[pieix]

> AI isn’t a co-pilot; it’s a junior dev faking competence. Trust it at your own risk.

This is a good take that tracks with my (heavy) usage of LLMs for coding. Leveraging productive-but-often-misguided junior devs is a skill every dev should actively cultivate!

[nrb]

> Leveraging productive-but-often-misguided junior devs is a skill every dev should actively cultivate!

Feels like this is only worthwhile because the junior dev learns from the experience; an investment that yields benefits all around, in the broad sense. Nobody wants a junior around that refuses to learn in perpetuity, serving only as a drag on productivity and eventually your sanity.

[simonw]

That's somewhere that the AI-as-junior-dev analogy breaks down a little.

There's still incredible accumulated value here, but it's at the other end. The more times you successfully use an LLM to produce working code, the more you learn about how to use them - what they're good at, what they're bad at, how to effectively prompt them.

[krageon]

It's worthwhile because you cannot do everything and often it is better to have someone far worse than you work on a problem than to just ignore it.

[codr7]

What you're doing is sacrificing learning for speed.

Which is fine, if it's a conscious choice for yourself.

[piva00]

I don't think GP was talking about themselves being a junior using LLMs, at least my interpretation was that devs should learn how to leverage misguided junior, and LLMs are more-or-less on the level of a misguided junior.

Which I completely agree, I use LLMs for the cases where I do know what I'm trying to do, I just can't remember some exact detail that would require reading documentation. It's much quicker to leverage a LLM rather than going on a wild goose chase of the piece of information I know exists.

Also it's a pretty good tool to scaffold the boring stuff, asking a LLM "generate test code for X asserting A, B, and C" and editing it to be a proper test frees up mental space for more important stuff.

I wouldn't trust a LLM to generate any kind of business logic-heavy code, instead I use it as a quite smart template/scaffold generator.

[codr7]

And the end result is you won't learn the details, so you will become more and more dependent on your magic piano.

[piva00]

I know the details, I've been through the wading, thrashing around the docs, the books, I just can't recall the right incantation at that moment and a LLM is more efficient than searching the web.

I still have the skills to search the web if the magic piano disappears.

Don't know why you are trying to come up with a situation that doesn't exist, what's your point exactly against this quite narrow use-case?

[pieix]

Thanks for explaining my intent, you nailed it.

[jdietrich]

It is quite remarkable that we are already at the stage where saying "this AI is about as competent as an inexperienced college graduate" constitutes criticism. It is entirely proper for people to be engaging sceptically with LLMs at their current level of capability, but I think we should also keep in mind the astonishingly rapid growth rate in their performance. LLMs were a toy two years ago, they're now a useful if flawed colleague, but what can we expect two years from now?

[acedTrex]

I mean 2 years ago they were at about the same place, theres been very little practical gain from gpt4 in my opinion. No matter the model the fundamental failure cases have remained the same.

[danielbln]

I disagree, context size alone has exploded from 8k to 200k now and that makes a huge difference. LLMs have also progressed significantly in many other metrics, code quality, understanding, etc. The recent reasoning models have upped the ante further, especially when combined with models that are good at editing code.

[jdietrich]

Reasoning models like o1 or QwQ absolutely destroy 4o in coding, let alone GPT-4 circa 2022.

[tkgally]

Minor correction: GPT-4 was announced on March 14, 2023, less than two years ago. I don’t remember how much LLMs had been discussed as coding assistants before then, but it was Greg Brockman’s demonstration of using it to write code that first brought that capability to my attention:

https://www.youtube.com/live/outcGtbnMuQ?si=oTMA02ns_BJDRS4c...

Advances since then have indeed been remarkable.

[acedTrex]

This is not my experience at all, o1 fails in the same way 4o does.