Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Loss of nearly a full decade of information from early days of Chinese internet

chinamediaproject.org

Post reply on HN
Print view View original

81–90 of 105 posts

Re: Loss of nearly a full decade of information from early days of Chinese internet

#81

Post by WalterBright »

Natsu wrote: post #74
WalterBright wrote: post #72

Earlier quoted context omitted.

I've love to have a single letter from some of my ancestors.

I have one, actually, from my grandpa's generation. He told another family member about his time growing up in the early 1900s, riding trolleys and eating Walnettos (a strange Walnut-based candy bar). Then the Spanish Flu came around and the eldest sister just died at the breakfast table one day. Later, the family rallied together to care for each other after his father lost his job due to automation. He moved on to…

I have some letters from 3 of my grandparents, but beyond that, it's pretty much nothing.

Re: Loss of nearly a full decade of information from early days of Chinese internet

#82

Post by gwern »

Wow - this explains a lot about why Chinese LLMs and AI struggle so much to get data, despite the supposed near-infinite databases: a lot of it is just gone, far more than anyone had ever proposed might be the case, because of neglect, censorship, walled gardens/apps killing everything trapped within, and chilling effects.

You can't scrape what is no longer there, or was never written to begin with...

Re: Loss of nearly a full decade of information from early days of Chinese internet

#83

Post by smcin »

Arnavion wrote: post #12
lelandfe wrote: post #7

> Within the selected date range of “May 22, 1998 to May 22, 2005” on Baidu, there is just one positive result for “Jack Ma” (dated May 22, 2024). [..] Click on the result and you’ll find it is an article posted in 2021 US Google: About 2,580,000 results A pretty remarkable scrubbing of history.

https://www.google.com/search?q=jack+ma&tbs=cdr%3A1%2Ccd_min... First result for me is https://www.scb.co.th/en/personal-banking/stories/business-m... which Google thinks is from 2003-03-15, except it mentions COVID-19 so it obviously isn't. Second result is https://www.instagram.com/jack_overpower/feed/ which Google thinks is from 2001-01-02, except Instagram didn't exist at that time. It might have pictures from 20…

> First result [scb.co.th] ... Google thinks is from 2003-03-15, except it mentions COVID-19 so it obviously isn't.

Interesting catch, seems Google grossly mistagged its date. IA confirms it was actually published 2021-09-06 [0], but that isn't tagged or referenced anywhere in the article text or HTML. I'm assuming Google misinferred the date as "2003-03-15" because the first two paragraphs talk about the SARS crisis, which was declared by the WHO around 2003-03.

> I don't doubt that some of those results are from 1998 to 2005, but the millions of results number specifically is meaningless.

Yes, seems there's not much QC on Google's date-inferencing of "old" articles. Hence the date-range is hit-and-miss, and search hit counts (which Google is eliminating anyway). I mean if anyone wanted to QC it, just search "old" internet for telltale terms like "COVID", "Nicki Minaj", "President Zelenskyy" etc. that should hardly generate any hits.

[0]: https://web.archive.org/web/20210601000000*/https://www.scb....

Re: Loss of nearly a full decade of information from early days of Chinese internet

#84

Post by smcin »

boomboomsubban wrote: post #27
Arnavion wrote: post #12

Earlier quoted context omitted.

https://www.google.com/search?q=jack+ma&tbs=cdr%3A1%2Ccd_min... First result for me is https://www.scb.co.th/en/personal-banking/stories/business-m... which Google thinks is from 2003-03-15, except it mentions COVID-19 so it obviously isn't. Second result is https://www.instagram.com/jack_overpower/feed/ which Google thinks is from 2001-01-02, except Instagram didn't exist at that time. It might have pictures from 20…

The "custom range" feature simultaneously feels broken, gamed by spammers, and intentionally being scrubbed. I'm surprised they haven't completely removed it yet.

In general I suppose, but per my comment above in this particular case of the scb.co.th article (which mentions SARS crisis), the article was actually published 2021 not 2003, there was no gaming going on, simply Google's data-inference code got it totally wrong on the Siam Commercial Bank article.

I don't want to see Google remove the search-by-date-range feature, it has tons of 100% serious legitimate uses (quote attribution, journalistic, historical, also debunking internet rumors and fresh reposts of old fake virals), but Google could estimate the errorbars on date-ranges and hitcounts, provide a disclaimer and feedback box to encourage users to flag/retag gross errors like this.

If anyone dug deeper into why the date inferencing is now getting broken, I'd speculate they find Google is nowadays getting reciprocally confused by site publishers and advertising networks changing or removing items which contain date; but that's in turn presumably for Google changes which downrank or uprank legit older content.

(I can't find the recent HN article here by an SEO expert with a bullet-point list advising website maintainers to remove all date information, among other things)

Re: Loss of nearly a full decade of information from early days of Chinese internet

#85

Post by guffins »

abecedarius wrote: post #65
lostemptations5 wrote: post #33

Intentionally or not this is -- exactly -- what 1984 is all about: changing our perception of history by rewriting or erasing previous writings. Unfortunately alot of it from the article seems typical: blogs going off line as bloggers move to new technolgoies, social media companies going defunct or just not keeping old content. Alot of these things can happen in the west. Remember these books you could read? "The Fe…

Tangential, but what is "The Feynman Letters" here? I know of a book of some of his letters, but not about censorship/loss thereof.

Perhaps referring to this? I’m not entirely sure. https://ahf.nuclearmuseum.org/ahf/history/mail-censorship/

Re: Loss of nearly a full decade of information from early days of Chinese internet

#86

Post by pennybanks »

lostemptations5 wrote: post #33

Intentionally or not this is -- exactly -- what 1984 is all about: changing our perception of history by rewriting or erasing previous writings. Unfortunately alot of it from the article seems typical: blogs going off line as bloggers move to new technolgoies, social media companies going defunct or just not keeping old content. Alot of these things can happen in the west. Remember these books you could read? "The Fe…

one link of jack ma between that time period on badu? bro that cant be no accident. if the chinese govt didnt do this id be more surprised i mean there already censoring most of the internet

Re: Loss of nearly a full decade of information from early days of Chinese internet

#87

Post by renonce »

Why expect search engines to return historical data accurately? Modern search engines have a lot of tasks like combating CEO and returning up-to-date data, and they have no incentive to preserve history as old as 2005 as it’s very likely that any page from that date has been superseded by more relevant articles. The task of preserving history should be delegated to archive.org and search engines are just not well positioned.

Re: Loss of nearly a full decade of information from early days of Chinese internet

#88

Post by yorwba »

demosthanos wrote: post #75
yorwba wrote: post #73

Earlier quoted context omitted.

The original post was about natural entropy on the internet. Websites from 2005 that have disappeared or been redesigned so that you can't find their old content anymore, and the uselessness of search engines, domestic or foreign, for date range queries reaching that far back into the past. Even on the Internet Archive, the earliest working snapshot of Baidu Tieba is from 2006. You may think that it's impossible for…

> The original post was about natural entropy on the internet. The post by He Jiayan was, but that post was taken down for violating regulations. TFA is largely about the censorship angle which He Jiayan specifically avoided talking about (not that it helped him). > If the disappearance of old websites were entirely deliberate, you'd also need to explain why the West is in on it. Name one figure who was prominent in…

> Name one figure who was prominent in between 1995-2005 who you can't find any content about from that era when using Google's date filters. A single figure.

The original post listed multiple people famous in China at that time (including Taiwanese celebrities) where even Bing and Google didn't get them old enough results. Sure, they return results that supposedly match the date filter, but if you actually read them, it becomes clear that Google got the publication date wrong, because much later events are mentioned in the text. Or e.g. a YouTube video from 2004, before YouTube even existed. (Actually uploaded in 2013.)

Re: Loss of nearly a full decade of information from early days of Chinese internet

#89

Post by rfoo »

gwern wrote: post #82

Wow - this explains a lot about why Chinese LLMs and AI struggle so much to get data, despite the supposed near-infinite databases: a lot of it is just gone , far more than anyone had ever proposed might be the case, because of neglect, censorship, walled gardens/apps killing everything trapped within, and chilling effects. You can't scrape what is no longer there, or was never written to begin with...

> supposed near-infinite databases: a lot of it is just gone

Pretty sure megacorps still have their near-infinite databases around, in walled garden as always, but that's another topic. Censored contents are almost certainly soft-deleted.

The problem is more about independent-ish sites, for example Tianya [1]. Think ancient BBS. Or newsgroups. If nearly all archive of these are lost it would indeed be very sad.

[1] https://en.wikipedia.org/wiki/Tianya_Club#cite_note-4

Post reply on HN