The beginning of this article turned me off it felt worst and full of jargon. For example, this is not Byrne's best writing (I only realized at the end that this Diff post was actually written by another contributor, Nikhil Davar):
This intelligent switchboard turns aggregated workflow intent into a dynamic economic inefficiency index that is programmatically extensible/legible to the growing universe of first party and third party, increasingly AI-powered applications purpose built to solve specific problems end-to-end.
I had to read it a number of times before I understood it, but what it says is interesting. And later on, there is a slightly better way of putting it:
OpenAI is admitting here that their strategy isn’t primarily rooted in getting better and better at gathering the correct context and invoking the correct first-party tools to solve your problems end to end inside ChatGPT. Instead, they’re saying it's about building a machine that deeply understands your problem, and knows how to make that understanding maximally legible and usable by the highest-fidelity (increasingly third-party, increasingly AI-powered) end-to-end solutions.
But as the article goes on, it gets to a number of observations that are good to think about (although, it feels a little like they just wanted to mention everything AI and econ that happened in the last few weeks...)
Here's one interesting thing: he goes on to hazard, by extension, this definition of AGI:
When they say they are building AGI (which Sam defines as AI that can automate most or even all economically valuable tasks), they don’t mean building one super intelligent model that suddenly qualifies; they mean building the coordination machine that systematically discovers and matches the economy’s highest value problems with the best, currently available solutions, ideally while continuing to improve itself.
And further:
we probably shouldn’t be thinking about AGI as something that will take all of our jobs. Instead, we should ask if it can get so good at understanding the economy that it will be able to route human talent, capital, and digital agents to their most productive uses.
As an example:
Today, cars are still a significant chunk of the economy, but a far larger portion of economic activity exists because people can drive to it.
Overall, I'm not sure The Diff makes the most compelling case for the revolutionary nature of ChatGPT being able to programmatically access third party apps and tools, but I appreciate the observations he surfaces in exploring the idea.
#NoHeroes, and definitely#NoEmperors