Wow, a unique algorithm is worth writing a paper for!

• Really, #NotCool!

Holy shit, what an incredible way to admit you're a talentless hack

My largest reason for eschewing LLMs is that the projects I am most proud of were the ones solving problems that no one had ever solved before. If it already existed, we would just use that!

Still, it motivated me to learn about narration, how to discuss and reason about it, how to show without telling, and when to tell without worry.

A few years later, while debugging a convoluted execution chain, I realized the same rules applied to code, tests, and documentation. After all, if it’s useful, a human’s gonna read it sooner or later.

@kubukoz

Their argument is also…just straight up false.

Like, it’s been demonstrated time and time again that _linguistic_ ability is tied much more closely to “programming” ability than mathematic ability is.

This is _why_ it’s interesting that LLMs can write any code _at all_. Their ability to write code is an _emergent property_ of the model, likely stemming from the fact that computer languages are, at their core, fundamentally the same as human language.

….but as we all keep saying, writing code isn’t the hard part. Just like writing the sentences of a novel isn’t the hard part.

Neither code nor prose changes people’s lives because of the mechanisms we lay on the page or in the editor, and it’s here that your OP is sort of right. Sort of.

The magic of software comes from the unique and interesting ways those mechanisms interact. LLMs do not exhibit this kind of thinking or creativity because they _don’t think_. The day we get a deterministic compiler from natural language to machine code will be a _huge_ win for society. It’s just also no where close to being here in any sort of sustainable way.

"Skill without imagination is craftsmanship and gives us many useful objects such as wickerwork picnic baskets."

Tom Stoppard, via Civilization VI.

@winter

> The idea→code pipeline in my head is not obviously distinguishable from the prompt->code process in an LLM

Frankly speaking, we can tell. It’s gratingly obvious when a code is written to just do something without any attempt to tell a story.

As an aside, the experience that contributed the most to quality of my work was an attempt to help edit a my little pony fanfic.

As it stands today it's utterly unethical however "good" the code turns out to be on grounds of water and energy waste alone. <-- see one aspect only

• The code quality is not good enough
• The resource usage is abysmal
• The dumbing down effect on the users is verified.

Maybe that's not the "tech to set us free"

@mjg59 that's your answer. Your take stinks.

Edit: it's an understandable shitty position considering that the "AI" goldrush shuffelmaker NV is your current employer.

cuius enim panem manduco, carmina canto