What am I doing wrong? Is the article not as interesting as I thought it is? Is HN not my audience?
Just looking for advice for future posts.
Probably nothing.
Is the article not as interesting as I thought it is?
Didn't read the article[1], but the title sounds interesting enough.
Is HN not my audience?
I don't think it's that. As others have said, the /new page gets a LOT of traffic, and not that many people read it and upvote stuff. So there's a lot of competition for not so much attention. Given that, luck is going to play a big factor. If somebody who's particularly interested in vector embeddings happens to see your post and upvotes it, that can trigger a cascade where it gets a bunch of upvotes. If they don't, it can remain at 0 upvotes forever until it vanishes off the first page of /new, where it's (much) less likely to ever be seen.
I wouldn't worry about it. Other than to say, if your marketing strategy depends on getting stories on the HN front page all the time, or on any kind of predictable basis, you're probably barking up the wrong tree. Submit stuff that you think is interesting and just let it be. Don't try to "game the system" or you'll wind up shadow-banned (either your user account, or your domain).
Note the following blurb from the HN Guidelines[2]:
Please don't use HN primarily for promotion. It's ok to post your own stuff part of the time, but the primary use of the site should be for curiosity.
[1]: https://aimd.app/blog/2024-01-16-using-ai-to-overengineer-40...
Both of those are treated as probability calibrated classifiers because of the extreme variations in outcomes. You might post something one day and get 0 upvotes, another day somebody might post it and get 500. Not that many people read the “new” page so it is really iffy if you get your first few votes.
In principle you could get more information out of the data treating it as a regression problem (e.g. give more points for predicting that an article gets 500 votes as opposed to 50) but one model I made never predicted that anything gets more than 16.7 votes which means the error function is going to always look awful and not really be a good tool for improving the model. The logit models I”m using, however, give realistic scores in the face of uncertainty.
titles=["How to use org mode in Hugo","Code reviews should be converging","How to profile cloudflare workers"]
>> y_vote
array([0.07707551, 0.11528609, 0.1505662 ])
>> y_comment
array([0.1886978 , 0.53672774, 0.20985652])
Also, if you dig deeper into what AIMD does, AIMD was built to combat spam. The one-pager is a good starting point.
Generally, the full AI generation of articles with minimal human intervention that your startup promotes is considered spam, even if it's more "honest" spam. Your one-pager has the line "Finally, AIMD takes over the writing process, crafting the article for you."
That alone may not affect people on HN upvoting an article on vector similarity but it could get it flagged if people started looking deeper.
AIMD is very different from your typical article generators (like https://writesonic.com/, which good God I don't know why anyone would use, but they are doing good, so good for them) that is just spinning out LLM-written content. AIMD orchestrates a multi-step process of topic discovery, source research, content summary, fact validation, hypothesis generation, outline planning, and only then writing. Content generation is a tiny surface area of AIMD, for which I am the least excited. It is really meant to generate a data-rich starting point for writing a high-quality article, rather than produce something that's ready to post. That is also reflected in the product: the content score is based (among other variables) on the % of insights added by the editor VS generated.
That may well be true, but my take would be that whether something qualifies as "SEO Spam" or not depends largely on the intent of the content. If it's intended to provide useful information that will benefit the reader first and foremost, and has a side effect of promoting your business, then that's good. But I'd consider content intended primarily to function as part of a "content marketing" strategy or SEO campaign to be "SEO spam" no matter how well written it is, and regardless of whether it's written by a human or an AI.
I mean to cast no judgment on your business (hell, I could see giving it a try at some point, truth be told) but just be aware that a LOT of people are going to have the "you're just pitching SEO spam" reaction seen above. Whether that matters to you or not, is up to you of course.
I don't believe that.
Especially considering
> Anti-AI Detection: Avoids detection by AI content detection tools.
At the end of the day, my job is to sell to the ICP that I am targeting, and I know that their concern (based on user interviews) is that content will be detected as AI detected. I need to get them into the product first before I can educate them.
I don't believe this either.
> Note that AIMD already implements techniques to evade AI detection. The techniques used in this demonstration are for illustrative purposes only. The actual techniques used by AIMD are proprietary.
> here is the same article scrambled using AIMD anti-AI detection techniques
> Anti-AI detection is only available to paid users of AIMD
This sounds to me like you specifically made something to avoid detection
https://aimd.app/blog/2023-12-14-tricking-ai-generated-conte...
As mentioned before, sales gen (which is the blog post you shared) and product are not going to match 1:1. You have to reach the audience at the points that you know they research and try to solve. That's not going to be a surprise to anyone who built and scaled products.
I have, it got stuck, and the little it did generate used some pretty terrible sources
Like:
- An announcement about a seminar relating to the subject
- A poorly written blog post by some company (numerous spelling mistakes, poor grammar and factual errors)
- A link to a store page for buying a book relating to the subject (a book that hasn't even been published yet)
RE terrible sources, the sources are chosen based on what Google identifies to be credible sources. I am aware that some of the sources Google chooses as the source of truth are not great though. I've not found a good way to filter them out.
Out of curiosity. Which source in particular stood out as low-quality?
The entire article is edited after the initial draft is written out.
Good feedback. Didn't think that many people will look at this first paragraph is indicative of the final result. Will think how to reflect that in the UI.