Comment by eru
2 days ago
I suspect this is where our AI agents will really come to shine: helping people who might have a technical bend but are not programmers do cool things.
2 days ago
I suspect this is where our AI agents will really come to shine: helping people who might have a technical bend but are not programmers do cool things.
Yeah. Anyone who hated writing code but had to, like biologists and chemists, now don't have to.
This is arguably a dangerous fallacy, particularly for anything critical. When the person doesn't know programming, it becomes harder to guide agents and do reviews, leading to overlooking issues and accepting mistakes. This gets amplified for any creative solutions or fixing future problems, where it is something the LLM was not trained on or has limited data about the subject. Plus add how much money spent on tokens.
Maybe a way for non-programmers to better understanding what they are saying, is to think about translation. It's kind of like saying a person doesn't need to know anything about Spanish to now sell LLM translated books from English to Spanish.
Yes, the LLM can do translations for you, and quickly. But, as you don't know Spanish, you won't know how good the translation is. You are less likely to catch all kinds of mistakes, from bad grammar to wrongfully translated idiomatic expressions. The less Spanish that you know, the more likely you are to be impressed or fooled into a false sense of security. Then when native and fluent Spanish speakers read your books, they may rate it as mostly gibberish and nonsense.
I think there is negative bias. The counterpoint being that with code you can test input vs output. With foreign language you cannot. Programmed system, either computes what you want or it doesnt. For Spanish you either have a person who can do testing for you. Good practises and cybersec questions matter in programming, but system doing what it is supposed to do (success path) matters more imho.
The current situation is that the person that doesn't know programming will do it anyway. They will likely make more mistakes than an LLM will.
Better still, the llms are easier to get on with than real developers. You explain to a real developer what you want, after a few weeks you find every word you said has been understood in the frankly weirdest way possible and every grey area, instead of asking, has gone way off piste.
The same goes for the LLM, but it replies in minutes and costs next to nothing, so you can correct it quickly, and you don't even have to be politely patient with it to avoid hurting its feelings
I'd argue we're there anyway with all kinds of non-programmers using languages like Python to achieve their goals.
On one hand, it results in some absolutely horrendous code that takes 100x longer to run than necessary, because it lacks the most basic principles of engineering. On the other hand....it works and does what is asked of it.
1 reply →
I'm mixed on this. They may not(or may) write elegant code, but they have clear intent about its parameters and purpose. I've seen LLMs mess up numbers by order of magnitude on multiple occasions, which could range anywhere from annoying to fatal, depending on the field.
Like all things, helpful as long as it's checked very carefully.
The less physicist code exists in the world the better.
The problem with super smart people is that they often succeed without having to or bothering to understand even the most basic common sense techinques.
I had a friend who told me a story, that when he was a kid, he found a copy of the Borland Pascal DOS compiler on his dad's machine. It came with a demo demonstrating the graphics library, but that was the extent of his introduction to Pascal, most of the syntax he figured out by trial and error.
He went ahead and build a quite elaborate tank game, with things like enemy AI, multiple weapons and destructible terrain.
The code was a fever dream, but the game was pretty impressive, and enjoyable.
One of the things he didn't intuit, was that you could not only use built in functions, but define your own. So when he needed a code block a second time, he just copied it, so you can imagine how pleasant that code looked.
They do still write code, to a point, but it's in human language with a humanized feedback loop instead of the lower level of code.
Yes. And in addition, Claude and Kimi and friends will happily do version control and write tests etc.
They'll even volunteer to do so and recommend.
Not really though. There's a whole product segment like LabView, which has existed for over three decades now, for people like biologists and chemists who do not like to code.
Those tools are built for quick iteration. Compared to that, using an LLM surely feels a lot like programming. Also a lot less deterministic.