4 ms·

Ask HN: What's the Secret to Ranking on HN?

2 points by adaboese 3y ago

I thought I wrote a really interesting article and shared to HN (Using Vector Embeddings to Overengineer 404 pages), but it got 0 upvotes. 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.

gigatexal 3y ago
Periodically reshare it. It could be time of day or really anything.
mtmail 3y ago
Not enough users browse the https://news.ycombinator.com/newest https://news.ycombinator.com/newest queue, there's 2 submissions per minute. I've seen articles ignored and then suddenly the third or fifth time it hits number 1 on the frontpage. It's much more random than you might expect.
- adaboese 3y ago
  so re-submitting after a couple of hours is really the way to go?
  - ThrowawayR2 3y ago
    Maybe once at most. Re-submitting more than once within a few days will likely get it flagged by users.
    - PaulHoule 3y ago
      Real SEO masters overcome this uncertainty by blogging more and recruiting other people to make links.
- PaulHoule 3y ago
  I have statistical models that can predict if a headline will get > 10 votes or if a headline that gets > 10 votes will get a comments/votes ratio > 0.5. (We can’t think of a better headline than “Richard Stallman is dead”) 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.
  - adaboese 3y ago
    How does that model rank the title I shared?
    - PaulHoule 3y ago
      8% likely to get >10 votes, 13% likely to get a big discussion, 23% likely to be "dead" but my "dead" model has problems.
  - pdyc 3y ago
    interesting, can you run it on my blog? i bet it would generate false negative for you.
    - PaulHoule 3y ago
      Got RSS?
      - pdyc 3y ago
        feed.xml is not there but individual pages have xml https://newbeelearn.com/index.xml https://newbeelearn.com/index.xml
        PaulHoule 3y ago
        For titles=["How to use org mode in Hugo","Code reviews should be converging","How to profile cloudflare workers"] I see >> y_vote array([0.07707551, 0.11528609, 0.1505662 ]) >> y_comment array([0.1886978 , 0.53672774, 0.20985652]) If you wrote 10 of those one would be a hit, I believe that. I think "code reviews should be converging" has good potential for a discussion. My model looks at the headline only and does not consider the quality of the post or any metrics thereof. I thought the article was basically right but the use of the word "converging" is unconventional.
jacquesm 3y ago
You're spamming SEO junk, please don't.
- adaboese 3y ago
  In what way on Earth is the post that I am referring to an SEO junk?
  - KomoD 3y ago
    Your entire business looks to me like SEO spam, specifically about generating it
    - adaboese 3y ago
      You are free to judge my business model (which, if you are referring to content generation, YC is heavily invested in), but nothing about the content I've shared is spam. Also, if you dig deeper into what AIMD does, AIMD was built to combat spam. The one-pager is a good starting point.
      - minimaxir 3y ago
        > AIMD was built to combat spam. 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.
        adaboese 3y ago
        If that's the criteria, then fair, but AIMD produces articles that are virtually indistinguishable from what an average copywriter would compose if you hired them. AIMD is very different from your typical article generators (like https://writesonic.com/ 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.
KomoD 3y ago
Post things people are interested in
- pdyc 3y ago
  that is necessary but not sufficient. I too find ranking to be unpredictable.
mindcrime 3y ago
What am I doing wrong? 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-404-pages https://aimd.app/blog/2024-01-16-using-ai-to-overengineer-40... [2]: https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html

KomoD 3y ago

> AIMD was built to combat spam I don't believe that. Especially considering > Anti-AI Detection: Avoids detection by AI content detection tools.

adaboese 3y ago
The secret sauce behind "anti-AI detection" is 'not writing overly generic, unreferenced content'. Not a single line in the code base tries to evade AI detection; it is just the side-effect of how the product works. 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.
- KomoD 3y ago
  > Not a single line in the code base tries to evade AI detection; it is just the side-effect of how the product works. 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-content-detection https://aimd.app/blog/2023-12-14-tricking-ai-generated-conte...
  - adaboese 3y ago
    Well, I am not here to win Internet arguments. Try the product to form an opinion, if you feel strongly to discredit what I am saying. 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.
    - KomoD 3y ago
      > Try the product to form an opinion 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)
      - adaboese 3y ago
        Just migrated to Google Cloud last night. Still ironing out some details. Will check what broke there. 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?
      - adaboese 3y ago
        I just realized – that you are looking at the draft, not the edited version. 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.
        adaboese 3y ago
        Product has been updated to reflect this.