 |
| Warranty Void If Regenerated (nearzero.software) | ||
| recommended by 3 users | 2026 | 5 points by Stwerner | books | original thread | | ||
As an experiment I started asking Claude to explain things to me with a fiction story and it ended up being really good, so I started seeing how far I could take it and what it would take to polish it enough to share publicly. Over the last couple months, I've been building world bibles, writing and visual style guides, and other documents for this project… think the fiction equivalent of all the markdown files we use for agentic development now. After that, this was about two weeks of additional polish work to cut out a lot of fluff and a lot of the LLM-isms. Happy to answer any questions about the process too if that would be interesting to anybody. | ||
| 3 users recommended this thread. What they said: |
| bostik Jul 30, 2026 | on: PostgreSQL's MVCC is bad. So is everyone else's | context |
|
ventana Jul 31, 2026 | on: The End of an Era | context
The problem of AI writing, to me, is its emptiness, not the fact that it's not written by a human. Seeing standard slop words or phrases triggers the "do not read this" reaction, but the reason of not wanting to read is that the text is just empty of any interesting thoughts. The majority of us read ChatGPT/Claude/Gemini responses, even though they are, by definition, slop, because we probably asked some question (the one we did not really care about and just wanted to explore) and the response has some valuable information in it (maybe true, maybe not). So we read it. If (or when?) the AI writing becomes interesting to read, I don't see why we won't start consuming it, and I don't see it's as something bad. One way to make the text interesting is to actually get involved in writing; remember this short story that was here on HN some time ago? [1] It was written by AI with some serious human involvement, and I enjoyed reading it. |
|
fellerts May 16, 2026 | on: Someone Shared a Real Monet Painting as AI and Asked for… | context
I'm reminded by this short story posted to HN a couple months ago: https://nearzero.software/p/warranty-void-if-regenerated The discussion is interesting as many people didn't catch that it was mostly written by Claude: https://news.ycombinator.com/item?id=47431237 |
Your post has a trove of really good data. A lot of care, skill and attention must have gone into building that kind of expertise. When you then use an LLM to produce a post about it, that comes off as lazy. It wouldn't have to be that way, but sadly very (very!) often is.[0]
Writing well is hard. Editing text to be readable is hard. Shortcutting that process is a signal to your audience that you do not appreciate their time. I use LLMs at work to generate client-facing analyses, but I never let a single version go out without doing at least some revising. LLMs love nothing more than to generate another token, which makes their output flowery and verbose. They overuse rule of three, without any of the underlying understanding why it works, let alone when (and how) to break it intentionally.
I spotted the familiar "seesaw cadence" in your post but ploughed through because the content itself was still damn good. It's a shame that you did not put the same level of care into your writing that has gone into developing the technical expertise. Getting the post flagged because of the generic anti-LLM sentiment is undesirable but expected. HN doesn't have a way to mark posts as "sloppy writing but content-wise really good".
It's the same kind of visceral reaction this crowd tends to have towards voice notes: "if you did not bother to condense your thoughts and instead dumped half an hour of rambling for me to dig through, why would I waste as long to listen to you not being able to say it clearly?" ["Podcast for none"]
0: The only exception I can point to is https://nearzero.software/p/warranty-void-if-regenerated (discussed at the time in https://news.ycombinator.com/item?id=47431237).