LLM Inevitabilism

In the 90s a friend told me about the internet. And that he knows someone who is in a university and has access to it and can show us. An hour later, we were sitting in front of a computer in that university and watched his friend surfing the web. Clicking on links, receiving pages of text. Faster than one could read. In a nice layout. Even with images. And links to other pages. We were shocked. No printing, no shipping, no waiting. This was the future. It was inevitable.

Yesterday I wanted to rewrite a program to use a large library that would have required me to dive deep down into the documentation or read its code to tackle my use case. As a first try, I just copy+pasted the whole library and my whole program into GPT 4.1 and told it to rewrite it using the library. It succeeded at the first attempt. The rewrite itself was small enough that I could read all code changes in 15 minutes and make a few stylistic changes. Done. Hours of time saved. This is the future. It is inevitable.

PS: Most replies seem to compare my experience to experiences that the responders have with agentic coding, where the developer is iteratively changing the code by chatting with an LLM. I am not doing that. I use a "One prompt one file. No code edits." approach, which I describe here:

https://www.gibney.org/prompt_coding

As someone who has historically been very much an LLM inevitabalism skeptic and has recently decided that we've crossed the breakeven point with indiscriminant use of Opus 4, eh, it's precisely because we're in late LLM === AGI hype world. They're actually cutting the shit on "this can do anything, and in a month, twice that!". This thing is crazy operator aligned, wildly SFT'd on curated codebases, and running a TTFT and cost that means it's basically Chinchilla maxed out, back to work boys, sell some NVIDIA stock.

This is precisely the opposite data point to the one you'd expect if the TESCREAL hype men were right: you do that when the writing is on the wall that this thing is uniquely suited to coding and the only way you'll ever do better than quantize and ad support it is to go after a deep pocketed vertical (our employers).

Nothing whatsoever to do with making a military drone or a car that can handle NYC or an Alexa that's useful instead of an SNL skit. That's other ML (very cool ML).

So the frontier lab folks have finally replaced the information commons they first destroyed, except you need a nuclear reactor and a bunch of Taiwan hawks that make Dick Cheney look like a weak-kneed feminist to run it at a loss forever.

The thing is, this kind of one ine itabalism isn't new: David Graeber spent a luminous career tearing strips off of hacks like Harari for the same exact moral and intellectual failure perpetrated by the same class warfare dynamics for the same lowbrow reasons.

Can you translate "SFT'd" and "TTFT" and "TESCREAL" for the less clued-in members of the audience? On "one ine itabalism" I just gave up.