That's awesome!
I watch the posts on libera IRC in ##hntop.
Keep hoping for a hexchat script to accomplish this...
Show HN: AI-Less Hacker News
11–20 of 32 posts
Re: Show HN: AI-Less Hacker News
#12I had no idea you could submit a post with a link and body-text. Is that new, or did I just never notice?
It's pretty new, but AFAIK only works for Show HNs.
Re: Show HN: AI-Less Hacker News
#13This is great
6 months of everyday AI spam is pretty annoying
Re: Show HN: AI-Less Hacker News
#14We need a tool that will filter out hn comments and posts generated by ai too.
Re: Show HN: AI-Less Hacker News
#15We need a tool that will filter out hn comments and posts generated by ai too.
Sure, now prove you're human. You have 30 seconds.
Re: Show HN: AI-Less Hacker News
#16This is great 6 months of everyday AI spam is pretty annoying
One person's spam is another person's interesting content.
Re: Show HN: AI-Less Hacker News
#17Just for fun I wanted to do a simple server-side version of this where the submissions would be truly hidden on my account, so it would take effect on mobile too. And avoid client side artifacts like messed up numbering. https://github.com/tfederman/hacker-news-topic-hider
I wish I had found this before I made my own client-side version! https://news.ycombinator.com/item?id=35352160
Re: Show HN: AI-Less Hacker News
#18This post made it through as well (though it's fine since it's not about LLMs) - "Segment Anything Model (SAM) can "cut out" any object in an image". Perhaps use GPT to determine whether a post title is ML / LLM related ?
That would be pretty cool, but I'm not sure how much money I'd need to pay per request for something like that. Maybe I can train a model offline and bundle it with the webpage somehow
GPT-3.5-turbo costs one dollar per 500k tokens so if you just analyzed the title the cost would be negligible. Probably a bit more if you wanted to include the first few paragraphs of content.
Honestly, though, you can probably use something like the Vicuna model that was on the front page yesterday without any fine tuning of your own. You could even add some RHLF feedback if anything slips through.