Hacker News Reader

Comment by LeoPanthera | Hacker News Reader

This feels very LLM. It has the shape of sense but doesn't actually make any coherent points, other than "how to draw the Sierpiński triangle". It even has some non-sensical AI diagrams to finish it off.

View on HN

I wrote this. If you ask an LLM what to make of IJ as a Sierpinski Gasket, it doesn't yield anything too interesting (it mainly talks about the voids representing the absence of plot points / missing characters). As far as I know the two-ways angle is novel.

The only AI piece is the banner image, because I can't draw. All other images are attributed (Wikipedia, YouTube).

I got the idea for this post after listening to Wallace's interview with bookworm (linked in the post), and subsequently researching Sierpinski Gaskets on YouTube (where I saw the Numberphile and ViralHog videos).

For what it's worth, I probably wouldn't have commented without the AI image. I just tend to assume that articles with AI graphics also have AI text. That assumption is almost never wrong.

This is a good point. The banner image is something I've been doing on all my articles, and it's creating the wrong perception. I'm probably going to stop adding these. All my other content is software engineering related, and I've started adding more visual content like diagrams, screenshots, and YouTube videos. So I probably don't need to lean on the banner images anymore, and it seems like they're doing more harm than good!

Edit: I actually went ahead and removed the banner image from this post. Thank you for pointing out this signal!

Yeah agreed - Just as a future context I tend to close out of articles quickly if they have an AI generated image. Highly reccomend something like Unsplash or Comsos.so - They both have some brilliant visuals, especially public.work - cosmos.so/public-work

Even old archives are great to search for visuals and shows the passion that you've put in to the writing.

This is really helpful, thank you! This post was one of my first with relevant, 'real' images, and I really like how it came out relative to the generated stuff. Besides, I feel like image generation has lost its way since DALL-E got deprecated. I've bookmarked these resources!

Y'all need sense of smell tho

*Has almost never been

Well the summary hit 100/0/0 on GPT-Z, let me see what the others say Sapling: 1.4%

Anyway, the text of the article doesn't feel LLM to me and on first scan, neither the structure. They don't care about minutiae in this way nor do they follow threads. Theirs is a bunch of oh-so-pompously-and-fake-eruditely dispensed rehashes with little pounce.

The hunt for LLM writing is almost as boring as LLM writing itself.

Reply on news.ycombinator.com