Here is the excerpt rewritten with improved language:
Testing one, two, three. Blocks, obviously, like, here's what I can make up. Here you go. We've had a good run as the apex intelligence on the planet. We've been the smartest things around for a long time. I'm Dylan Curious from the YouTube channel Bill and Curious, and I'm really excited to talk with you, Wes. I'm glad to have you here. Let's get started. I'd love to start by asking when you had a defining moment where you thought AI might be worth all this effort?
For most people, I feel the ChatGPT moment was the start of something. For me, I saw Elon Musk tweet that ChatGPT was getting scary good, so I went to check it out. It was okay, definitely had something to it, but wasn't magical or mind-blowing. I was like, "Alright, this is interesting."
Next, I started learning more about OpenAI's work on reinforcement learning, where they were training AI agents to play hide-and-seek. It wasn't anything new or mind-blowing, but they did a great job presenting it in an engaging way, with little blue and red people running around. The premise was simple - the seeker team got points for keeping the hiders in sight, while the hiders got rewards for evading the seekers.
In the early simulations, the AI agents couldn't do anything - just randomly mashing buttons. Seeing that live was really interesting, and it tied into what's happening now. We didn't give them any human data to start - we just said, "Figure it out, and we'll reward or punish you." Over millions of iterations, you started to see intelligence emerge.
That was the second big piece for me - realizing this was genuine machine learning, not just programming. There was something resembling the human brain, gathering information and finding creative solutions the developers didn't anticipate. I was like, "Okay, this is learning. There's no way around it."
Then a few weeks after ChatGPT, GPT-4 came out, and that was the final nail in the coffin for me. Seeing the Microsoft paper on "sparks of AGI" - this was a prototype of general AI that would attempt any task, maybe succeed, maybe fail, just like a human. Combining the reinforcement learning and the more general LLM capabilities, I realized we were onto something big, and I went all in on it. Just eating some food now. We'll get back to what we're doing soon. I've got a feeling I should do my tax return, at least get money coming in, update my bank details, make a login defiance, and I hope someone's been checking these metadata why I couldn't log in after one attempt at the questions instead of two. We need to get proof of that so that we can let Sean know he's full of shit, my case manager.
