frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

X is selling existing users' handles

158•hac•11h ago•77 comments

Tell HN: Apple development certificate server seems down?

107•strongpigeon•1d ago•39 comments

Ask HN: Is Claude down again?

84•coderbants•17h ago•71 comments

Ask HN: What Are You Working On? (March 2026)

285•david927•3d ago•1098 comments

Ask HN: How are people doing AI evals these days?

30•yelmahallawy•2d ago•33 comments

Ask HN: How do we build a new Human First online community in the LLM age?

7•bluefirebrand•8h ago•7 comments

Ask HN: Remember Fidonet?

119•ukkare•1d ago•67 comments

Ask HN: What are you using to mitigate prompt injection?

5•ramoz•9h ago•2 comments

Ask HN: How to be alone?

680•sillysaurusx•3d ago•556 comments

Ask HN: What on this "List of Unsolved Problems in Physics" Has Your Attention?

3•ghastmaster•6h ago•1 comments

Ask HN: Please restrict new accounts from posting

711•Oras•3d ago•505 comments

Ask HN: Most beautiful personal blog UI you have ever seen?

145•ms7892•3d ago•54 comments

Happy Birthday YC/HN

6•ellis0n•7h ago•4 comments

Ask HN: Can I repurpose a Bluetooth voice remote as input device for a PC?

15•albert_e•3d ago•20 comments

Tell HN: I'm 60 years old. Claude Code has re-ignited a passion

1069•shannoncc•5d ago•978 comments

IdeaRank – Startup Analysis Engine

2•TMDev•18h ago•0 comments

Maybe we can keep on coding? pseudo code project

8•EmptyDrum•1d ago•13 comments

Ask HN: How do you review gen-AI created code?

6•captainkrtek•1d ago•10 comments

Ask HN: How are people forecasting AI API costs for agent workflows?

5•Barathkanna•1d ago•20 comments

Why is GPT-5.4 obsessed with Goblins?

16•pants2•2d ago•11 comments

Ask HN: Is GitHub getting less reliable, or is it just me?

13•_pdp_•2d ago•9 comments

Tell HN: Vertical tabs has arrived (behind a flag) in Chrome stable

6•crummy•1d ago•3 comments

Ask HN: Is Starlink still being jammed in Iran?

3•Jblx2•1d ago•1 comments

Ask HN: How to "make it" as a newlygrad/junior?

4•kartoffelsaft•1d ago•11 comments

Ask HN: Finding a purpose after tech layoffs

8•fud101•16h ago•8 comments

Ask HN: Does automatic multilingual support make sense for a launch platform?

2•LeanVibe•1d ago•3 comments

Unlocked SaaS, file source as truth?

2•abmmgb•1d ago•1 comments

The Architecture of an Exit Scam: A Technical Audit of Zszrun

5•cappyfjao•1d ago•0 comments

Ask HN: Since a week HN keeps logging me off every few days, why?

5•epolanski•1d ago•2 comments

Ask HN: What AI content automation stack are you using in 2026?

3•jackcofounder•1d ago•3 comments
Open in hackernews

Ask HN: How do we build a new Human First online community in the LLM age?

7•bluefirebrand•8h ago
I basically grew up online and I have some lifelong friends I met online. I cherish the real potential for international community that can be built on the web, there really is nothing like it

I have been finding myself feeling very bitter about AI lately. I'm angry about how it's seeping into every aspect of life. Not just my work and my hobbies but it also seems to be creeping into many online communities (including this one!)

I have been thinking a lot about how we could possibly build any of the trust that we used to have online. Yes, bots have been a problem for a long time but this is so much further beyond spam posting. LLMs have poisoned the commons online At Scale and there's likely no going back. It has made me very bitter, I won't lie.

However that doesn't mean we can't find a way forward with something new that is somehow resistant to LLMs. I'm not sure what exactly that might look like but I'm curious what ideas others have had.

My wish list would be something that

* Is resistant to LLM "infiltration" for lack of a better word. We should be able to be relatively confident that people on the other end are real humans

* Does not require giving up all anonymity. It will likely require some identity authority but interactions between users should/could be pseudonymous at least

* Ideally is also resistant to LLM scraping. I personally find the thought of sharing work publicly now so LLMs can ingest it is demoralizing

I know it's a big ask and maybe not realistic. I'm curious what HN thinks about this possibility though

Edit: This was partially inspired by the recent mod post discussed here: https://news.ycombinator.com/item?id=47340079

I respect that HN's mod team is willing to sort of leave this up to the honor system, but I think in the future we are going to need some serious ideas to strictly prevent this unwanted behavior, not just hope people will play nice

Comments

up-n-atom•8h ago
wearables. u need to authenticate a literal pulse. essentially a passkey for being alive. we can’t allow a corporate entity control, it needs to be truly open and not like fido.
FloatArtifact•6h ago
I think it's important to focus on the local geographically close group of people that we have relationships first.
bluefirebrand•6h ago
I think you're right, but I've been pretty isolated from local people since I moved to a new city

Time to get back out there and meet people I guess

Johnny_Bonk•3h ago
Check out my project :) feelfamiliar.com
allinonetools_•4h ago
One thing that helped older communities was friction — things like slower posting, reputation built over time, and real participation history. When identity grows from consistent behavior instead of instant access, it is much harder for bots to blend in. Communities used to value that patience a lot more.
yukapero•1h ago
dead end. human verification just leads to a digital prison of ids

the real issue isn't bots, it's humans using ai. i'm doing it right now. English isn't my first language so i used an llm to translate my thoughts for this post. if the tech is this useful for bridging gaps, you can't really filter for a "soul" anymore. the line is already gone.

scraping is a lost cause too. if a human can read it, a model can ingest it

i guess the only fix is to stop scaling. go back to small, private, invite-only groups. intentional friction and making things "inconvenient" is the only filter left that actually works

raw_anon_1111•40m ago
Interesting, your comment doesn’t read like AI slop.