Item — HN Reader

There’s also a class of “detailed agent written technical blog” that’s you can just scan and get the shape that it’s long paragraphs, tables of statistics, and dry LLM writing. You notice the blog has one or more such posts a day.

It’s a turnoff immediately if someone shares what their agent wrote.

It also has happened a few times they contain a fair number of misrepresentations and mistakes.

2 comments

On the other hand, it could be great to have collapsed sections clearly marked as LLM with the name of the model, the prompt, the assumption the author somewhat checked it and didn't use it to write the article.

You can put it inline or at the end of the article: references, citations, further reading, "AI".

(It might also be the only way to stop training models on their own output)

Are you speaking in the abstract, or would you, personally, like that when reading a blog?

Speaking for myself, I have absolutely no interest in reading anything written by an LLM that I didn't prompt myself, under really any circumstances I can think of.

> Are you speaking in the abstract, or would you, personally, like that when reading a blog?

It's better than the current status quo, where people post LLM writing without attributing it to the LLM, hoping you'll get tricked into reading.

At least if it's labeled like the GP describes, you know what you're getting and can make your own decision.

On the other hand:

> (It might also be the only way to stop training models on their own output)

On the other hand, I think it's a moral imperative to make it so the models train on their own output. I have absolutely no interest in giving them clean input data. After reading about how ISBNdb is helping them destroy old pre-LLM books for input, I totally think someone who does not have small children should start creating print-on-demand forgeries of such books (with the same ISBN numbers) filled with the most garbage-quality LLM text possible.

I don't mind it for footnote/reference-book type material, where you want to have some information available for the curious but 99% of your readers won't care.

A few dense paragraphs about a database implementation that's ancillary to a story about the content of the database, for example; a small percentage of readers might want to reimplement or host A Thing locally and it could be helpful to know that it's this type of database with that number of tables and takes up x gigbytes on disk etc.

You are thinking of something nice to have. At least half the idea here is that it would be nice not to have something.

Usually there are many buttons and links one wouldn't click on.

If authors can't resist having chunks of copy paste slop in the middle of their article they can do it in a way I don't have to see it.

They also get a fair opportunity to impress me with a prompt that looks interesting enough.

And they get to distance themselves from the idea that copying [say] a Wikipedia article into a blog posting is the same as writing something.

I don't see a lot of value in that, but one could supply a lot of semi structured extra information which could aid the readers LLM to answer questions the reader might have with more accuracy / factual grounding.

It's a turn off both because you don't know if it's accurate at all, and because LLMs have a way of turning 1 sentence ideas into pages of diluted slop.

Even if it used to be accurate. Percona were a great resource and now it's all giving LLM vibes. Maybe it's still accurate, it probably is, but it also feels like they don't care