Jordan Morgan✍️ New Post ✍️
The Abstraction Layer: https://www.swiftjectivec.com/The-Abstraction-Layer/
Matt Massicotte@jordanmorgan very interesting!!
So, today, do you think of programmers as abstracting away the code from product designers?
@mattiem It's hard to say, what made me think of it originally was when I was chatting with my professor from college. I asked her what it was like trying to teach the next wave of programmers. She said it's been extremely difficult because they all just want to use AI to do homework, etc, for them. So, if the models don't keep improving, we could look at a whole generation of software developers who don't really understand code as well as we do. Interesting times!
@jordanmorgan yes I think that future is a certainty, unfortunately. I don't see how it would be limited to just software developers either...
It's possible the whole question about abstraction is pure semantics. I do not think AI is an abstraction (and absolutely not a compiler). I don't have answers, but I definitely do enjoy thinking about these things philosophically!
Kyle Hughes@mattiem @jordanmorgan I waffled on which direction to try to hijack this conversation toward and landed on:
Elon—paraphrasing, and I can’t believe I’m invoking him—predicted that by the end of this year LLMs will be directly producing binaries more efficiently than any compiler. On one hand it’s a farfetched timeline, and the data collection is hard to reason about, but on the other hand I don’t see why that isn’t a curve that can be fit to with enough effort. What does that make them?
@kyle @jordanmorgan I'll take this bet! I think that prediction is nonsensical (and not just because of who made it, but that certainly does hurt its credibility)
The amount of context loss for such an arrangement would be tremendous. And it would simultaneously also remove virtually all well-defined language-based guardrails. And all for what for benefit exactly?
But still not an abstraction! "delegation" is the only word I have ever see that feels satisfactory.
@mattiem @jordanmorgan thankfully I have a derivative Polymarket contract paying out if you take this bet
Dominic Hopton@mattiem @kyle @jordanmorgan if you can generate a binary from a prompt directly, abstractions don’t matter because you can just generate it every time! and who needs language guardrails when you can generate a binary directly?
@grork @kyle @jordanmorgan I suspect this will actually be possible, but what I bet is that the result will be worse in many dimensions except perhaps for the nerd cred of it being possible.
Manuel Carlos@grork @mattiem @kyle @jordanmorgan why generate a binary at all? If we get to a point where that would be a better approach , we'd likely get whatever we're looking for directly from the AI - a binary would be a quaint thing at that point
@manuel_carlos_ @grork @mattiem @jordanmorgan This is the direction I enjoy thinking about. It reminds me of this article I recently read: https://whtwnd.com/austegard.com/3mgxahx5axp2c
Yes, LLMs Can Be Computers. Now What? | Oskar 🕊️
Yes, LLMs Can Be Computers. Now What? A raven's-eye view of validating Percepta's claims — and the questions that raises On March 11, 2026, Percepta published "Can LLMs Be Computers?" The post makes a bold claim: you can compile a program interpreter directly into a transformer's weight matrices, ...