costco

25 ms·

361.
by costco 4y ago
Apologies for the downtime. Something crashed while I was asleep, should be working now. Not really sure how because the log indicates that uwsgi "gracefully exited," but I'm looking into it.
362.
by costco 4y ago
Are you serious? That is way cooler than the alt detector site. Depending on how much extra work you're willing to do I think TAILS is actually looking for a program to do this because they had some problems with Anonymouth.
363.
by costco 4y ago
Not sure if GPT3 at least if prompted right would have clearly identifiable style. Could probably detect converted call centers in Russia or Cambodia where 50 employees post on 10000 accounts though.
364.
by costco 4y ago
> Edit: by bigrams and trigrams, I mean reducing word to their parts of speech labels and using THOSE as word tokens. You’ll find that native English speakers have higher weights on some phrase construction patterns than, say, folks from
365.
by costco 4y ago
Sorry for any misunderstanding, read https://news.ycombinator.com/item?id=33756725
366.
by costco 4y ago
It seems the accuracy for nonnative speakers is not nearly as good as it is for native speakers. The algorithm could definitely use some work.
367.
by costco 4y ago
That's the lowest I've seen yet. You must write uniquely :)
368.
by costco 4y ago
But a 0.8 or 0.9 match and something like Tor usage could be enough to justify a warrant. That's why I'm not sure I want to open source the code because I don't want to normalize this.
369.
by costco 4y ago
With all due respect: https://news.ycombinator.com/item?id=16785542 https://twitter.com/paulg/status/1362369484036653058 I think you are very likely to be wrong.
370.
by costco 4y ago
Really cool idea. I'd need to upgrade the VPS though so all the vectors would fit in memory but it probably wouldn't be too hard (right now I'm just storing a map of username string -> array of 20 username strings because
371.
by costco 4y ago
There are factors that make me think it is more likely than not (just scrolled through the comment history, don't feel like linking everything) that he is pg. - Is bolded on pg's page - Mentions yoga - Talks about Lisp often - Tal
372.
by costco 4y ago
I was aware there was a HN dataset on BigQuery but I had never used a library to work with it before and when I played around on the website the posts I got were all from 2015 at the latest. It probably would've made my work easier bu
373.
by costco 4y ago
Exactly. HN emphasizes long-form posts much more than other forums which makes the commenters here very susceptible to this kind of analysis. Plus you can fit every single HN comment in RAM on a mid tier gaming laptop so it's even ea
374.
by costco 4y ago
All words have the same weighting. I don't ignore stop words, in fact most of the ngrams I use are compromised almost entirely of stop words. Maybe it'd be more effective if I ignored them.
375.
by costco 4y ago
Quoted replies shouldn't be included unless there's a bug on my end. Submission text is not included though I probably should have.
376.
by costco 4y ago
> On the other hand, I'm nearly as correlated with peterwwillis: 0.5485 -- who has no comments and no submissions. This is due to the Firebase API not updating when users ask the admins to move their comments to another account.
377.
by costco 4y ago
I think people are sort of confused at what this tool is supposed to be which I will concede is partially my fault. The results of this tool are by themselves not indicative of having an alternative account. It generates the 20 most simil
378.
by costco 4y ago
That post was actually what motivated me to make this. I'm on your email list :)
379.
by costco 4y ago
False positives become an increasingly difficult problem the more and more potential authors you introduce. If I had wrote a fancier model it probably wouldn't be as much of a problem but what can you do.
380.
by costco 4y ago
I could have used a part of speech tagger, looked at time of day a user posts, capitalization, spelling errors, etc. From what I understand the state of the art is lightyears ahead of this, there are even companies with actual linguists w
381.
by costco 4y ago
Huh... seems there are some inconsistencies between what's presented on news.ycombinator.com and the Firebase API. Glad it matches for you though :)
382.
by costco 4y ago
There is no threshold. This site does not make any call as to whether a user is an alt or not. It just gives the users with the most similar word choice and from there it is up to you to decide (is there a very specific detail that both a
383.
by costco 4y ago
> Can you clarify what this means and why it would result in a ban? I have seen dang respond to users multiple times asking them to stop making new accounts especially but not always if it's to avoid rate limiting. I don't kno
384.
by costco 4y ago
Say you see user2 listed in bold on user1's page. That means that user1 is also in user2's top 20 users. In my experience it is often an indicator of a good match (but not always).
385.
by costco 4y ago
The funny thing is that I thought of it while eating dinner last night :)
386.
by costco 4y ago
I'm not that smart - my site is basically just doing some calculations on word frequencies. You can read https://academic.oup.com/dsh/article-abstract/17/3/267/92927... and https://
387.
by costco 4y ago
I turned off nginx logging if that makes you feel any better. Of course there's no way for you to verify that because I'm just a random guy on the internet but I will tell you that I am a civic minded citizen who is concerned abo
388.
by costco 4y ago
> motrose is almost definitely pg. Someone who talks about ancient history, Occam’s razor, VCs and startups, uses the phrase “YC cos” (relatively uncommon), etc. https://news.ycombinator.com/item?id=17112567 I had this h
389.
by costco 4y ago
I obviously don't expect you to help me but do they have at least >10000 characters written and are you varying your writing style in any way?
390.
by costco 4y ago
Current minimum is 10000 characters. In my own tests accuracy was still pretty good at 3-5000 but I instituted the 10000 minimum to reduce false positives. Yes it would, if you read the advice page on avoiding detection that is one of the

More ›