@kkarhan @LillyHerself I'd be very wary of using anthropomorphic terms like "want" and "trying" as it can make people think that transformer algorithms like LLMs have a consciousness (something people are already doing).

On another note, though, "sentience" is often misused, but not here. It just means being aware of one's own condition. A cow knows when it's wet, or cold, or hot, so a cow is sentient. LLMs generally have no such awareness, they are not sentient.

@cg @Ninjaneers Hahaha, that reminds me of a saying they have in Swedish "En mans bröd är en annan mans död"

[it rhymes in Swedish, so non-literal translation: "one man's death is another man's breath" ]

@wall_e @LillyHerself Exactly.

My dad is a pretty competent coder. He did it professionally for longer than I've been alive. He's worked in hardware design for a number of years now, but his coding skills are still fairly sharp. If I ask him for help solving a computer science problem, he's happy to give it to me, and he often has valuable insights -- but I find it often takes me twice as long to explain my problem to him than it would have to solve it on my own.

And he's a human. Why in the ever-loving sweet fuck would I do that with a machine that cannot synthesize solutions to novel problems at all and is pretty bad at synthesizing solutions to ones it's already seen?

@darkmatter_ @LillyHerself LLMs are an extension of the "magic" UI trend we first saw with stuff like Siri.

It's an infinite oracle that speaks natural language! Just keep talking until you get lucky!

Historically, that's not how professional tools worked. You get stuff like "documentation" and "explicit limits".

This creates an environment where we blame the users when the system fails them: they were unfaithful and refused to spend 2 hours coaxing ChatGPT to implement a 15-minute fix.

@dominykas @LillyHerself

Have you ever read " The Colour of Blood" by Brian Moore? It conjures such a vivid picture that after I'd read it, I started again, to see how he'd managed that. And was surprised to discover there is barely an adjective from page 1 to the end.

None of the adjectives in the quoted piece is superfluous. LLMs, by contrast, add an adjective to nearly every noun and it's wearisome reading. I assume they trained them on tabloid newspapers and airport novels.

@LillyHerself

AI coding tools Introduce low quality code in large volumes. It spagettifies existing code and makes code reviews so much more painful. It ignores quality gates, demonstrates lack of understanding of how code works, and would probably be the type to ask "what's a design pattern?" because it clearly doesn't know anything about them for practical use. Not to mention that much of the code is simply wrong.

When adding it into a process, you may as well douse us in maple syrup on a hot humid day in a swampy area. Junior devs who still aren't familiar with coding practices and understanding the codebase try to rely on AI slop and it is immediately obvious because it's a lot of fluff that doesn't actually do what you asked them to do. But it looks right if you don't know what's going on.

2/3