zlacker

[parent] [thread] 58 comments
1. babbli+(OP)[view] [source] 2025-10-28 15:26:49
Especially relevant today with the release of grokipedia
replies(7): >>bright+T3 >>mapont+d4 >>embedd+Z4 >>dreamc+75 >>jalape+F7 >>z7+8b >>george+Hc
2. bright+T3[view] [source] 2025-10-28 15:45:10
>>babbli+(OP)
It’s so good. I spent 2 hours reading articles on there last night and the consistency was excellent, although a little verbose at times.
replies(5): >>GaryBl+85 >>djeast+I5 >>radley+P6 >>coolel+n7 >>nojone+Nc
3. mapont+d4[view] [source] 2025-10-28 15:46:33
>>babbli+(OP)
oh...no.

I thought this was a joke, but I googled it and it's not.

4. embedd+Z4[view] [source] 2025-10-28 15:49:55
>>babbli+(OP)
> The articles in Grokipedia indicate that they have undergone fact-checking by the Grok model.[3] Visitors to Grokipedia cannot make edits, though they can suggest edits via a pop-up form for reporting wrong information.[5] Musk positioned Grokipedia as an alternative to Wikipedia that would "purge out the propaganda" in the latter.[1] Articles have been described as manipulated to promote right-wing perspectives and Elon Musk's views,[4][7] medical misinformation,[7] and for removing content disfavored by Musk.[8][9] - https://en.wikipedia.org/wiki/Grokipedia

Seems like a great platform, here's to hoping it costs a lot to run and doesn't influence too many humans to drink bleach.

5. dreamc+75[view] [source] 2025-10-28 15:50:27
>>babbli+(OP)
https://en.wikipedia.org/wiki/Grokipedia
replies(1): >>HPsqua+ma
◧◩
6. GaryBl+85[view] [source] [discussion] 2025-10-28 15:50:35
>>bright+T3
If it works, it works.
replies(2): >>IAmBro+q5 >>neaden+a8
◧◩◪
7. IAmBro+q5[view] [source] [discussion] 2025-10-28 15:51:55
>>GaryBl+85
Unsupervised Source of Truth(tm), what could possibly go wrong?
replies(2): >>sunaoo+ea >>GaryBl+Xh
◧◩
8. djeast+I5[view] [source] [discussion] 2025-10-28 15:53:24
>>bright+T3
Nevermind the Grok-ness of it, I can't seriously believe a thinking human being would spend 2 hours knowingly reading something written by AI.
replies(3): >>delich+b9 >>add-su+z9 >>bright+oA1
◧◩
9. radley+P6[view] [source] [discussion] 2025-10-28 15:57:07
>>bright+T3
Of course, it cribbed the best!
◧◩
10. coolel+n7[view] [source] [discussion] 2025-10-28 15:58:57
>>bright+T3
It took me less than a minute on the site to run into factually wrong information, broken citations, etc. Cannot imagine rotting my brain with knowingly bad information for over 2 hours
replies(2): >>antony+58 >>vitro+ml
11. jalape+F7[view] [source] 2025-10-28 15:59:54
>>babbli+(OP)
Someone please enlighten me: what is the point of an AI-generated Wikipedia when we're all now using (Wikipedia-trained) AIs directly instead of Wikipedias?
replies(9): >>milton+l9 >>j2kun+y9 >>Spivak+W9 >>this_u+ld >>etchal+Ed >>moritz+Oy >>rsynno+iI >>Yizahi+9o1 >>breppp+Mi2
◧◩◪
12. antony+58[view] [source] [discussion] 2025-10-28 16:01:19
>>coolel+n7
I’ve run into the exact same set of issues on Wikipedia.
replies(2): >>gdulli+0c >>gamerd+ig
◧◩◪
13. neaden+a8[view] [source] [discussion] 2025-10-28 16:01:36
>>GaryBl+85
Yeah, but it doesn't work. It's full of inaccuracies.
◧◩◪
14. delich+b9[view] [source] [discussion] 2025-10-28 16:07:14
>>djeast+I5
I'm unsurprised that a human being would glibly dismiss the utility of the most powerful new form of knowledge representation since the written word, since we are all deeply in the grip of motivated reasoning.
replies(2): >>PaulDa+we >>djeast+Jp
◧◩
15. milton+l9[view] [source] [discussion] 2025-10-28 16:07:44
>>jalape+F7
From the wikipedia on grokipedia:

"Following the public launch of Grokipedia, it was criticised for publishing false information. Wired reported that "The new AI-powered Wikipedia competitor falsely claims that pornography worsened the AIDS epidemic and that social media may be fueling a rise in transgender people."

So, it's a way of Musk using AI to propagandize on a large scale.

replies(2): >>reaper+Ok >>benphl+Ky
◧◩
16. j2kun+y9[view] [source] [discussion] 2025-10-28 16:08:19
>>jalape+F7
Perhaps to muddy future corpora
◧◩◪
17. add-su+z9[view] [source] [discussion] 2025-10-28 16:08:26
>>djeast+I5
It's for the intersection of people who want LLM summarization and people who want an assurance of confirmation of bias explicitly built in. It's not for thinking people.
replies(1): >>etchal+3e
◧◩
18. Spivak+W9[view] [source] [discussion] 2025-10-28 16:10:43
>>jalape+F7
It's Conservapedia, authored by AI, and exists to present the world as seen through the eyes of Mr. Musk. The hope is that through AI it can be comprehensive enough to be useful and if enough people adopt it you can quietly put your thumb on the scale to make truth what you say it is.
◧◩◪◨
19. sunaoo+ea[view] [source] [discussion] 2025-10-28 16:12:03
>>IAmBro+q5
Just like Wikipedia?
replies(2): >>neaden+Bd >>IAmBro+Kn
◧◩
20. HPsqua+ma[view] [source] [discussion] 2025-10-28 16:12:36
>>dreamc+75
That references another article "Views of Elon Musk" ... Quite an unusual format, "views of [public figure]". I don't remember seeing anything quite like it on Wikipedia.

https://en.wikipedia.org/wiki/Views_of_Elon_Musk

replies(1): >>embedd+wl
21. z7+8b[view] [source] 2025-10-28 16:16:06
>>babbli+(OP)
Here's the Grokipedia submission (currently censored / flagged):

>>45726459

replies(1): >>tim333+JC
◧◩◪◨
22. gdulli+0c[view] [source] [discussion] 2025-10-28 16:19:31
>>antony+58
Suggesting that people being able to make mistakes means that there's no qualitative and quantitative difference in how AI makes mistakes is either disingenuous or stupid. I don't know which place you're coming from or what kind of gotcha you think you pulled, but it doesn't create a strong argument either way.
23. george+Hc[view] [source] 2025-10-28 16:23:00
>>babbli+(OP)
I would really, really like to see the prompts used for this next to each article.

Since "The Algorithm" at Twitter was supposed to be open sourced, surely that wouldn't be controversial.

And I genuinely do find it absolutely fascinating and somewhat shocking how LLMs can follow such long and complex prompts and respond so well.

◧◩
24. nojone+Nc[view] [source] [discussion] 2025-10-28 16:23:18
>>bright+T3
I checked it out based on this comment. It's funny how in some ways it feels like a lazy student-assignment copied from Wikipedia: the subheadings and the structure are exactly the same as the Wikipedia article on the topic, and sometimes it even leaves in the citation numbers as normal text like a careless copy paste.

However, it also seemed less eurocentric, mentioning non-Greek non-Roman side of origins of fields where relevant, when the corresponding Wikipedia article doesn't. Wikipedia is generally pretty bad at this, but I had expected "Grokipedia" to be worse, not better in this regard!

replies(1): >>ZeroGr+jJ
◧◩
25. this_u+ld[view] [source] [discussion] 2025-10-28 16:25:23
>>jalape+F7
So that new AI models can then be trained on the AI-generated Wikipedia.
◧◩◪◨⬒
26. neaden+Bd[view] [source] [discussion] 2025-10-28 16:26:29
>>sunaoo+ea
With Wikipedia there is the talk page which will alert you to controversies about topics, as well as checking the citations. While Grokopedia has "citations" when I checked many of them didn't actually have anything to do with what they were supposed to be citing.
◧◩
27. etchal+Ed[view] [source] [discussion] 2025-10-28 16:26:42
>>jalape+F7
Grok trained on Wikipedia so it could generate a version of Wikipedia that reflected Musk's views that could then be used to train future versions of Grok.
replies(1): >>downri+ly
◧◩◪◨
28. etchal+3e[view] [source] [discussion] 2025-10-28 16:28:39
>>add-su+z9
"A machine which simulates thought for people who don't want to think" is an adequate summation of LLM-generated text.
◧◩◪◨
29. PaulDa+we[view] [source] [discussion] 2025-10-28 16:30:55
>>delich+b9
> the most powerful new form of knowledge representation since the written word

1. the LLM model is a representation of language, not knowledge. The two may be highly correlated, but they are probably not coterminant and they are certainly not equivalent.

2. the final "product" is still the written word

3. whether LLM's are or are not the most powerful new form of knowledge representation or not, their output is so consistently inconsistent in its accuracy that it makes that power difficult to utilize, at best.

◧◩◪◨
30. gamerd+ig[view] [source] [discussion] 2025-10-28 16:38:39
>>antony+58
Did you fix them?
replies(1): >>hagbar+a63
◧◩◪◨
31. GaryBl+Xh[view] [source] [discussion] 2025-10-28 16:44:58
>>IAmBro+q5
I don't understand what supervision you want. Worried about inaccuracies? Double check and use it in conjunction with other sources.
replies(1): >>IAmBro+1p
◧◩◪
32. reaper+Ok[view] [source] [discussion] 2025-10-28 16:55:38
>>milton+l9
social media may be fueling a rise in transgender people

So one part of the Musk empire is fueling a thing that another part of the Musk empire doesn't like.

Seems like the problem is in one hand, and the solution is in the other.

replies(1): >>embedd+8m
◧◩◪
33. vitro+ml[view] [source] [discussion] 2025-10-28 16:57:45
>>coolel+n7
Maybe he meant that the information was consistently incorrect, which was entertaining..
◧◩◪
34. embedd+wl[view] [source] [discussion] 2025-10-28 16:58:52
>>HPsqua+ma
Seems there are a couple of pages like that, if you enter "views of" without hitting enter in the Wikipedia search bar, you get some suggestions. Seems there is similar pages for Kanye West, Richard Dawkins and some more. Many of the pages are redirects back to the main page of the person though, so seems they're maybe disappearing or exclusively used for people who are very outspoken about lots of different things.
replies(1): >>exaspe+4D
◧◩◪◨
35. embedd+8m[view] [source] [discussion] 2025-10-28 17:00:50
>>reaper+Ok
> Seems like the problem is in one hand, and the solution is in the other.

I think that's a common thread with what Musk does. On one hand, his companies rely on money from the US government, then with the other he's helping firing a lot of people in the government supposedly to save money.

On one hand, he's trying to run for AGI and manage a LLM company that use vast amount of resources. On the other hand he's trying to sell electric vehicles because vast amount of resources are being used.

I guess it kind of makes sense in some way, but also he could probably better help those efforts by just stopping doing the other thing, but that probably conflicts with his other more important goals.

replies(1): >>hedora+Xx
◧◩◪◨⬒
36. IAmBro+Kn[view] [source] [discussion] 2025-10-28 17:06:42
>>sunaoo+ea
Wikipedia has 7 billion potential editors. Grokapedia explicitly says we can't edit it.

So, absolutely the opposite thing.

◧◩◪◨⬒
37. IAmBro+1p[view] [source] [discussion] 2025-10-28 17:12:20
>>GaryBl+Xh
I want some supervision. Why is that hard for you to grasp?
replies(1): >>GaryBl+Ux
◧◩◪◨
38. djeast+Jp[view] [source] [discussion] 2025-10-28 17:14:53
>>delich+b9
No one is being glib here, this is a serious concern. Think about it, please. A human being choosing to spend hours of their time reading something produced by something that is an amorphous, unanswerable, unaccountable agglomeration of weights formed not by a human's lived experience, but by a for-profit company's selection of inputs and tuning. It's completely dystopian.
◧◩◪◨⬒⬓
39. GaryBl+Ux[view] [source] [discussion] 2025-10-28 17:48:33
>>IAmBro+1p
There is supervision, just not community-based.
◧◩◪◨⬒
40. hedora+Xx[view] [source] [discussion] 2025-10-28 17:48:47
>>embedd+8m
His behavior only made sense to me after I realized his dynastic wealth came from apartheid-era mines.
◧◩◪
41. downri+ly[view] [source] [discussion] 2025-10-28 17:50:28
>>etchal+Ed
newspeak
◧◩◪
42. benphl+Ky[view] [source] [discussion] 2025-10-28 17:51:23
>>milton+l9
Are these false claims? From Grokipedia:

> This marked the onset of what would become a devastating crisis disproportionately affecting gay male communities, where behaviors idealized in pornography—such as unprotected receptive anal intercourse and multiple anonymous partners—aligned directly with primary transmission routes, leading to rapid seroconversion rates.

This sounds plausible. Is it factually incorrect?

replies(2): >>slater+7A >>stubis+fc1
◧◩
43. moritz+Oy[view] [source] [discussion] 2025-10-28 17:51:31
>>jalape+F7
It makes sense to people who don't know what an encyclopedia or an AI chatbot are, respectively.

It's ideal to poison the web with arbitrarily distorted texts that are a mix of facts and lies, and will be picked up by others, from AI to Zoomer school essay.

There is no point except for manipulation. Right now, you have to be pretty inept to think that a language AI could contribute anything valuable to an encyclopedia.

But maybe, this will change, the group of people who consider Chatbot output as insightful about the real world seems to be growing.

◧◩◪◨
44. slater+7A[view] [source] [discussion] 2025-10-28 17:57:10
>>benphl+Ky
Why not have the decency to respond with your real HN account, instead of making a new account just to post tedious sealioning?

[this is the point at which you swear up, down and sideways that you've never ever in your whole life had a HN account, this is your first account ever, how dare i, etc. etc. etc.]

Edit to answer your totally-asked-in-good-faith question: Causation != correlation.

replies(1): >>benphl+pC
◧◩◪◨⬒
45. benphl+pC[view] [source] [discussion] 2025-10-28 18:04:21
>>slater+7A
It is a genuine question. If you don't want to engage in intellectual curiosity on this topic, please do not respond at all. Thank you.
◧◩
46. tim333+JC[view] [source] [discussion] 2025-10-28 18:06:03
>>z7+8b
It seems a bit unfairly flagged. After all the first AI written encyclopedia seems a new and interesting thing.
◧◩◪◨
47. exaspe+4D[view] [source] [discussion] 2025-10-28 18:07:57
>>embedd+wl
They are largely used to put the Talk: flamewar in a box, I think, so they can separate out the mundane fact edits from the rest.
replies(1): >>embedd+H41
◧◩
48. rsynno+iI[view] [source] [discussion] 2025-10-28 18:32:45
>>jalape+F7
Safe space for Musk. He's been upset about Wikipedia for years (I suspect because it refuses to buy into his headcanon that he founded Tesla.)
replies(1): >>dreamc+D31
◧◩◪
49. ZeroGr+jJ[view] [source] [discussion] 2025-10-28 18:38:03
>>nojone+Nc
If I was rewriting Wikipedia pages with an LLM I'd maybe use all the different languages' Wikipedias as input.
◧◩◪
50. dreamc+D31[view] [source] [discussion] 2025-10-28 19:53:53
>>rsynno+iI
His own entry in grokpedia [0] says two conflicting things:

"Musk founded SpaceX in 2002 as CEO and chief engineer, Tesla in 2003 where he became CEO in 2008..."

and later on the same page,

"...the company [Tesla] had been founded in 2003 by engineers Martin Eberhard and Marc Tarpenning with a focus on high-performance EVs."

Grok can't seem to keep its story straight.

[0] https://grokipedia.com/page/Elon_Musk

◧◩◪◨⬒
51. embedd+H41[view] [source] [discussion] 2025-10-28 19:57:57
>>exaspe+4D
That's actually kind of neat, at least for us that don't care much of the political views of celebrities.
replies(1): >>exaspe+hp1
◧◩◪◨
52. stubis+fc1[view] [source] [discussion] 2025-10-28 20:31:51
>>benphl+Ky
Yes, it is factually incorrect.

Now you have two unsubstantiated opinions contradicting each other.

◧◩
53. Yizahi+9o1[view] [source] [discussion] 2025-10-28 21:28:28
>>jalape+F7
It reduces a barrier to entry, so less knowledgeable people can access the same information without inputting a prompt and then corrections. Also a person using an LLM directly may accidentally produce a progressive/liberal result, which is not good. So while for now it seems Elonopedia is mostly automated, in the future I foresee that young energetic party members will vet the most popular articles, to follow the party line.
◧◩◪◨⬒⬓
54. exaspe+hp1[view] [source] [discussion] 2025-10-28 21:34:55
>>embedd+H41
I think it makes sense for some politicians, too —- especially politicians who have an extensive life story otherwise.

Not least because it bumps the topic up one heading level, as it were, which means more possible uses of mediawiki formatting to break it up than if it were a section of another page.

◧◩◪
55. bright+oA1[view] [source] [discussion] 2025-10-28 22:40:10
>>djeast+I5
I decided to read through a subject I already knew a lot about.
◧◩
56. breppp+Mi2[view] [source] [discussion] 2025-10-29 05:30:41
>>jalape+F7
Generally the UX could be an advantage, the hypertext format of wikipedia and length is nicer when you want to go on a random knowledge walk compared to LLMs.

Regarding the Wikipedia trained Ouroboros models, you can argue that the Wikipedia training is mostly there to learn to summarize sources and translate, and once you have the original sources the LLM might do a better job than humans

◧◩◪◨⬒
57. hagbar+a63[view] [source] [discussion] 2025-10-29 12:36:50
>>gamerd+ig
This is sometimes hard when the editors keep on reversing edits which attempt to fix those errors. It will be interesting to see how Grokipedia - a bad name, surely they can come up with something better - deals with this.

I often come across out-of-place or clearly ideologically driven content on Wikipedia and normally just leave this alone - I have better things to do with my limited time than to fight edit wars with activist editors. Having said that I did a number of experiments some 5 years ago with editing Wikipedia where I removed clearly ideologically driven sections out of articles where those sections really had no place. One of these experiments consisted of removing sections about ´queer politics and queer viewpoints' from articles about popular cartoon characters. These sections - often spanning several paragraphs - were inserted relatively recently into the articles and were nothing more than attempts to use those articles to push a 'queer' viewpoint on the subject matter and as such not relevant for a general purpose encyclopedia. I commented my edits with a reference to the NPOV rules. My edits were reversed without comment. I reversed the reversion with the remark to either explain the reversion of leave the edits in place and was reversed again, no comments. I reversed again with an invitation to discuss the edits on the Talk pages which was not accepted while my edits were reversed again. This continued for a while with different editors reversing my edits and accusations of vandalism. Looking through the 'contribs' section for the users responsible for adding the irrelevant content showed they were doing this to hundreds of articles. I just checked and noticed the same individuals are still actively adding their 'queer perspectives' to articles where such perspectives are not relevant for a general-purpose encyclopedia.

replies(2): >>bright+Np3 >>gamerd+X77
◧◩◪◨⬒⬓
58. bright+Np3[view] [source] [discussion] 2025-10-29 14:24:12
>>hagbar+a63
Correct. That's the main reason I dove into reading subjects I was already knowledgeable of to see how it did.
◧◩◪◨⬒⬓
59. gamerd+X77[view] [source] [discussion] 2025-10-30 16:02:49
>>hagbar+a63
Do you happen to remember any of the articles where you performed this experiment? I ask because specifically around 5 years ago, I know there were a number of cartoons where the creators intentionally wrote characters with queer representation in mind (She-Ra is the first to come to mind). So, if the sections you were removing had been properly cited and relevant to the actual series, then the removal for being "nothing more than attempts to use those articles to push a 'queer' viewpoint on the subject matter" probably did not represent a neutral viewpoint.

Of course, this depends on you opening up your research to some peer review.

[go to top]