Bursting the AI Bubble?

A Telling Interview Goes Viral

This past week, Andrej Karpathy appeared on Dwarkesh Patel's podcast. Karpathy, if you don't know, is one of the most respected names in artificial intelligence. Along with being on the founding team at OpenAI, he was Tesla's director of AI and Autopilot Vision.

While many podcasts are more entertaining than informative, I found this conversation to be really enlightening. Not only is he a deep thinker, but he is clearly willing to say things that others in the industry aren't willing to say.

For instance, Karpathy pumped the brakes on some of the most aggressive forecasts on artificial general intelligence. In his opinion, it's going to take at least one more decade to get even close to something like this.

That said, in my mind, the more interesting argument centered on agents. While he touted agents like Claude and Codex, he threw some cold water on the actual effectiveness of many agents today. Ultimately, the potential is there, but the timeline is super aggressive.

Right now, agents don't have continual learning. Often, they don't remember the context that you have provided. These are real issues that make it much tougher for them to actually be used in a B2B setting.

The herding effect around AI is (still) all too real. Being "AI-first" is a necessary signal to not only investors and employees, but to potential customers. But at the same time, the net effect behind the scenes may not be as bright. I keep thinking about a recent MIT study finding that 95% of workplace Generative AI pilots fail.

I think that the discourse is like most of the arguments that occur in public markets. The major disagreements center on timing. Agents (and AI in general) will be much more sophisticated in five to ten years.

Nonetheless, there are real limitations right now. Those limitations are often put aside and overshadowed by the hype and excitement around the technology's potential. Consequently, that hype and excitement have inflated some asset prices. A lot has to go right—sooner, rather than later.

Ultimately, I recommend that you listen to the Karpathy podcast. No matter where you fall on the hype spectrum, there are plenty of nuggets here that will help you see where we are and where we're going. Let me know what you think!

Prompt of the Week

This is more of a fun one. ChatGPT suggested Japan and New Zealand for me. Interesting choices!

"With what you know about me, pretend that you are a travel agent. Think of two places for me to travel (outside of the US) that would lead to extremely memorable vacations. I have no major preferences, but will be traveling with two young children."