ChatGPT, 2022: When AI Became Everyone's Business

“OpenAI publishes a blog post with the understated title ‘ChatGPT: Optimizing Language Models for Dialogue.’… No press release. No influencer campaign. No coordinated media effort. Just a blog post and a link. Within five days, ChatGPT has one million users. Within two months, it has one hundred million users — the fastest adoption of a consumer product in history.”

San Francisco, California. November 30, 2022. A Wednesday. 11:43 AM Pacific time.

OpenAI publishes a blog post with the understated title “ChatGPT: Optimizing Language Models for Dialogue.” The post describes a new conversational AI system, accessible through a simple web interface. “We’ve trained a model called ChatGPT which interacts in a conversational way,” the post explains. “The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.”

No press release. No influencer campaign. No coordinated media effort. Just a blog post and a link.

Within five days, ChatGPT has one million users.

Within two months, it has one hundred million users — the fastest adoption of a consumer product in history.

Within a year, it has changed the AI industry, triggered an educational crisis, sparked a creative renaissance and a creative panic simultaneously, generated more venture capital than any previous technology, destabilised several professions, and become the most consequential product launch of the decade.

The world has been waiting for this moment for decades without knowing it. And now that it is here, nobody is quite sure what to do with it.

---

Before the Launch: What Made ChatGPT Different

ChatGPT was not the first conversational AI system. It was not the first large language model. It was not the first AI chatbot. What made it different was a specific combination of capabilities and accessibility that had not previously existed.

The capabilities were rooted in GPT-4’s predecessor model — specifically in InstructGPT, the model that OpenAI had trained using reinforcement learning from human feedback (RLHF) to follow instructions and produce helpful, harmless, honest outputs. InstructGPT had been a significant technical advance over GPT-3’s base model, and the improvements were specifically in the dimensions that made AI systems useful for general purposes: instruction following, factual accuracy relative to the base model, and the ability to maintain a coherent, helpful persona across an extended conversation.

The accessibility was the chat interface itself. Previous AI systems with similar underlying capabilities had been accessible primarily through APIs — requiring technical knowledge to use, limiting their availability to developers and researchers. ChatGPT wrapped the underlying model in a simple, free, web-based interface that anyone with an internet connection could use immediately, without any technical knowledge and without creating an account. You typed a message, the model responded, you typed another message. The interaction model was familiar — it was just like a text message exchange — and the simplicity of the interface allowed the extraordinary capabilities of the underlying model to be experienced directly, without any technical barriers.

The combination of RLHF-trained capabilities and a simple chat interface created something that felt qualitatively different from any previous AI system. Not because the underlying technology was more capable than GPT-3 in an absolute sense — the improvements were significant but incremental. Because the specific capabilities being unlocked — instruction following, helpful conversation, coherent explanation — were the capabilities that made the system immediately useful to ordinary people doing ordinary tasks.

---

The First Wave: Writers, Students, and Office Workers

The people who adopted ChatGPT in the first weeks fell into several overlapping categories, and their specific experiences of the technology are worth examining because they reveal what the system actually did and why it mattered.

Writers and content creators. For writers who produced large volumes of text — journalists, bloggers, content marketers, social media managers — ChatGPT offered something that felt, in the first days, like a miracle: the ability to generate competent prose on any topic, quickly, without any particular effort. You described what you wanted, and the system produced a draft. The draft was not always excellent, but it was often adequate for the purpose, and it could be edited and refined much faster than writing from scratch.

The first reaction of many professional writers was either excitement (this will make me more productive) or alarm (this will replace me) — and often both simultaneously. The productivity boost was real. The threat was also real, or at least plausible. Content that could be produced in hours by a writer could now be produced in minutes by an AI. The question of what that meant for the economics of writing — for the people who earned their livings producing content — was raised immediately and has not been resolved.

Students. For students at every level from high school to graduate school, ChatGPT offered something even more immediately useful and more immediately problematic: the ability to produce competent essays, problem set answers, and code on demand. A student who was supposed to write a five-paragraph essay about the causes of the First World War could ask ChatGPT to write it for them, and the result would be — depending on the teacher’s standards — passable, competent, or genuinely impressive.

The educational crisis that ChatGPT triggered was immediate and loud. Teachers discovered that the essays submitted were suddenly better than the students’ previous work — and not just marginally better, but suspiciously polished, suspiciously confident, suspiciously free of the specific errors that characterised each student’s writing. Plagiarism detection software was rendered useless overnight, because ChatGPT didn’t reproduce existing text — it generated new text that was the product of training on existing text, detectable only by tools specifically designed for the purpose.

Schools and universities responded with a range of policies: banning AI use entirely, requiring in-class writing, redesigning assignments to make AI less useful, requiring documentation of AI use. None of these responses was entirely satisfactory, and the conversation about how to teach writing, thinking, and learning in a world where AI can produce first drafts on demand is still ongoing.

Office workers. For people who worked in offices and spent significant portions of their time producing text — writing emails, drafting documents, summarising meetings, preparing presentations — ChatGPT offered a productivity improvement that was visible and measurable. The specific tasks where AI assistance was most valuable — writing a polite but firm email, summarising a long document, generating options for how to phrase a difficult message — were tasks that took up real time and that ChatGPT could assist with immediately and effectively.

The office worker adoption was less fraught than the student adoption because the productivity improvement was the point. Using AI to be more productive at work was not cheating — it was work. The early adopters in office settings quickly became evangelists, because the productivity gains were genuine and because the barrier to demonstrating those gains was low.

---

The Viral Moment: Why ChatGPT Spread So Fast

The spread of ChatGPT from its November 30 launch to one million users in five days and one hundred million users in two months was driven by specific dynamics of viral adoption that are worth examining.

The immediacy of the experience. The first time a person used ChatGPT, they had an immediate, visceral experience of its capabilities. Not a description of what the system could do, not a benchmark score, not a technical explanation — an actual demonstration, tailored to their specific interests and needs. The first-time user experience of ChatGPT was powerful enough that a significant fraction of users became immediate advocates, sharing the experience with friends and colleagues.

The shareability of examples. ChatGPT outputs were highly shareable. When ChatGPT wrote a sonnet in the style of Shakespeare about the challenges of doing laundry, or explained quantum computing using only words a five-year-old would understand, or wrote a cover letter that was exactly what the user needed, the output was a natural thing to share — on Twitter, on LinkedIn, in group chats, at the dinner table. The shareability of outputs drove discovery at scale.

The news cycle. Technology journalists, education journalists, business journalists, and cultural commentators all found in ChatGPT a story that was compelling to their specific audiences. The educational crisis angle attracted education journalists. The automation-of-work angle attracted business journalists. The implications for creative fields attracted culture journalists. The AI safety angle attracted technology journalists. The simultaneous relevance of the story to multiple different beats and audiences produced an unusually extensive and sustained media cycle that introduced ChatGPT to people who had not discovered it through direct experience or social sharing.

The free access. ChatGPT was initially free to use without creating an account. This eliminated the friction that normally limits consumer technology adoption — no payment, no commitment, no personal information required. Anyone could try it immediately. The low barrier to first experience meant that adoption was limited primarily by awareness, not by willingness to engage.

---

The Reactions: A World Processing a New Reality

The reactions to ChatGPT across different professions, industries, and communities reveal the breadth of its impact and the diversity of the responses it triggered.

In journalism and media. The reaction ranged from genuine excitement about AI-assisted reporting tools to existential dread about the future of the profession. Some news organisations immediately began experimenting with AI assistance for specific writing tasks — summarisation, boilerplate content, data-driven stories. Others published pieces warning about the risks of AI to journalism, the threats to employment, the potential for AI-generated misinformation.

The specific threat to journalism was real. The content that AI could produce most easily — brief, factual summaries; formulaic news stories about earnings or sports scores; simple explainers on common topics — was exactly the content that lower-paid journalists and freelancers produced at scale. The economic model for high-volume, low-prestige content production was immediately challenged by technology that could produce similar content at near-zero marginal cost.

In law. Lawyers discovered that ChatGPT could assist with legal research, contract drafting, and memo writing at a level that was useful for specific tasks. The notorious case in which a lawyer submitted legal briefs citing cases that ChatGPT had fabricated — confidently producing non-existent legal precedents, a textbook example of AI hallucination — became a cautionary tale about the risks of uncritical reliance on AI for legal work. But it did not stop the adoption: law firms across the country began developing AI policies and AI assistance programmes within months of the ChatGPT launch.

In education. The educational response was more fraught and more prolonged than in most professional fields. The debate about AI use in education touched on fundamental questions about the purpose of education — whether education was primarily about producing outputs (essays, problem sets, code) or about developing the thinking processes that produced those outputs. If AI could produce the outputs, what was the educational value of requiring students to produce them?

These questions did not have easy answers, and the educational community’s response has been characterised by productive experimentation alongside genuine uncertainty. The most thoughtful educational responses have tried to redesign assessments and assignments in ways that leverage AI as a learning tool rather than treating it as a threat — using AI to provide personalised feedback, to generate practice problems, to help students learn from their mistakes rather than hide them.

In software development. Programmers found in ChatGPT a coding assistant of unusual capability. GitHub Copilot, the AI coding assistant built on OpenAI’s Codex model, had already demonstrated that AI could assist with code writing, but ChatGPT’s ability to explain code, debug errors, translate between programming languages, and generate code from natural language descriptions was more flexible and more immediately accessible. Developer productivity improvements from AI assistance were widely reported, and the debate about whether AI would replace programmers or augment them was renewed with specific evidence on both sides.

In creative fields. Artists, musicians, and creative writers responded to ChatGPT (and to the simultaneously developing image generation AI, which was its own distinct phenomenon) with a mixture of creative excitement and economic anxiety. Some writers found AI assistance genuinely useful — for brainstorming, for overcoming writer’s block, for generating first drafts that could be refined. Others found the prospect of AI-generated content threatening to the economic model that made creative careers viable and to the cultural value of human creative expression.

The creative anxiety was not irrational. If AI could produce competent prose, the economic value of competent prose was reduced. If AI could produce publishable poetry, the market for poetry was changed. The specific creative fields most threatened were those where the value had been primarily in the execution of predictable forms rather than in genuine originality — genre fiction, commercial copywriting, advertising, certain kinds of journalism. Fields where genuine originality and individual voice were the primary value proposition were less immediately threatened but not immune.

---

The Technical Reality: What ChatGPT Actually Was

The public experience of ChatGPT was sometimes at odds with the technical reality of what the system was doing, and understanding the gap is important for understanding both the system’s capabilities and its limitations.

ChatGPT was not accessing the internet. It was not looking things up. It was generating text based on patterns learned during training — patterns in the enormous corpus of internet text on which it had been trained, combined with the specific fine-tuning that the RLHF process had produced. When it appeared to “know” something, it was more accurate to say that it had learned, from training data, what kinds of text were associated with what topics, and was producing text that fit those patterns.

This had specific implications for reliability. ChatGPT could produce text that sounded authoritative and accurate even when it was not — the hallucination problem that had been documented for GPT-3 was still present in ChatGPT. The difference was that ChatGPT’s more consistently helpful tone and instruction-following capability meant that users were more likely to trust its outputs without verification. The combination of improved capability and increased user trust made hallucination more dangerous, not less, in some contexts.

ChatGPT also had a training data cutoff. Its knowledge of the world stopped at a specific date — the date when its training data had been compiled — and it was not reliably aware of events that had occurred after that date. Users who asked about recent events could receive confident but potentially outdated responses, and the system was not always reliable at flagging the temporal limits of its knowledge.

The context window — the amount of text the model could process in a single conversation — was limited. In early versions, ChatGPT could not maintain coherent context across very long conversations, leading to the appearance of “forgetting” earlier parts of the conversation. The context limitation was a practical constraint on the system’s usefulness for extended, complex tasks.

These limitations were real and important. They did not prevent the system from being extraordinarily useful for a wide range of tasks — but they meant that understanding what the system was and was not doing was essential for using it well and avoiding the specific failure modes that could cause harm.

---

The Educational Reckoning: Teaching and Learning After ChatGPT

The educational response to ChatGPT was the most prolonged and the most consequential of the immediate reactions, and it deserves extended treatment.

The initial panic in schools and universities was, in retrospect, understandable but somewhat misdirected. The initial concern was primarily about cheating — about students using ChatGPT to produce work that was supposed to be their own. This concern was real, but it focused on a symptom rather than the underlying question: what was the purpose of the assignments that ChatGPT could do?

If the purpose of a five-paragraph essay was to develop the student’s ability to structure arguments, use evidence, and write clearly, then the fact that ChatGPT could produce a five-paragraph essay did not make the assignment pointless — it made it essential to redesign the assessment so that it could not be satisfied by a ChatGPT submission. If the purpose of a programming assignment was to develop the student’s ability to think algorithmically and solve problems in code, then the availability of AI coding assistance required redesigning the assessment to focus on the dimensions of that capability that AI could not replicate.

The deeper question was what the AI era changed about the fundamental goals of education. If AI could produce competent first drafts of most kinds of text, what was the educational value of developing the ability to produce first drafts? If AI could write working code for most common programming tasks, what was the educational value of developing the ability to write code from scratch?

The educational community’s evolving answer has been roughly: the value is in developing the higher-order capabilities that underlie the production of first drafts and the writing of code — the ability to evaluate outputs critically, to identify what is good and what is not, to refine drafts into excellent work, to specify problems clearly enough that AI can assist effectively, to understand the domain well enough to catch AI errors. These higher-order capabilities are more valuable in an AI-assisted world, not less, and developing them requires practice that is not undermined by AI assistance — it is, in some cases, enhanced by it.

The universities and schools that navigated the ChatGPT challenge most effectively were those that asked not “how do we prevent AI use?” but “what should our students be able to do, and how do AI tools change what we need to teach and assess?”

---

The Labour Market Question: What Happens to Jobs?

The most socially consequential question that ChatGPT raised — the question that connected the technical capability of the system to the lives of the largest number of people — was the labour market question: what would happen to the jobs that AI could now do?

The historical precedents for major technological transformations of the labour market — the industrial revolution, the mechanisation of agriculture, the computerisation of routine cognitive work — provided a range of possible outcomes. The most optimistic precedent was that new technology created more jobs than it displaced, because it increased productivity in ways that generated new economic activity and new kinds of work. The most pessimistic precedent was that the transition was painful and extended, with specific workers and communities bearing concentrated costs while the benefits were diffuse.

The specific impact of ChatGPT on specific types of work was visible relatively quickly in some categories. Content writing and copywriting — the production of formulaic, low-differentiated text — showed immediate competitive pressure from AI. Customer service — the answering of common questions and the handling of routine queries — was an obvious application for AI assistance. Translation — a task where AI had been improving for years but where ChatGPT’s conversational interface made AI assistance more accessible — showed accelerated adoption.

The impact on higher-skill work was less clear initially and remains contested. The lawyer who uses ChatGPT to draft a first version of a contract is more productive than the lawyer who doesn’t. Whether this means fewer lawyers are needed, or whether the same number of lawyers can do more work, or whether the productivity increase generates enough additional legal work to absorb the freed capacity — these are empirical questions that economic research was beginning to address but had not definitively answered.

The research that has emerged since the ChatGPT launch on the labour market impacts of AI has produced complex results. Some studies have found that AI assistance significantly increases the productivity of lower-skilled workers on specific tasks — reducing the productivity gap between high-skilled and low-skilled workers. Others have found that the workers most exposed to AI automation show wage and employment effects consistent with displacement. The overall picture is one of significant sectoral variation, with some job categories clearly more exposed than others, and with the ultimate labour market effects depending on the pace of AI improvement and on economic and policy responses that are still developing.

---

The Geopolitical Dimension: ChatGPT as a Moment in AI Competition

ChatGPT’s launch and its extraordinary adoption were not just a technology story. They were a geopolitical story — a demonstration of American leadership in a technology that was rapidly becoming strategically important, and a specific data point in the competition between the United States and China for AI dominance.

In China, the reaction to ChatGPT combined admiration for the technical achievement with urgency about competitive positioning. Chinese technology companies — Baidu, Alibaba, Tencent, and a large number of startups — accelerated their large language model development programmes in response to ChatGPT’s success. Baidu’s ERNIE Bot, launched in March 2023, was explicitly positioned as China’s answer to ChatGPT.

The Chinese government’s response was complicated by the specific sensitivity of conversational AI systems. A system that could discuss any topic freely was a potential challenge to information control in ways that the Chinese government was not prepared to accept. The regulatory framework for large language models in China, which developed rapidly after ChatGPT’s launch, required AI systems to align with “socialist core values” and to produce content that was consistent with government-approved information. This regulatory requirement created a specific challenge for Chinese AI companies: building competitive systems while complying with constraints that American companies did not face.

In Europe, the response was primarily regulatory. The EU AI Act, which was being negotiated at the time of ChatGPT’s launch, was accelerated and expanded to address the specific capabilities that ChatGPT demonstrated. The regulation of general-purpose AI systems — systems capable of a wide range of tasks, including tasks not anticipated at the time of their development — became a central concern of the regulatory framework.

In the United States, ChatGPT’s launch triggered congressional attention to AI regulation that had been building since the AlphaGo match in 2016 and the growing awareness of AI’s social impacts. Sam Altman’s Congressional testimony in May 2023 was widely covered and was notable for the consensus between Altman and Congress that regulation of frontier AI was necessary — though the specific form of regulation remained contested.

---

The Creative Revolution: What AI Unlocked for Millions

Alongside the anxieties about jobs, education, and geopolitics, ChatGPT also triggered something that deserves recognition: a genuine creative renaissance for millions of people who had previously lacked the tools to express themselves effectively.

The specific capabilities that ChatGPT unlocked for non-professional creators were significant. People who had ideas for stories but lacked the writing skills to execute them found in ChatGPT a collaborator who could help them develop and articulate their ideas. People learning a new language found in ChatGPT a patient interlocutor who could help them practice conversation. People who needed to write formal communications for the first time — cover letters, professional emails, grant applications — found in ChatGPT a guide who could help them navigate formats they had never encountered.

The democratisation of writing assistance that ChatGPT represented was genuinely significant. Access to good writing assistance had previously been correlated with education and socioeconomic status — people with more education knew how to write in formal registers, and people with more money could hire editors and writing consultants. ChatGPT made a version of this assistance available to anyone with an internet connection.

The creative applications were also genuine and varied. Roleplay and collaborative storytelling communities adopted AI enthusiastically, finding in language models willing creative collaborators for interactive fiction. Game designers used AI to generate flavour text, dialogue, and world-building details at a scale that would have been impossible manually. Hobbyist writers used AI to develop and expand their creative ideas, producing work they were proud of even when the final text was a collaboration between their vision and the model’s execution.

The tension between the democratisation of creative tools and the economic impact on professional creators is real and unresolved. The same capability that allows a non-writer to produce acceptable prose also reduces the market for professional content writing. The same capability that helps a language learner practice conversation also competes with human language tutors. The distribution of benefits and costs from AI assistance is not uniform, and the people who bear the costs are not always the people who receive the benefits.

---

The Safety Conversation Goes Mainstream

One of the specific effects of ChatGPT’s mainstream adoption was to bring the AI safety conversation — previously primarily a concern of technical researchers, AI ethicists, and certain policy communities — to a much wider audience.

Before ChatGPT, the average person’s experience of AI was primarily indirect: algorithms that recommended content, algorithms that assessed creditworthiness, AI-powered virtual assistants with limited capabilities. The safety and alignment concerns raised about these systems were real but abstract — it was hard for most people to directly experience why they might be worried about the long-term trajectory of AI development.

After ChatGPT, the safety conversation was more concrete. You could ask ChatGPT to help you do something harmful and observe how it responded. You could observe it produce confident misinformation. You could probe the limits of its safety guardrails. You could experience directly the capability of a system that could be used for beneficial purposes and harmful ones simultaneously. The capability that made safety concerns urgent was now something people could touch and interact with.

The specific safety concerns that became most salient were not the long-horizon catastrophic risks that the alignment research community had focused on. They were the near-term, concrete harms: misinformation, fraud, manipulation, harassment, the automation of harmful content at scale. These concerns were more immediate, more visible, and more susceptible to policy intervention than the longer-horizon risks, and they drove most of the regulatory and policy response to ChatGPT’s launch.

The March 2023 open letter calling for a six-month pause in AI development — signed by prominent AI researchers, technology executives, and public intellectuals — was the most visible expression of the safety concerns that ChatGPT’s capabilities had made more urgent. The letter was contested, signed by some of the most prominent figures in AI research and dismissed by others as impractical, counterproductive, or misguided. But its publication was itself significant: it demonstrated that the safety concerns were serious enough to attract broad public attention, and it contributed to the regulatory and governance discussions that followed.

---

The Products That Followed: The Cambrian Explosion of AI Applications

ChatGPT’s success triggered what technology journalists described as a “Cambrian explosion” of AI applications — an extraordinary proliferation of AI-powered products built on top of large language model APIs, applying the underlying capability to hundreds of specific use cases.

The applications ranged across virtually every domain of human activity.

In healthcare, AI assistants for clinical documentation reduced the administrative burden on physicians. AI tools for medical education helped students prepare for examinations. AI systems for patient triage helped healthcare organisations manage demand. The specific applications were careful to stay within appropriate bounds — AI as a tool to assist healthcare professionals, not to replace clinical judgment — but the productivity improvements were real and significant.

In legal services, AI tools for legal research, contract analysis, and document review reduced the time required for routine legal work. The specific applications were careful about the hallucination problem — the legal community’s experience with ChatGPT producing fabricated case citations had established that AI-generated legal content required careful verification. But the productivity improvements from AI assistance in appropriate contexts were substantial.

In software development, the AI coding assistant market exploded. GitHub Copilot expanded to cover more programming languages and more types of coding tasks. Competitors including Amazon CodeWhisperer, Tabnine, and dozens of others competed for the developer market. The productivity improvements from AI coding assistance were empirically documented — studies showed developers completing tasks faster with AI assistance than without — and the adoption was rapid across the industry.

In customer service, AI chatbots capable of handling complex, conversational queries replaced simpler decision-tree chatbots in many applications. The improvement in handling natural language, in understanding context, and in providing helpful responses rather than scripted ones made AI customer service more effective for a wider range of queries.

In education, AI tutoring applications capable of explaining concepts, providing personalised feedback, and adapting to individual student needs represented a genuine capability improvement over previous educational technology. The applications ranged from tools for helping students understand difficult concepts to tools for providing writing feedback to tools for personalised language learning.

---

The Culture Shift: AI as Background Radiation

Perhaps the most significant effect of ChatGPT’s mainstream adoption was not any specific use case or any specific industry disruption. It was the shift in the cultural background — the change in what felt normal about the role of AI in daily life.

Before ChatGPT, AI was something that happened to you — algorithms that shaped your social media feed, your search results, your product recommendations — but that you did not typically interact with directly and consciously. After ChatGPT, AI was something you could talk to, something you could ask for help, something you could collaborate with on tasks that mattered to you.

This shift from passive subject to active participant in AI interaction changed the relationship between ordinary people and AI systems in ways that are still playing out. People who had never thought seriously about what AI was or what it could do now had direct experience of a specific AI system’s capabilities and limitations. The abstract became concrete. The theoretical became personal.

The cultural shift also changed the political and policy conversation about AI. When AI was something that happened in the background of digital life, the people most concerned about its implications were specialists — researchers, policy advocates, technology journalists. When AI was something that hundreds of millions of people were actively using daily, the implications became something that the general public had views about and that politicians needed to respond to.

The result has been a broader, more intense, and more politically engaged conversation about AI — its benefits and risks, its economic implications, its governance needs — than had previously been possible. Whether this conversation will produce the policy responses that the technology’s potential requires is still to be determined. But the conversation is happening, at a scale and with a breadth that would not have been possible without ChatGPT’s mass adoption.

---

The Moment That Defined an Era

ChatGPT’s launch was, like AlexNet’s 2012 ILSVRC victory, a moment that defined an era — a moment that could be pointed to as the beginning of a recognisably new period in AI’s relationship with society.

Before ChatGPT, AI was a specialised technology with specific applications, impressive in narrow domains, inaccessible to most people, and understood primarily by specialists. After ChatGPT, AI was a general-purpose tool with applications in virtually every domain, accessible to anyone with an internet connection, and experienced directly by hundreds of millions of people.

The transition was not sudden — it had been building for years, through the deep learning revolution, through GPT-2 and GPT-3, through the quiet deployment of AI in products people used without recognising them as AI. ChatGPT was the moment the transition became visible, and therefore the moment it became consequential for society in ways that the underlying technical progress had not yet been.

What happens next — how the technology continues to develop, how society adapts to it, how governance frameworks evolve, how the promise and the risks are navigated — is the story that the final section of this series will tell. The foundation for that story was laid on November 30, 2022, with a blog post and a link, and within two months, one hundred million people were part of it.

---

Further Reading

---

---