hyperultra

As a software engineer, I can accept AI-generated code because the code is not the point of software; the point is software that works, and the code is just a means to an end. Code is usually a terrible mess and essentially no one just reads code, like, for fun. In the spirit of this article, I would love for someone more knowledgeable to articulate what the end product of mathematics is. Is it, as this article mentions, the abstractions that it produces? I've always been fascinated by how the Fourier transform and the quaternion were invented as part of pure math and then used for computer graphics stuff over a century later. To put it another way: should the proofs of theorems that AI is now generating actually be considered end products of mathematics, or are they turning out to be more of a step along the way to developing mathematical ideas, kind of like the code in software engineering?

You have to admit, this guy has created a powerful brand. He can get people to use an agent metaharness by putting T3 in the name just like terminal.shop can use SSH to get people to buy generic coffee.

I think that if you make your request too specific, like in the second example, you might seem less approachable, and you'll cutting yourself off from a lot of potential serendipity. The first question could result in "No, but I know someone else you could talk to." The second question requires the person you're messaging to be willing and able to go down a very specific path.

For about thirty years, the Internet was a substantial store of different kinds of value that mostly weren't being monetized. People are slowly figuring out how to capture and extract that value, which is kind of inevitable. They're slowly catching up.

If you've tried to set up your own email server in the last ten years, you know how it's somewhere between tricky and impossible to get an email delivered without it being blocked by the countless layers of reputation checks and spam filters that are set up to keep public email inboxes usable. It seems like we might be heading that way for all public web traffic, especially "writes" (as opposed to "reads")

There are all kinds of technical details in software that almost no programmers understand - stuff like how their operating system's process scheduler works, or how the algorithm used by the network stack confirms that a packet was received. To have control and direction over a product, humans have always just needed to understand the details of how that product works well enough to deal with whatever can't be taken care of automatically. We will see if that still includes "the code" in a few years time. I think it will still include "the spec," in whatever level of detail that that needs to be in.