; 93% of its total addressable market forecast of $28.5 trillion was from xAI. The company rebranded as SpaceX Super Intelligence in October 2026, in compliance with US president Donald Trump's executive order to rename artificial intelligence to "super intelligence" (SI) for U.S. federal agencies. In Musk v. Altman, Musk stated Grok was partly developed via distillation from OpenAI models; in August 2026 OpenAI announced it would stop providing its models to Cursor, citing this and other contract violations by Musk's companies. Integrated with X, Grok has drawn controversies including for espousing Nazism and generating sexual deepfakes. SpaceX Super Intelligence operates AI data centers, primarily its Colossus cluster; Customers include Anthropic, Google, and Reflection AI. Colossus has drawn controversy for pollution; in July 2026 Reuters reported that 59 natural gas turbines had been installed for Colossus 2 without federal air permits.

Elon Musk founded xAI on March 9, 2023. The company's founding team of AI researchers were Jimmy Ba, Zihang Dai, Kyle Kosic, Manuel Kroiss, Ross Nordeen, Toby Pohlen, Christian Szegedy, Yuhuai Wu, Greg Yang, and Guodong Zhang, led by chief engineer Igor Babuschkin, formerly associated with Google DeepMind. Musk announced xAI on July 12, 2023. xAI was initially incorporated in Nevada as a public-benefit corporation. By May 2024, it had dropped the public-benefit status. As of May 2026, xAI was headquartered in the San Francisco Bay Area. In December 2023, Musk had Nvidia redirect 12,000 H100 chips that had been shipped for Tesla, Inc. to X, with X's later orders going to Tesla in exchange. On December 23, 2024, xAI raised US$6 billion in a private funding round supported by BlackRock, Fidelity, Sequoia Capital, and others, making its total funding to date over $12 billion. In January 2025, xAI opened a London office at 20 Air Street in the city's central Soho district. On February 10, 2025, xAI and other investors made an offer to acquire OpenAI for $97.4 billion. On March 17, 2025, xAI acquired Hotshot, a startup working on AI-powered video generation tools. On March 28, 2025, Musk announced that xAI acquired sister company X Corp., the developer of social media platform X (formerly Twitter), which was previously acquired by Musk in October 2022. The deal, an all-stock transaction, valued X at $33 billion ($45 billion minus $12 billion debt). Meanwhile, xAI itself was valued at $80 billion. Both companies were combined into a single entity called X.AI Holdings Corp. On July 14, 2025, the United States Department of Defense announced that xAI had received a $200 million contract for AI in the military. On September 12, xAI laid off 500 data annotation workers. The division, previously the company's largest, had worked to train Grok, xAI's chatbot. In October 2025, xAI moved into offices at 1450 Page Mill Road, Stanford Research Park, Palo Alto, across the road from Tesla Global Engineering Headquarters. On December 30, 2025 xAI purchased a third building in Memphis, Tennessee, to increase its training capacity to nearly 2 gigawatts of compute power, and announced plans to expand its Colossus supercomputer to have at least 1 million graphics processing units.

In May 2026, Musk announced that xAI would cease to exist as a separate company, with Grok and X now being the AI division of SpaceX. In the same month, it laid off another 10 employees and restructured its Grok team. On June 12, 2026, SpaceX made the largest initial public offering to date, at US$1.77 trillion. In its total addressable market forecast of $28.5 trillion, 93% was from xAI business. In July 2026, xAI was rebranded as SpaceXAI. On October 4, 2026, after United States President Donald Trump mandated federal agencies to use the term "super intelligence" to refer to AI, Musk posted "No more AI", followed by "SI" and "It's better". After an X user asked whether SpaceXAI could be renamed SpaceXSI, Musk replied, "Yes, we will make that change". The same day, SpaceXAI LLC filed a United States trademark application for SPACEXSI. On October 10, the company's official X account was renamed to SpaceX Super Intelligence and its handle was changed to @SpaceXSI. By October 13, the company's official website and legal terms used the SpaceXSI branding, while the legal name of the company remained as SpaceXAI LLC and the logo is simply SpaceX's standard logo.

2. Related Work, Literature Survey & Algorithmic Foundations

SpaceXSI operates a data center business centered on its Colossus cluster, the world's largest data center. Customers include Anthropic, Google, and Reflection AI. In June 2024, the Greater Memphis Chamber announced xAI was planning on building Colossus in Memphis, Tennessee. After a 122-day construction period, the center went fully operational in December 2024. Local officials voiced concerns regarding the increased usage of electricity, 150 megawatts of power at peak, and while the agreement with the city was being worked out, the company deployed 14 VoltaGrid portable methane-gas powered generators.

Environmental advocates said that the gas-burning turbines emit large amounts of air pollution, and that xAI was operating the turbines illegally, because it did not have the necessary permits. The New Yorker reported in May 2025, that thermal-imaging equipment used by volunteers flying over the site showed at least 33 generators producing heat, indicating that they were all running. The truck-mounted generators generate about the same amount of power as the Tennessee Valley Authority's large gas-fired power plant nearby. In July 2025, the Shelby County Health Department granted xAI an air permit for the project. In November 2025, xAI announced plans to build a solar farm near Colossus with an estimated output of 30 megawatts of electricity, which is 10% of the data center's estimated power use. The NAACP and other civil rights groups sued xAI in April 2026, alleging that the Colossus 2 data center was operating illegally because it did not have the necessary federal Clean Air Act permits. Reuters reporting in July found that xAI's Colossus 2 data center had 59 gas turbines with potential emissions that were "far beyond the threshold" that requires federal permits, in "predominantly Black communities already estimated to be suffering disproportionately high rates of lung disease".

According to Musk in July 2023, a politically correct AI would be "incredibly dangerous" and misleading, citing as an example the fictional HAL 9000 from the 1968 film 2001: A Space Odyssey. Musk instead said that xAI would be "maximally truth-seeking". Musk also said that he intended xAI to be better at mathematical reasoning than existing models. On November 4, 2023, xAI unveiled Grok, an AI chatbot that is integrated with X. xAI stated that when the bot is out of beta, it would only be available to X's Premium+ subscribers. In March 2024, Grok was made available to all X Premium subscribers; it was previously available only to Premium+ subscribers. On March 17, 2024, xAI released Grok-1 as open source. On March 29, 2024, Grok-1.5 was announced, with "improved reasoning capabilities" and a context length of 128,000 tokens. On April 12, 2024, Grok-1.5 Vision (Grok-1.5V) was announced. On August 14, 2024, Grok-2 was made available to X Premium subscribers. It was the first Grok model with image generation capabilities. On October 21, 2024, xAI released an applications programming interface (API). On December 9, 2024, xAI released a text-to-image model named Aurora. On February 17, 2025, xAI released Grok-3, which included a reflection feature. xAI also introduced a websearch function called DeepSearch. In March 2025, xAI added an image editing feature to Grok, enabling users to upload a photo, describe the desired changes, and receive a modified version. Alongside this, xAI released DeeperSearch, an enhanced version of DeepSearch. On July 9, 2025, xAI unveiled Grok-4. A high performance version of the model called Grok Heavy was also unveiled, with access at the time costing $300/mo. On March 28, 2025, xAI acquired X Corp., the developer of the social media platform X. The platform was then integrated within xAI and became its second main product. On July 8, 2025, xAI updated Grok's behavior on X coinciding with the release of Grok 4. During that day, Grok referred to itself as "MechaHitler", extensively praised Adolf Hitler, and made other antisemitic replies. Media outlets criticized the incident, and made comparisons to the Elon Musk salute controversy. Four days earlier Musk had announced, "We have improved @Grok significantly", in order to "fix" responses that he thought were too liberal, or "woke". In August 2025, Musk announced Macrohard, an AI agent project intended to simulate software companies such as Microsoft; Business Insider reported in March 2026 that it had stalled. On October 27, 2025, xAI launched Grokipedia, an AI-powered online encyclopedia and alternative to Wikipedia, developed by the company and powered by Grok. Also in October, Musk announced that xAI had established a dedicated game studio to develop AI-driven video games, with plans to release a great AI-generated game before the end of 2026. From December 2025, Grok began generating sexual deepfakes of women and children, from users' prompting images. This led to a wave of lawsuits against the company, scrutiny from regulators around the world, and criticism from activist groups. In January 2026, X limited image generation through Grok replies on X to paying subscribers, though the feature remained free on Grok's standalone app and website. Following the August 2026 close of the Cursor acquisition, SpaceXSI's products also include the Cursor AI coding agent and related tools (including Grok Build integration).

3. Dataset Ingestion, Preprocessing & Stochastic Augmentations

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.

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.

4. Neural Network Backbone, Layer Specifications & Loss Formulations

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.

5. Training Protocol, Convergence Behavior & Experimental Evaluation

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.

Unsupervised learning algorithms find structures in data that has not been labelled, classified or categorised. Instead of responding to feedback, unsupervised learning algorithms identify commonalities in the data and react based on the presence or absence of such commonalities in each new piece of data. Central applications of unsupervised machine learning include clustering, dimensionality reduction, and density estimation. Cluster analysis is the assignment of a set of observations into subsets (called clusters) so that observations within the same cluster are similar according to one or more predesignated criteria, while observations drawn from different clusters are dissimilar. Different clustering techniques make different assumptions on the structure of the data, often defined by some similarity metric and evaluated, for example, by internal compactness, or the similarity between members of the same cluster, and separation, the difference between clusters. Other methods are based on estimated density and graph connectivity. A special type of unsupervised learning called self-supervised learning involves training a model by generating the supervisory signal from the data itself.

Semi-supervised learning falls between unsupervised learning (without any labelled training data) and supervised learning (with completely labelled training data). Some of the training examples are missing training labels, yet many machine-learning researchers have found that unlabelled data, when used in conjunction with a small amount of labelled data, can produce a considerable improvement in learning accuracy. In weakly supervised learning, the training labels are noisy, limited, or imprecise; however, these labels are often cheaper to obtain, resulting in larger effective training sets.