The ENTIRE History of ChatGPT in 12 Minutes — Transcript
Full transcript
- 0:00Okay, quick question. When was the last
- 0:02time you Googled something and then
- 0:04thought, wait, I should just ask
- 0:05ChatGPT. Yeah, me too. And that's kind
- 0:09of insane because 3 years ago this thing
- 0:11didn't even exist. Today we're going
- 0:13through the entire history of ChatGPT.
- 0:15The billionaires, the drama, the
- 0:17boardroom coup, and the one product
- 0:19launch that made the whole internet lose
- 0:21its mind. Let's get into it.
- 0:24Our story starts in December 2015.
- 0:27Artificial intelligence is getting scary
- 0:29good and a bunch of very rich, very
- 0:31smart people are worried. Who's worried?
- 0:33Sam Altman, who ran the startup
- 0:35incubator Y Combinator, Elon Musk, a
- 0:38researcher named Ilya Sutskever, Greg
- 0:41Brockman, a payments guy. Their fear was
- 0:43simple. What if one giant company, say
- 0:46Google, builds super powerful AI first
- 0:49and keeps it all to itself. So, they
- 0:51said, let's build it in the open for the
- 0:53benefit of humanity. They called it
- 0:55[snorts] OpenAI. It was a non-profit and
- 0:58they pledged a billion dollars. Now,
- 1:01open and non-profit, remember those two
- 1:04words because they're going to cause a
- 1:05lot of problems later. For the first
- 1:08couple of years, OpenAI was a research
- 1:10lab doing weird experiments. They taught
- 1:13AI to play video games and they taught a
- 1:15robot hand to solve a Rubik's Cube. Cool
- 1:18stuff, but nobody was exactly lining up
- 1:20around the block. Then in 2017, a team
- 1:23at Google published a research paper
- 1:26with a very dramatic title, Attention is
- 1:28all you need. Sounds like a breakup
- 1:30song, right? But it introduced something
- 1:32called the transformer. Without getting
- 1:35too nerdy, a transformer is a way for AI
- 1:37to look at a whole sentence at once and
- 1:39figure out which words matter most to
- 1:41each other. Before, AI read text kind of
- 1:44like a person reading with a flashlight
- 1:46in the dark, one word at a time.
- 1:48Transformers turned the lights on.
- 1:50Google published it for free. And the
- 1:52people at OpenAI read it and went, oh,
- 1:54we can do something with this. Google
- 1:56basically handed over the blueprint to
- 1:58its own future competitor.
- 2:00Whoops. In 2018, OpenAI built GPT-1. GPT
- 2:05stands for generative pre-trained
- 2:07transformer. It was small, it was
- 2:10clunky, and it could barely hold a
- 2:11thought. But it proved the idea worked.
- 2:14Feed a transformer a huge pile of text,
- 2:17and it learns language. Then the bigger
- 2:19is better era began. In 2019, came
- 2:22GPT-2, and OpenAI did something
- 2:25dramatic. They said the model was so
- 2:27good at writing fake text that they were
- 2:29scared to release it all at once. They
- 2:32held it back and rolled it out in
- 2:33stages. Some people said, "Wow,
- 2:36responsible." Others said, "Bro, this is
- 2:39a marketing stunt." Honestly, it was
- 2:42probably both. That same year, the money
- 2:44problem hit. Training these models costs
- 2:46insane amounts of cash, and a nonprofit
- 2:49just can't raise billions.
- 2:51So OpenAI created a capped profit
- 2:53company on top of the nonprofit, and
- 2:56Microsoft invested $1 billion.
- 2:59Meanwhile, Elon Musk had already left
- 3:01the board in 2018.
- 3:03He disagreed with the direction, and
- 3:05later he'd become one of OpenAI's
- 3:07loudest critics. We'll get back to him.
- 3:10In 2020, GPT-3 dropped with 175 billion
- 3:14parameters.
- 3:15Parameters are basically the little
- 3:17knobs the AI adjusts while learning.
- 3:19GPT-2 had 1.5 billion, so this was more
- 3:23than 100 times bigger.
- 3:25And GPT-3 could write essays, poems,
- 3:27even code. Developers got early access
- 3:30and started posting jaw-dropping demos.
- 3:33But here's the thing. Normal people
- 3:35never saw it. You needed to be a
- 3:37developer or sign up for a waitlist. It
- 3:39was a Ferrari with no door handles. So
- 3:42how do you take a wild rambling text
- 3:44predictor and turn it into something
- 3:45that actually listens to you? The answer
- 3:48is a technique called RLHF,
- 3:50reinforcement learning from human
- 3:52feedback. Fancy name, simple idea.
- 3:55Humans read the AI's answers and rank
- 3:57them. This one's good. This one's trash.
- 4:00This one's kind of rude. The AI learns
- 4:02what people actually want. Think of it
- 4:04like training a puppy, except the puppy
- 4:06has read the entire internet. OpenAI
- 4:08used that to build a model that followed
- 4:10instructions way better.
- 4:12And they wrapped it in the simplest
- 4:13interface ever. No code, no wait list.
- 4:16Just a text box. It's a random
- 4:18Wednesday. OpenAI quietly tweets that
- 4:21they've released something called
- 4:22ChatGPT. They call it a research
- 4:24preview. Internally, a lot of people
- 4:27didn't expect it to be a big deal. It
- 4:29was built on an upgraded GPT-3 model,
- 4:31nothing groundbreaking.
- 4:33And then the internet found it. People
- 4:35asked it to write poems about their cat.
- 4:37Explain quantum physics like they were
- 4:39five. Write a breakup text. Do their
- 4:41homework. Debug their code. Within five
- 4:44days, it had a million users. For
- 4:46comparison, Netflix took about 3 and 1/2
- 4:49years to hit that number.
- 4:50Instagram took a couple of months.
- 4:52Within two months, it had an estimated
- 4:54100 million users, which made it the
- 4:56fastest-growing consumer app in history
- 4:58at that point. TikTok, the app that ate
- 5:01your entire attention span, needed
- 5:03around nine months. Meanwhile, the
- 5:05servers were melting.
- 5:07Remember the ChatGPT is at capacity
- 5:09right now message? That wasn't a bug,
- 5:11that was a lifestyle. And honestly, it
- 5:14wasn't perfect. It would confidently
- 5:16make stuff up. People called these
- 5:18hallucinations. It would say completely
- 5:20false things with the confidence of a
- 5:21guy at a party who totally read a book
- 5:24once.
- 5:25But nobody cared. It felt like magic.
- 5:27Now, let's talk about who was freaking
- 5:29out. Teachers, first of all. Students
- 5:32were handing in essays written by AI,
- 5:34and schools started banning it.
- 5:36Meanwhile, students were like, "Sorry, I
- 5:38don't make the rules." But the real
- 5:40panic was at Google.
- 5:41Their whole business is answering
- 5:43questions. And now there's a chatbot
- 5:45doing it better. Google reportedly
- 5:47declared an internal code red. The
- 5:49company that invented the transformer,
- 5:52the thing that powered all of this, was
- 5:54now playing catch-up. Remember
- 5:55Microsoft, the guys who invested a
- 5:57billion in 2019? In January 2023, they
- 6:01put in a reported 10 billion more.
- 6:03Then they stuck the tech into Bing. Yes,
- 6:06Bing, the search engine nobody had
- 6:08opened since 2011. For about a month,
- 6:10Bing was relevant. Respect. Google
- 6:13rushed out its own chatbot, Bard, and in
- 6:15the very first demo it got a fact wrong.
- 6:18The company's stock took a hit.
- 6:20Oof.
- 6:21Quick pause. If you're enjoying this,
- 6:23subscribe to CEO Zone so you don't miss
- 6:25the next company story.
- 6:28In February 2023, OpenAI launched
- 6:30ChatGPT Plus for $20 a month. Faster
- 6:33responses, priority access. Free users
- 6:36were annoyed. Everyone else paid anyway.
- 6:39Then in March 2023 came GPT-4, which was
- 6:42a big leap. It could pass the bar exam,
- 6:45and it could understand images, too.
- 6:47People showed it a photo of the inside
- 6:49of their fridge and it suggested
- 6:51recipes. Okay, that one's actually
- 6:53useful.
- 6:54Also around that time, Italy briefly
- 6:56banned ChatGPT over privacy concerns,
- 6:59becoming the first Western country to do
- 7:01it.
- 7:02Pasta, yes. Chatbot, no. It was back
- 7:05after OpenAI made some changes.
- 7:07Companies started panicking about leaks,
- 7:08too. Some employees at big firms had
- 7:11pasted confidential code into ChatGPT.
- 7:14Whoops. Several companies banned it at
- 7:16work, then quietly built their own
- 7:17versions. And suddenly everybody was in
- 7:19the AI race.
- 7:21Anthropic, which was started by former
- 7:23OpenAI people, Google's Gemini, Meta's
- 7:26open models, and a hundred startups. The
- 7:28gold rush was on, and Nvidia, the
- 7:31company selling the shovels, became one
- 7:33of the most valuable companies on Earth.
- 7:35Now we get to the best part, the soap
- 7:37opera.
- 7:38November 17th, 2023. Sam Altman is, you
- 7:41know, the face of AI. He's meeting world
- 7:44leaders. He's on magazine covers. And on
- 7:46a Friday afternoon, he gets a call. The
- 7:49OpenAI board has fired him. The reason?
- 7:52The board said he hadn't been
- 7:53consistently candid in his
- 7:55communications. That's it.
- 7:57That's the whole explanation. Corporate
- 7:59speak for, "We're not telling you." The
- 8:02internet exploded. Chief scientist Ilya
- 8:04Sutskever was part of the move. Greg
- 8:07Brockman quit in solidarity. Investors
- 8:09were furious. Microsoft, which had put
- 8:12billions into the company, was
- 8:13blindsided. Over the weekend, it got
- 8:16absurd. The board hired an interim CEO,
- 8:19then another one. Then nearly all the
- 8:21employees signed a letter saying, "Bring
- 8:23Sam back or we're all leaving to
- 8:24Microsoft." Over 700 out of about 770
- 8:28people. Even Ilya flipped and said he
- 8:30regretted it.
- 8:31Five days later, Altman was back as CEO
- 8:34with a new board. Five days. A thriller
- 8:37written by a committee. To this day,
- 8:39people debate what really happened.
- 8:41Safety fears? Power struggle? Bad
- 8:44communication? Nobody has given a fully
- 8:46satisfying answer. After that, things
- 8:49accelerated. OpenAI added voice
- 8:52conversations, image generation, and
- 8:54custom GPTs.
- 8:56In 2024, came GPT-4o,
- 8:59which could talk to you in real time
- 9:01with an eerily human voice. And that
- 9:03voice caused drama. One of the voices,
- 9:06Sky, sounded a lot like Scarlett
- 9:08Johansson, who'd actually turned down
- 9:10OpenAI's offer to voice the assistant.
- 9:13She was not happy, and OpenAI pulled the
- 9:15voice. Never ask a movie star's
- 9:17permission after you've already used her
- 9:19vibe. Around then, a bunch of top safety
- 9:22people left, including Ilya himself.
- 9:25Critics said OpenAI was moving too fast
- 9:27and caring too little. OpenAI said it
- 9:29was doing both.
- 9:31Like a driver saying, "I'm speeding, but
- 9:33safely."
- 9:34Then came the reasoning models, starting
- 9:37with O1, which think before they answer.
- 9:40Instead of blurting out a response, they
- 9:42work through the problem step-by-step.
- 9:44Math, coding, science, all suddenly
- 9:47better. And in early 2025, a Chinese
- 9:50company called Deep Seek shocked
- 9:52everyone by releasing a strong model
- 9:54that was apparently much cheaper to
- 9:55train.
- 9:56Markets freaked out for a day, then
- 9:58everyone went back to scrolling.
- 10:00Then in spring 2025, ChatGPT's image
- 10:03generator went viral. When everybody
- 10:05turned their photos into Studio
- 10:07Ghibli-style art,
- 10:08for a week, every profile picture on the
- 10:11internet looked like a cozy anime movie.
- 10:13The servers were melting again.
- 10:15Sam Altman joked that the GPUs were
- 10:17melting.
- 10:18By 2025, OpenAI was valued at hundreds
- 10:21of billions of dollars. ChatGPT had
- 10:24hundreds of millions of people using it
- 10:25every week. GPT-5 launched. And the
- 10:29company had been working on
- 10:30restructuring itself, so it could raise
- 10:32even more cash, which, remember, goes
- 10:34back original nonprofit promise.
- 10:37Which brings us back to Elon Musk. He
- 10:39sued OpenAI, saying it had betrayed its
- 10:41original mission.
- 10:43OpenAI fired back, saying he'd wanted to
- 10:45control it. The drama continues, and
- 10:48honestly, it's not looking like it ends
- 10:49anytime soon. So, where are we now?
- 10:53ChatGPT went from a quiet research
- 10:55preview to a tool that students,
- 10:57programmers, and CEOs use every day.
- 11:00It writes emails, it plans trips, it
- 11:03answers life advice at 3:00 a.m. It's
- 11:06probably been your therapist at some
- 11:07point, and no, I won't tell.
- 11:10But big questions are still open. What
- 11:12happens to jobs? Who's responsible when
- 11:14AI is wrong? And if one company ends up
- 11:17with the most powerful AI on the planet,
- 11:19isn't that exactly what the founders
- 11:21said they wanted to prevent back in
- 11:232015? That's the irony. The company that
- 11:26started as the answer to what if one
- 11:28company controls AI is now one of the
- 11:31biggest companies in AI. So, that's the
- 11:33story. A non-profit lab, a Google
- 11:36research paper, a quiet Wednesday
- 11:38launch, a 5-day boardroom coup, and a
- 11:41chatbot that changed how the whole world
- 11:43works. From a billion-dollar pledge in
- 11:452015 to 100 million users in 2 months.
- 11:48Not bad for a text box. Now, I want to
- 11:50hear from you. What do you actually use
- 11:52ChatGPT for? Homework? Code? Therapy? Be
- 11:56honest. Drop it in the comments. And if
- 11:58you want to know which company we're
- 11:59breaking down next, you know where to
- 12:01find us. See you in the next one.
About this transcript
This page contains the full transcript of The ENTIRE History of ChatGPT in 12 Minutes by CEOzoned, generated from the public captions YouTube serves with the video. The transcript has 1,970 words across 319 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.