@plexus Language is a tool. It’s a versatile tool, but tool. Not every word needs to be skillfully crafted and sweated out for weeks before it’s committed to the page.

I’m in no position to tell you how to feel about LLMs. Your position is certainly valid and I share it _in some contexts_. LLM is unlikely to write your favorite novel. Not this year, anyway. But it’s probably OK for all the bulshit elsewhere. It can write a decent marketing copy. Results are not stellar but you probably wouldn’t be able to tell the difference from something written by a person. And like it or not there’s a lot of it out there and demand doesn’t seem to shrink any time soon.

Since most of us are geeks around here let me suggest an analogy. Back in the day we used to craft software in machine code. That was to hard so we came up with assembly which was pretty close to machine code but easier to work with. And what we could do with assembly! Read source code of PacMan, or Zork, or Impulse Tracker. Those are works of art.

But then heathens came up with compilers. How those messed up our machine code! All those automatic optimizations… unrolled loops, and constant elimination, and reordering, and speculative execution, and undefined behavior exploits… The blasphemy! The garbage those compilers spit out is unreadable. Will a compiler ever come up with a fast inverse square root on its own?!

Well, it might since it’s us who’s in charge of teaching that optimization to the compilers.

Machine code/language is a tool. It’s primarily use to achieve some goal that is mostly unrelated to the tool itself. It is OK to automate production if the output quality doesn’t suffer for the purpose. Like compilers, LLMs are fine for some use cases.

We still have people who write ingenious assembly and beautiful poetry and we always will.

Let me give you another analogy. In games there’s a notion of skill floor. Effectively, it’s the basis where a player can play the game. We have tutorials in games to ensure everyone has the basic skills in order to not get stuck. There’s also skill ceiling. It’s how many unessential skill you can get before becoming the best player you can be. Games want this interval to be as wide as possible. This would mean that many people can start playing but also they won’t get bored fast as there’s new skills to learn and we tend to be excited about new things.

LLMs are effectively widening this gap with language. Ceiling remains high. But floor gets lower. People who couldn’t write marketing copy now can guide an LLM to produce something acceptable. People who are just starting to learn English and are unsure of their skill now can write an OK blog post on the topic they care about. More people can use language effectively with LLMs. At the same time the value of good prose/poetry remains the same.

I see why you might think the language is devalued, though. Providers of LLMs promise limitless ability to do anything with their product. In that light results are certainly disappointing. But if we look at LLMs and see what they actually are we might get some use of them.