If you couldn't bother to read it, why should I?
doesn't follow. You don't know that they didn't read it! Only that they didn't write it. If the point of things like this is to be careful, then be careful.
I like Oxide's version of this: <a href="https://rfd.shared.oxide.computer/rfd/0576#_llms_as_writers" rel="nofollow">https://rfd.shared.oxide.computer/rfd/0576#_llms_as_writers
The old social contract was that the writer had done more work than they were asking of the reader, but this is torn to shreds with LLM-generated text.
That is entirely invalid considering the existence of ghost writers and such.
No one reads a piece of text BECAUSE the writer had more work writing it than the writer will have reading it (this could end up not being true regardless of AI either way).
The “old social contract” is that readers trust the writers had something to say, and said it. The inner workings have always been irrelevant.
Trouble with genText is that it’s easy for your point to get lost in the narrative exactly because you didn’t write it—it’s like when you know something but can’t put it into words. Your own voice can get lost and so does the appeal to the readers. Nothing to do with effort, entirely to do with the content itself.
I don’t want to read your genAI text because I don’t trust that you managed to get your point across in prose, but I don’t mind reading an LLM’s output because I just wanted a piece of information.
The old social contract was between two humans. Effort was a consequence, not the target.
Fundamentally if I could have got the same outcome from an LLM as the “author” of an AI-generated piece, I value their effort at zero.
The more generative AI content there is in a piece, the less value it has per word.
Deep down our brains seem to know this, too.
LLM content is smooth, without novelty or surprises or shared amusement, and in text we have not summoned up ourselves, we are looking for that as a proxy for connection with the author.
I have used this analogy before but it’s like a rollercoaster without jerk or jounce; there are no thrills or interesting quirks for our brains.
So much so that when you see an AI-assisted piece of any length that has human-written sentences in it, they tend to stick out more, even before you have fully read the text, whereas the LLM words scan past so easily.
The thing that brings this home for me is that the final output of an LLM very rarely makes me chuckle at whimsy or weirdness, but the more disordered thinking traces more often do.
> The old social contract was that the writer had done more work than they were asking of the reader, but this is torn to shreds with LLM-generated text.
The problem is that your biases kick in long before you read the stuff. And I do not see a solution for this.
I write my devlog by hand and then run it through an LLM for a quick sanity check. Grammatical errors and spelling mistakes in stuff you publish in this age of AI is lazy. But I don't do it when I comment on HN/Reddit etc because this is stream-of-consciousness stuff and using AI actively interferes with my thought process.
My code is almost entirely written by LLMs today, but no one can judge how much time I spend on it unless they actively follow my comments on WA, Reddit, HN, my devlog and other places. All they would see is a repo with commits done in the space of a day with nice multi-para commit messages.
This is because I get an LLM to rewrite the entire commit tree once I stabilize a project and have it reproduce the final state in a sequential manner with bite-sized commits. My own development process involves 3-5 word opaque commit messages and a lot of squashing/force pushes: impenetrable/cryptic for anyone other than myself.
There is also the problem with pushing people --- for whom English is a second/third language --- out of entire conversations. Last year, I was talking (on Reddit) to someone from South America who was using LLMs to write in English and a few people were so dismissive of the messages.
I see this as an unstated case of preening.
For better or for worse, the world now communicates in English. You cannot shut people out --- because LLM --- when you yourself would not take the time to read someone if they wrote in Spanish or Russian or Japanese (or the ~25 commonly used languages from my own country).
tptacek · · focus · HN ↗
If you couldn't bother to read it, why should I?
doesn't follow. You don't know that they didn't read it! Only that they didn't write it. If the point of things like this is to be careful, then be careful.
slongfield · · focus · HN ↗
The old social contract was that the writer had done more work than they were asking of the reader, but this is torn to shreds with LLM-generated text.
oejgirjdiejfiej · · focus · HN ↗
No one reads a piece of text BECAUSE the writer had more work writing it than the writer will have reading it (this could end up not being true regardless of AI either way).
The “old social contract” is that readers trust the writers had something to say, and said it. The inner workings have always been irrelevant.
Trouble with genText is that it’s easy for your point to get lost in the narrative exactly because you didn’t write it—it’s like when you know something but can’t put it into words. Your own voice can get lost and so does the appeal to the readers. Nothing to do with effort, entirely to do with the content itself.
I don’t want to read your genAI text because I don’t trust that you managed to get your point across in prose, but I don’t mind reading an LLM’s output because I just wanted a piece of information.
The old social contract was between two humans. Effort was a consequence, not the target.
dofm · · focus · HN ↗
The more generative AI content there is in a piece, the less value it has per word.
Deep down our brains seem to know this, too.
LLM content is smooth, without novelty or surprises or shared amusement, and in text we have not summoned up ourselves, we are looking for that as a proxy for connection with the author.
I have used this analogy before but it’s like a rollercoaster without jerk or jounce; there are no thrills or interesting quirks for our brains.
So much so that when you see an AI-assisted piece of any length that has human-written sentences in it, they tend to stick out more, even before you have fully read the text, whereas the LLM words scan past so easily.
The thing that brings this home for me is that the final output of an LLM very rarely makes me chuckle at whimsy or weirdness, but the more disordered thinking traces more often do.
sieve · · focus · HN ↗
The problem is that your biases kick in long before you read the stuff. And I do not see a solution for this.
I write my devlog by hand and then run it through an LLM for a quick sanity check. Grammatical errors and spelling mistakes in stuff you publish in this age of AI is lazy. But I don't do it when I comment on HN/Reddit etc because this is stream-of-consciousness stuff and using AI actively interferes with my thought process.
My code is almost entirely written by LLMs today, but no one can judge how much time I spend on it unless they actively follow my comments on WA, Reddit, HN, my devlog and other places. All they would see is a repo with commits done in the space of a day with nice multi-para commit messages.
This is because I get an LLM to rewrite the entire commit tree once I stabilize a project and have it reproduce the final state in a sequential manner with bite-sized commits. My own development process involves 3-5 word opaque commit messages and a lot of squashing/force pushes: impenetrable/cryptic for anyone other than myself.
There is also the problem with pushing people --- for whom English is a second/third language --- out of entire conversations. Last year, I was talking (on Reddit) to someone from South America who was using LLMs to write in English and a few people were so dismissive of the messages.
I see this as an unstated case of preening.
For better or for worse, the world now communicates in English. You cannot shut people out --- because LLM --- when you yourself would not take the time to read someone if they wrote in Spanish or Russian or Japanese (or the ~25 commonly used languages from my own country).