alt Hacker NewsRe the OP's methodology for detecting bots, down in the 7th paragraph they say it's a conservative lower-bound:
- they label everything that failed the anti-GPT test 'bot' and everything else unambiguously 'human' (even if might be inauthentic or compensated human, a non-GPT bot or a bot with some basic input filter to catch anti-bot challenges). For example commenter Emmanuel/@techinnovatorevp doesn't fail the anti-bot test, but posts two chatty word-salad comments 10min apart that contradict each other, so is at minimum inauthentic if not outright bot.
- even allowing there are other LLMs than GPT, or that filtering the input for 'GPT' after an '---END OF TEXT---' to catch anti-bot challenges
- why not label everything in-between as Unconfirmed/Inauthentic/Suspicious/etc.?
- makes you wonder how few unambiguously human, legit accounts are on ProductHunt.
Replies
I see plenty other non-bot weirdness with ProductHunt's site:
- if you search for "iPhone", click the 2nd tab "Launches" then click to sort by Launch date, the only launches listed since 2019 are: "iPhone 15 Pro Max" (June 5th, 2024) with only 8 upvotes(!), "iPhone 11" + "iPhone 11 Pro" (Sept 10th, 2019) with 208 + 446 upvotes. No launches shown for iPhone 16, 14, 13, 12. (There are some product pages, but not launch pages). Compare to 2,878 upvotes for iPhone X back on Sept 12th, 2017. So it seems the site's been declining for nearly a decade.
Emmanuel/@techinnovatorevp was detected as a bot because they voted on a launch with over 70% bot votes, but yes I don't detect all the bots... not even close actually.
I thought about just marking any account that comments as bot, because that's more accurate than my current formula ;)