Ask HN: Will LLMs increase or decrease the value of human-written text? | Hacker News Reader

Ask HN: Will LLMs increase or decrease the value of human-written text?

3 pointsgolly_ned6 comments

LLMs let us generate plausible sentences that may as well have been written by a human. I’m expecting there to be a flood of LLM-generated text on the internet in the coming years.

I’ve been wondering about the value of human-written text. Specifically, both Twitter and Reddit have closed off programmatic access to their vast amounts of text, having supplied training data for free for so long. Do you expect there to be a premium on human-written text? Or for LLMs to displace the need for human-written text?

View on HN

Call me crazy but I don't think it is possible to overestimate the value of language, especially written language. Language is a tool for communication between humans. An LLM can enhance a human's ability to communicate but it can't take a thought or a feeling and express it in language without a human instructing the LLM on what to say.

In the same way the paper and pen, the typewriter, the telegraph, or the internet changed how people exchanged various kinds of language, LLMs will greatly change how we communicate, but LLMs are fundamentally about generating text - so I would think it would increase the value of very good written language. Hopefully it will also greatly help people who struggle with communicating themselves with written language.

LLM's need to be trained on something, and the more the 'market' for text is flooded with ChatGPT output, the more relatively scarce human-written text will be. However, if in the long-term, LLM's hinder peoples' ability to string together sentences or otherwise affect the quality of their writing, then (modern) human-written text might not be so valuable. In this case, the value of past human-written text will skyrocket (Twitter, Reddit, etc. will make a killing).

It will be both. "Content", meant in a derogatory way, will go to zero, ideally on the demand side as well as the supply side because I think the existence of llms drives down the value of filler content in the wild as well. And thoughtful analysis of good writing that llms are not capable of will become increasingly sought after. I can actually see a future where llms help clean up the internet by driving demand away from crap meant to draw viewers toward ads and towards actual human viewpoints and analysis.

That only seems possible if generative models suddenly "break" the attention/ad driven internet as we know it and force in a better model

Its possible that the internet settles into a mediocre quagmire, or breaks into an even worse model (like siloed off chatrooms and subnets)

Category mitosis.

I said something similar in my response. I think your first point depends on how much reliance on LLM's determines things like quality of writing and general creativity. For now, comedy (which is a big source of reposted content) probably will always come from humans. To the extent that LLM's have invaded comedy, it's from humans using them in funny ways and reposting that.

Reply on news.ycombinator.com