hola!

I'm Angel Ponce, a web developer by accident. I work at Justia as a Senior Front-end Engineer. I love the internet, movies, and specialty coffee. Fascinated by Language Models, offended by grifters.

Elsewhere·GitHub·Mastodon·LinkedIn·X

Angel Ponce’s feed

· Link

Founder Pronounces Programming (and Bread Making) Dead.

I try to avoid writing angry, but sometimes writing angry is the only way to restore your sanity. I was reading the linked Paul Dix piece (The end of programming). It’s the kind of writing I dislike the most, bad epistemology trying to disguise itself as thoughtful thinking.

I’ll limit myself to criticizing the parts that I find so baffling that they cannot be thought of as anything other than comedy. I of course encourage you to read the entire piece. It’s important to read bad frames to be able to develop and sharpen one’s own.

If you can build a verification system and give proper direction, AI can produce a highly complex, highly sophisticated piece of software and it can continue to refine it until it just works.

There you have it, friends. If you solve the hardest problem, then AI can handle the rest, please be amazed. Building rockets is easy, all you have to do first is define how to build them safely, if you do that, building them is no problem at all.

If you take what they say publicly at face value, they are no longer doing very close review of every line of code that gets shipped.

Another wisdom gem: you must take at face value the statements of people and companies that have a vested interest in you taking their statements at face value. By that measure North Korea is a paradise, per Kim Jong Un’s publicly made statements.

But perhaps the most aggravating one is the one that gave the title to Paul’s piece:

And we will have arrived at the point where we have more running, production software written by AIs than written by humans. It’ll be the end of programming.

You’d have to be a non-programmer to write that; only such a person would think programming is merely the act of typing code. And typing code is the easy part. Well, it isn’t, but it isn’t the hardest part either (I’m being dramatic). The hard part has always been what happens before the typing: the analysis, the decisions, the tradeoffs, weighing when a simple fix is enough and when to take the time to lay a foundation for what comes later. Agents excel at the typing, not at what comes before. Paul’s conclusion is so wrongheaded that you only need to change the domain to see it:

And we will have arrived at the point where we have more bread produced by machines than by humans. It’ll be the end of bread making.

· Quote

And this was a debug session from hell, enormously helped by an AI doing much of the grunt-work.

I’d like to call it my tireless helper, but the AI several times stated flat out that this was impossible and unsolvable and that we should just write a report about it.

I suspect those things have been trained by people who may not be quite as stubborn as I am.

But while the AI was ready to give up several times, it did keep adding debug code and analyzing it faithfully when I pushed. So credit where credit is due and I let the AI write the commit message above.

This is basically a one-liner fixing a bogus “round_up()” to a “round_down()”, but there were 24 patches adding more and more debug information to this, and 18 kernel boot to finally narrow it down to this.

Linus Torvalds · Linus Torvalds on the Debug Session from Hell

We need more of these kinds of honest accounts that clearly show how LLMs aren’t magical “thinking” beings. They are pieces of software that produce their best work when supplied with intent and direction from us, either during training or in user sessions.

I think this part is rather interesting:

“I suspect those things have been trained by people who may not be quite as stubborn as I am.”

I read stubbornness as a proxy, a diagnosis of what is more likely a product of how the models are trained to favor quick solutions. If the right solution isn’t immediately obvious or available after a few tries, most models (at the moment of writing) won’t stop and ponder whether taking a step back is necessary to find the root cause. Other problems that fall into this category are ones where the solution isn’t more code, but less.

· Post

Everything is a Context Problem

I've come to believe that everything about LLMs is a context problem, even when it doesn't look like one. The kind of thing that's hard to unsee. And once you see it, you start noticing who benefits when you don't.