“It’s basically a smarter Google.”
That’s how a senior allocator described AI to me over dinner last weekend.
By the next morning, our fund-monitoring agents had processed new documents across the funds we track, checked the extracted data against our rules, and posted exceptions to a Teams channel. And all this before my first coffee.
We were both talking about AI, but we were drawing on very different experiences.
Andrej Karpathy recently put some back-of-the-envelope numbers on that gap:
- ~6 billion people have barely encountered an LLM.
- 1–2 billion use free chatbots for search and writing.
- ~20 million, largely in code and maths, delegate substantial projects to advanced agents.
- ~5,000 have access to systems inside frontier labs.
His argument is that the gap in understanding is widening, and our reality depends on the problems we feed these systems, how we can verify the results, and which systems we can access.
For asset owners, that gap can cost us twice: once by underestimating what AI could do to a business we own, and again by believing a fund manager’s AI story we cannot properly assess.
The answer is not to become more enthusiastic about AI, but to become better at judging it. I tried to make the case over dinner, but there is a limit to what an explanation can bring across.
Give an agent a problem that matters, then check the work: what did it get right, what did it miss, and how much supervision did it actually need? That experience gives you a better basis for questioning both the skeptics and the sales pitches.