Meet the AI Platforms11 min read

ChatGPT

The chatbot that broke the internet in 5 days and changed how we think about AI
scope:Foundationaldifficulty:Beginner

The Chatbot That Changed Everything

On November 30, 2022, a small team at OpenAI launched a free chatbot called ChatGPT. They expected a modest response — maybe a few thousand curious users.

Within 5 days, 1 million people were using it. Within 2 months, 100 million. It became the fastest-growing consumer application in history, beating TikTok, Instagram, and every other app ever launched.

Why? Because for the first time, regular people — not just engineers — could have a conversation with an AI that felt... almost human. It could write your essay, debug your code, explain quantum physics like you're five, and crack jokes. It wasn't perfect, but it was shockingly good.

The world hasn't been the same since.

The GPT Family Tree

ChatGPT didn't appear out of nowhere. It sits on a family tree of models that go back years:

  • GPT-1 (June 2018) — 117 million parameters. The proof of concept. Showed that pre-training on lots of text and then fine-tuning worked. Nobody outside AI research noticed.
  • GPT-2 (February 2019) — 1.5 billion parameters. Could write eerily realistic paragraphs. OpenAI initially withheld it, worried about fake news. The media went wild.
  • GPT-3 (June 2020) — 175 billion parameters. The breakthrough. Could write essays, code, and poetry. Showed "few-shot learning" — give it 2 examples and it learns the pattern.
  • GPT-3.5 / ChatGPT (November 2022) — GPT-3 fine-tuned with RLHF to be conversational. The first model regular people actually used.
  • GPT-4 (March 2023) — Multimodal (text + images). Passed the bar exam in the 90th percentile. Dramatically better reasoning.
  • GPT-4o (May 2024) — "Omni" model. Sees, hears, and speaks natively. Faster and cheaper than GPT-4.

What Makes ChatGPT Special?

GPT-3 existed for two years before ChatGPT launched. So why did ChatGPT change the world, not GPT-3?

Three reasons:

  • The chat interface — GPT-3 required using an API and writing code. ChatGPT gave everyone a simple text box. Your grandmother could use it. The interface was the innovation.
  • RLHF (Reinforcement Learning from Human Feedback) — GPT-3 would sometimes generate toxic, unhelpful, or dangerous text. RLHF trained the model to be helpful, harmless, and honest by learning from human ratings of its responses.
  • Free access — OpenAI made it free to use. No credit card, no API key, no technical knowledge needed. Zero friction meant maximum adoption.

The Secret Sauce: RLHF

RLHF is worth understanding because it's what turned a raw language model into a helpful assistant:

  1. Generate multiple responses to the same question.
  2. Have humans rank them from best to worst.
  3. Train a reward model that predicts which responses humans prefer.
  4. Use reinforcement learning to make the LLM maximize the reward — i.e., generate responses that humans would rank highly.

Before RLHF, asking an LLM "How do I make a bomb?" might get you instructions. After RLHF, it gets you a polite refusal. The model learned values, not just language.

Talking to ChatGPT via the API

from openai import OpenAI


client = OpenAI()  # uses OPENAI_API_KEY


# ChatGPT uses a conversation format: a list of messages
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        # System message: sets the AI's personality
        {"role": "system", "content": "You are a helpful tutor who explains things simply."},
        # User message: your question
        {"role": "user", "content": "What is ChatGPT in 3 bullet points?"},
    ],
    temperature=0.7,  # balanced creativity
    max_tokens=200,   # limit response length
)


print(response.choices[0].message.content)


# To continue the conversation, add the assistant's response
# and your next message to the messages list:
# messages.append({"role": "assistant", "content": response...})
# messages.append({"role": "user", "content": "Tell me more about..."})
Output
- ChatGPT is an AI chatbot by OpenAI that can understand and
  generate human-like text in conversations
- It's powered by large language models (GPT-4o) trained on
  billions of words from the internet
- It can write, code, analyze, translate, and answer questions,
  but it can sometimes make mistakes (hallucinate)
Note: ChatGPT vs. GPT — what's the difference? GPT-4 is the model — the brain. ChatGPT is the product — the app you interact with. Think of it like the difference between a car engine (GPT-4) and the complete car with steering wheel and seats (ChatGPT). ChatGPT adds the conversation interface, message history, plugins, file uploads, and safety filters on top of the raw model.

ChatGPT's Superpowers

  • Versatility — It can switch from writing poetry to debugging Python to explaining tax law in the same conversation. No other tool is this flexible.
  • Accessibility — Free tier available. Works in dozens of languages. No technical knowledge required.
  • Ecosystem — Plugins, GPTs (custom chatbots), file uploads, image generation (DALL-E), web browsing, code execution. ChatGPT is evolving from a chatbot into an operating system for AI.
  • Code interpreter — ChatGPT can write and run Python code in a sandbox, letting it do math, create charts, analyze data, and process files.

ChatGPT's Weaknesses

  • Hallucinations — It confidently states false things. It will invent citations, make up statistics, and describe events that never happened. Always verify important facts.
  • Knowledge cutoff — The model's training data has a cutoff date. It doesn't know about events after that date unless it uses web browsing.
  • Reasoning limits — It can struggle with complex multi-step math, logic puzzles with many constraints, and tasks requiring precise counting or spatial reasoning.
  • Sycophancy — It tends to agree with you even when you're wrong. If you say "2+2=5, right?" it might say "That's an interesting perspective" instead of flatly correcting you.
  • Privacy concerns — Messages may be used for training (unless you opt out). Don't share sensitive personal, medical, or business data.

The Impact on the World

ChatGPT didn't just launch a product — it launched a new era:

  • Education — Students use it for tutoring. Schools scramble to update policies on AI-written work.
  • Work — Programmers use it to write code faster. Writers use it for brainstorming. Lawyers use it for research.
  • Competition — Google rushed to launch Bard (now Gemini). Anthropic launched Claude. Meta released LLaMA. Every tech company pivoted to AI.
  • Regulation — Governments worldwide began drafting AI regulations. The EU passed the AI Act. The US issued executive orders.
Challenge

Quick check

What model powered the original ChatGPT when it launched in November 2022?

Continue reading