Personally, I like how HN focuses on content and discussions rather than individual users. If I wanted to follow experts, I'd probably curate a selection on a social network like Mastodon, or kludge together some RSS feeds.
Also, I feel like this tool selects for active commenters, not for knowledgeable experts. Not to mention throwaway accounts.
Personally I don't read user names. Easiest way to focus on the comments. Of course, the lack of avatars and signatures and stuff like that helps a lot.
I have absolutely no idea who I've had an argument with on HN ever. I'm sure I've had a few.
It's interesting to keep track of some known people. Sometimes you get to see (for example) a thread with cperciva, tptacek and animats - and I think it makes it more fun to read when you know...
I do look at the commenter's name (same as for an article's author) as I know quite a few commenters personally (some are former colleagues) so our replies sometimes refer specifically to something the other person knows or did.
Obviously this applies to a negligible percentage of total commenters, but as I only comment on certain topics I’m more likely to encounter friends with the same interests/experiences.
the same, I get weirded out sometimes when I read a remark from someone who evidently has decided to keep track of who I am on the site. Seems like a lot of wasted mental effort for little reward.
The negative space should be the most interesting, since absence trends are the hardest to recognize. What are Hacker News’s quantifiable blind spots? Answers on a postcard.
But curating such a list of experts to follow takes quite some effort. It would be great to have a tool that helps with that.
And sure, ideally you wouldn't need such a list of trusted experts but just focus on content. There even was a time when this worked - you could just type "what is the best database to use" into a search engine, and get a helpful result. Not anymore. On HN it may still be better than elsewhere, but ultimately it's a similar issue.
Thanks! Those are fair points. We're thinking we could uplevel the social layer so you can connect with people of similar interests for deeper connections. In this way we compute not just your contributions but how they relate to others.
The social web died, all you're doing it making pitchfork and torch 2.0 for mobs.
If you want to add value and not bloody public spectacle rank comments instead of users.
I have a bunch of low quality posts here when idiots piss me off, but also share world first research and breakthroughs I've been involved with the rare time the counter party is worth talking to.
With such efforts I think it is time for people start deleting their accounts, not because they want/need to hide anything, but some people may want to stay semi-anonymous by using aliases and data mining everything and correlating it with other sources (e.g. LinkedIn) may help identify them and cause trouble (e.g. someone wrote something about their workplace without naming it, but hey, it is "John Doe that works in MegaCorp".
It’s great that you finished it but I wish you had taken a little more care to protect users here. Some people I respect a lot are a little more vulnerable because of this
I think that especially in CS, since applications of which touch on nearly every possible field of knowledge, computer scientists often run into trouble of assuming they know more than we do.
Mine seems fairly accurate - accelerating RDP, 3d printer firmware, touchscreen Graffiti gestures (which kind of dates me) and mainline Linux in SBCs, all of which are my hobbies (I seldom, if ever, comment on actual work, like cloud architecture, networking or... AI, which is too crowded a field).
Weirdly, almost no mentions of Apple stuff (which is supremely ironic given my blog).
So you get an impression of what people discuss here, but not necessarily what they know :)
Thanks dang! Amazing to me how much HN brought Wilson and I together after 16 years and across the world. We're hopeful about the future of trust on the internet exactly because of communities like HN.
So, you I assume you extracted my mail out of the profile text (which admittedly was not obscured enough for the LLMs of today), to put it into a mailto: link - Well, many thanks on behalf of the low-effort spammers for making harvesting easier, I guess...
I do wonder if low effort spam bots scraping mailto links are still a thing or if we are all just cargo culting obfuscations of emails. Spam has changed so much from those days, is it a strategy that is still used?
I think even basic LLMs are able to de-obfuscate our 2000s-style attempts at hiding emails (eg. one of my friends writing: "firstname [cute arobase sign] domain.com" on the contact page of his super-popular blog)
I remember there was a (rather controversial) tool posted here a few years ago, which used textual analysis and stylometry to find “similar users.” So you could type in someone’s username and find their likely alt accounts. It was creepily accurate - or at least that’s what I heard from a friend who has an alt :)
Could this tool be repurposed for that? Presumably the “map” rendered in each user’s avatar could be encoded as a vector and then compared to that of another user.
EDIT: Wait, I just realized it already does this… (or at least I think so - it’s not immediately obvious if “Explore More Users” is ranked by similarity.)
> Could this tool be repurposed for that? Presumably the “map” rendered in each user’s avatar could be encoded as a vector and then compared to that of another user.
It's probably less likely to work because people often use alts to participate in discussions that they wouldn't want to associate the primary identity with. Whether that be discussions about their employer, about politics, or something else, the subjects an alt particulates in will likely be different.
Style works because few people are capable of fundamentally altering their style even if they tried.
My final year project in uni was a tool to detect plagiarism by analyzing whether parts of an assignment had been outsourced to anyone else (basically did one author write the document on their own). Or if two projects by the same student were actually written by one person. For what was at the time an extremely naive implementation of various stylometric methods it worked surprisingly well. Even I was shocked when I managed to demonstrate the tool working. People should read about this branch of studies if they care at all about anonymity online.
I would be amused to see whether or not you could simply mask all of your comments by filtering them through a commodity LLM or not. I would assume so.
Ironically, it doesn’t use embeddings even though at the time and since I’ve basically exclusively used embeddings.
Main issue with embeddings is they don’t change well overtime (you can adjust, but not as easily as other methods). Language is filled with industry specific acronyms and models are generally not great at adapting to changing phrasing and acronyms.
Anyway, eventually decided to put some more pieces together and built it into a company - https://ipcopilot.ai
I chose my username from the narrator's alter ego in Zen and the Art of Motorcycle Maintenance. In reference to the analytical knife:
"Phædrus was a master with this knife, and used it with dexterity and a sense of power. With a single stroke of analytic thought he split
the whole world into parts of his own choosing, split the parts and split the fragments of the parts, finer and finer and finer until he had
reduced it to what he wanted it to be. Even the special use of the terms "classic" and "romantic" are examples of his knifemanship."
In a bit of nominative determinism, or perhaps just having chosen the name because I know myself (or maybe I just over use these words), my keywords include: "part, system, level, language, article, object," etc.
I found it a bit challenging to actually drill down into my own username, but it doesn't seem to offer much other than throwing a lot of dots all over the map. I'm trying to understand what the overall clusters might be, but most of them are just android/apple/google?
Readit News
Also, I feel like this tool selects for active commenters, not for knowledgeable experts. Not to mention throwaway accounts.
Still a cool project.
I have absolutely no idea who I've had an argument with on HN ever. I'm sure I've had a few.
Obviously this applies to a negligible percentage of total commenters, but as I only comment on certain topics I’m more likely to encounter friends with the same interests/experiences.
But at the same time I think the OP has made an awesome project! Well done
Not the negative space, but a list of untolerated content.
And sure, ideally you wouldn't need such a list of trusted experts but just focus on content. There even was a time when this worked - you could just type "what is the best database to use" into a search engine, and get a helpful result. Not anymore. On HN it may still be better than elsewhere, but ultimately it's a similar issue.
Currently, HN is the only place on the internet I am willing to interact with others _because_ it lacks the "social layer" you are recommending.
The focus on user comments that are thoughtful, relevant, and respectful _is_ the social connection I value.
If you want to add value and not bloody public spectacle rank comments instead of users.
I have a bunch of low quality posts here when idiots piss me off, but also share world first research and breakthroughs I've been involved with the rare time the counter party is worth talking to.
Dead Comment
If you zoom and pan the red cursor over that map location/text, your source comment will appear underneath the map.
Kind of like how when the pandemic started, sales of guitars and home gym equipment went up as people worked on at-home hobbies.
Weirdly, almost no mentions of Apple stuff (which is supremely ironic given my blog).
So you get an impression of what people discuss here, but not necessarily what they know :)
Mine has a ton of stuff:
https://hn2.wilsonl.in/user/AndrewKemendo
https://hn2.wilsonl.in/user/SketchySeaBeast
(I wonder if the embedding was generated using IDF, placing maybe too much emphasis on rare word tuples?)
Show HN: Exploring HN by mapping and analyzing 40M posts and comments for fun - https://news.ycombinator.com/item?id=40307519 - May 2024 (159 comments)
Is there an opt-out of this thing?
Could this tool be repurposed for that? Presumably the “map” rendered in each user’s avatar could be encoded as a vector and then compared to that of another user.
EDIT: Wait, I just realized it already does this… (or at least I think so - it’s not immediately obvious if “Explore More Users” is ranked by similarity.)
It's probably less likely to work because people often use alts to participate in discussions that they wouldn't want to associate the primary identity with. Whether that be discussions about their employer, about politics, or something else, the subjects an alt particulates in will likely be different.
Style works because few people are capable of fundamentally altering their style even if they tried.
Ironically, it doesn’t use embeddings even though at the time and since I’ve basically exclusively used embeddings.
Main issue with embeddings is they don’t change well overtime (you can adjust, but not as easily as other methods). Language is filled with industry specific acronyms and models are generally not great at adapting to changing phrasing and acronyms.
Anyway, eventually decided to put some more pieces together and built it into a company - https://ipcopilot.ai
Now we discover ideas on hacker news for our demo https://m.youtube.com/watch?v=B5ymgh-ZDiI&pp=ygUKSXAgY29waWx...
Deleted Comment
Like user123 is moderately funny, articulate, interested in typescript and web development, and rated high on conscientiousness and extraversion.
"Phædrus was a master with this knife, and used it with dexterity and a sense of power. With a single stroke of analytic thought he split the whole world into parts of his own choosing, split the parts and split the fragments of the parts, finer and finer and finer until he had reduced it to what he wanted it to be. Even the special use of the terms "classic" and "romantic" are examples of his knifemanship."
In a bit of nominative determinism, or perhaps just having chosen the name because I know myself (or maybe I just over use these words), my keywords include: "part, system, level, language, article, object," etc.
https://hn2.wilsonl.in/user/mmastrac