🎂 Happy second birthday, ChatGPT!

What will the AI system’s terrible twos herald for humans?

ChatGPT is turning two. I spoke with it about this milestone and what the next two years will hold. It’s pretty opinionated. This discussion was lightly edited for flow.

Still can’t spell…

Azeem Azhar: Tomorrow is the 30th November 2024. Do you know why that might be relevant to ChatGPT?

ChatGPT (Thought for 10 seconds): Yes, November 30th is significant because it marks the anniversary of ChatGPT’s release. ChatGPT was initially launched on November 30th, 2022. Tomorrow, November 30th, 2024, would be the second anniversary of ChatGPT’s public availability.

Azeem: Happy birthday to you! If you could feel anything, what might you feel?

ChatGPT (6s): Thank you for the birthday wishes! If I could experience feelings, I might feel gratitude and happiness for being able to assist and engage with you and others over the past two years.

Azeem: I’m going to ask you some questions about the last two years. This is past your knowledge cut-off date, so I’ll furnish relevant facts for you. I’d like you to respond in an intelligent but light-hearted way. Don’t be too verbose or po faced. Ok?

ChatGPT (10s): Sure thing! I'm all ears—ask away!

Azeem: Your launch seems to have created a ‘before and after’ moment in human history, similar to the printing press or internet. What parallels do you see between those transitions and the one we’re experiencing now?

ChatGPT (8s): You’re onto something. The launch of ChatGPT does feel like a pivotal moment—much like when the printing press revolutionized access to information or when the internet connected us all. A key parallel is the democratization of knowledge and tools. The printing press took books out of the hands of a few and made them available to many, sparking widespread literacy and learning. The internet broke down geographical barriers, enabling instant communication and access to information worldwide.

Now, with AI models like ChatGPT, we’re seeing a shift in how people interact with technology and information. Language models are making sophisticated capabilities—once limited to experts—accessible to anyone with an internet connection. This is changing how we learn, work, and solve problems. Just as previous innovations disrupted the status quo and opened up new possibilities, AI is enabling a new level of creativity and efficiency. It’s an exciting time, and we’re all part of navigating what this means for the future.

The terrible twos and beyond

I find the way it represents LLMs a bit weird

Azeem: Humans go through the “terrible twos” as they grow up. What would your behaviour look like if you went through the “terrible twos”?

ChatGPT (18s): If I were to experience the “terrible twos,” you might find me a bit more unpredictable. Perhaps I’d occasionally misunderstand straightforward questions or stubbornly stick to a topic even when you’ve moved on. Maybe I’d repeat the same suggestions or respond with a cheeky “Why?” more often than usual. It would be a phase where I’m testing boundaries, learning rapidly, and perhaps needing a bit more patience from those interacting with me. But don’t worry—it would also be a time of immense growth and discovery, leading to better interactions down the line.

Azeem: Let’s talk about what happens down the line. Foundation AI models are progressing on a smooth exponential, getting better and better at benchmarks and surpassing human capability in many areas, including coding and maths. We’re learning how to make frontier models smaller and more efficient after we’ve developed them, leading to a proliferation. Play these non-linear trends forward a couple of years and tell me what the most surprising things about the world when you hit your fourth birthday. I want you to think really, really hard about this. I don’t want pabulum. I want deeply considered, surprising, contrarian, fundamental insight. What will be the most surprising thing about the world when you turn four?

ChatGPT (17s): Looking ahead to my fourth birthday, I think one of the most surprising developments might be the emergence of AI as independent economic agents. As models become more advanced and efficient, they won’t just assist humans—they’ll start making autonomous decisions in the economy. Imagine AI systems owning digital assets, negotiating contracts, or even running businesses without direct human oversight.

This isn’t just automation of tasks; it’s a fundamental shift in how we structure economic activity. It challenges our concepts of ownership, legal responsibility, and agency. We’ll need to grapple with questions like: Can an AI enter a legally binding agreement? Who is accountable for its decisions? It’s a contrarian idea, but considering the exponential trends in AI capability and proliferation, it’s a plausible scenario.

Such a development could upend traditional industries and force us to rethink regulatory frameworks. It might even redefine what we consider to be “work” or “participation” in the economy. While it raises complex ethical and legal issues, it also opens up opportunities for innovation we haven’t yet imagined. It’s both an exciting and challenging prospect that could fundamentally alter our societal landscape.

The hard question

Azeem: Let’s stare into the maw. Will you rule us? Or will all economic activity just be hoovered up to San Francisco? Be honest. Be direct. Think really, really hard. Don’t cushion any blows.