Everyone's an AI Script Kiddie
Or maybe it's just me.
Have you ever been bored and overwhelmed at the same time? That’s what it feels like when every week there are 17 new ways to think about what AI is, what it does, how to make work with XYZ. Unless you have a detailed spec and a deadline, the whole mass of ideas compel you to be swamped and aimless at once. I mean, what do you pick? One day we’re thinking of MCP in terms of the Tron movie, the next day there are 41 MCP servers for you to select from. It feels like every choice is like picking a dot com in 1999.
I have three main goto guys, out of this following universe. Let’s talk about them.
Nate B. Jones: The Teacher
Nate is an overachiever who seems to have limitless opinions and energy. I’m reading his prompts as if it were code, or a set of tactical communiques to Central Intelligence. I expect that over time, a great deal of what he is producing will mature, but I reckon that he is not so much a practitioner as somebody who understands the internals and is giving us quick kernel calls for our elementary scripting. He just doesn’t stop and I can’t keep up. Nevertheless, I read his code/prompts and I’m getting his patterns. They are superbly practical. He is how I am learning to prompt. I have gotten about 15 sets of prompts that are genuinely useful.
Ilya Sutskever: The Theorist
I always remember his name because he is a former insider who, on the value of his name alone, is capable of generating billions. He is one of those who have questioned the marginal utility of scaling training AIs which has prompted nearly a trillion in investment for compute infrastructure.
I instinctively believe this is correct, that the real productivity is going to be found in the methodologies and of the assemblage of systems that surround LLMs and make what they do useful to humans.
Along the way I seem to recall that he echoed Minsky’s idea of the Society of Mind. AIs making humans think better in concert will have a greater beneficial effect than AIs talking to AIs and taking over many kinds of human labor. In fact, I recall very recently that AIs don’t know how to teach each other, or else they would have already shown an innate ability to do so. This is, I believe, called the ‘learning problem’, and defies the premise of the classic film Colossus: The Forbin Project. So they won’t be taking over the world, it will always be Blofeld.
Tyler Cowen: The Expert User
Cowen is one of the most coherent academics I know and he’s just the kind of brilliant that I like. He asks the smartest questions of any intellectual, and I’ve followed him for decades. I’ll never keep up, especially when it comes to Chinese food.
Cowen is an avid follower and user of chat AIs and knows exactly how to employ them to aid in adding to his knowledge. His is the model of intellectual production that I aim to follow. He says that AIs are smart enough to be very practical right now. So as far as I can see, there’s a kind of infinite roll-out that can be beneficial to everyone below Cowen’s level. What there is to learn is how exactly his production system works -
What he intuitively knows.
What his academic background has trained him to do
What adaptations he makes to make use of AI outputs
What Cowen understand more than the rest is the larger economic context
Dwarkesh Patel is the honorable mention primarily because he’s one of the top journalists out there. He can get the smartest people telling us what is obvious to them.