#12 - Humanity's Enduring Edge Over AI

Shooting around at the YMCA while listening to a Sam Altman interview is one hell of a way to brainstorm new ideas about the future.

Today I’m thinking about how humans stay relevant in a world with super-intelligent AI agents.

Key Takeaway

I used to think Transformer-based AI models would stop improving at some point.

The math says otherwise.

There’s something called The Universal Approximation Theorem, and everyone and their grandmother should learn about it.

It states that a neural network with only one layer can model any function if it has enough neurons.

What does that mean? Even if we never innovate on the underlying AI architectures ever again, we’ll keep building smarter AI models as long as we keep making them bigger.

And we will make them bigger.

Meta supposedly trained their Llama model with 2,000 NVIDIA A100 GPUs. Elon’s AI startup, xAI, just recently turned on “Colossus”: a supercomputer made up of 100,000 NVIDIA H100 GPUs.

The investments into AI infrastructure are staggering. GPUs are getting faster and the training clusters are getting bigger.

OpenAI’s o1 model already has an IQ of 120, making it smarter than 91% of people. Soon, there will be no task that an AI agent can’t execute faster and better than a human.

But there is one advantage humans have over AI agents:

Agency.

AI can generate a landing page, but it can’t buy a domain name. It can build a case, but it can’t represent someone in court. It can identify sales leads, but it can’t have dinner with the prospect.

These limitations are mostly self-imposed. These are social and/or software barriers we put up (or leave up) simply because we haven’t yet built enough trust in AI systems to act on our behalf.

Some of these gaps will be filled with robotics, giving AI a body with which to interact with the real world. But even then, we will restrict on where they can go and what they can do in the name of safety and privacy.

The more powerful the technology, the more guardrails we put up around it.

I believe we will continue to grant ourselves more freedom to roam than AI chatbots/robots, even when their performance matches and exceeds ours. At least this will be true for the next couple of decades.

It is in that agency gap that human work will still have value.

Quick reminder - If you like my newsletter, please reply to this email or “add to address book”. These are “positive signals” that help my emails land in your inbox. If you don't like my newsletter, you can unsubscribe below.

See you next Monday — Rayhan

P.S. Check out this Sam Altman interview. Great stuff!