Jason GormanSomeone now trying to argue that some of the most renowned software developers in the world think that LLM-generated code is "just fine".
Steve Fenton ➜@jasongorman my experience is that a talented developer fixes the code as they go, but still attributes the result to the LLM. So they overestimate how good it would be for a developer who can't "make it good" as they go along.
@stevefenton I literally just posted something on LinkedIn about "Clever Hans", the horse who could "do maths" 🙂
@jasongorman I just went and found it, boosted it, and reported the two comments from a fake account pretending to be LinkedIn! Doing Asherah's work!
@stevefenton I've seen quite a few developers displaying confirmation bias when we discuss sessions at the end. I ask how much of it they think the tool did. They remember it doing most of it, but I've just watched them correcting the model every other interaction, getting stuck and editing the code themselves many times. I also see it a lot in video tutorials.
The "I" in "AI" is *us*
Jeff GriggLLMs are trained on code that I think is crap. Sure, I use code from Stack Overflow and other sources. But I refactor it to within an inch of its life before moving on. How could I expect any better from LLMs?
Milan Todorov@JeffGrigg, the majority of the code out there in the open is really crap, so that's what the LLMs have seen, and that's what they are going to spew. Tons of it.