Hello everyone,
TLDR; Wikimedia is participating in the Outreachy Round 27 internship
program <https://www.mediawiki.org/wiki/Outreachy/Round_27 > [1].
Outreachy's goal is to support people from groups underrepresented in the
technology industry. Interns will work remotely with mentors from our
community. We are seeking mentors to propose projects that Outreachy
interns can work on during their internship. If you have some ideas for coding
or non-coding (design, documentation, translation, outreach, research)
projects, share them by Sept. 29, 2023 at 4 pm UTC here as a subtask of
this parent task: <https://phabricator.wikimedia.org/T343871 > [2]
Program Timeline
As a mentor, you engage potential candidates in the application period
between October–November (winter round) and help them make small
contributions to your project. You work more closely with the accepted
candidates during the internship period between December–March (winter
round).
Important dates are:
-
Aug. 22, 2023 at 4pm UTC - Live Q&A for Outreachy mentors
<https://www.youtube.com/@outreachyinternships>
-
September 29, 2023 at 4pm UTC - Project submission deadline
<https://www.outreachy.org/communities/cfp/wikimedia/>
Guidelines for Crafting Project Proposals
* Follow this task description template when you propose a project in
Phabricator: <
https://phabricator.wikimedia.org/tag/outreach-programs-projects> [3]. You
can also use this workboard to pick an idea if you don't have one already.
Add #Outreachy (Round 27) tag.
* Project should require an experienced developer ~15 days and a newcomer
~3 months to complete.
* Each project should have at least two mentors, including one with a
technical background.
* Ideally, the project has no tight deadlines, a moderate learning curve,
and fewer dependencies on Wikimedia's core infrastructure. Projects
addressing the needs of a language community are most welcome.
Learn more about the roles and responsibilities of mentors on
MediaWiki.org: <https://www.mediawiki.org/wiki/Outreachy/Mentors> [4][5]
Cheers,
Onyinye & Sheila (Wikimedia Org Admins for Outreachy Round 27)
[1] https://www.mediawiki.org/wiki/Outreachy/Round_27
[2] https://phabricator.wikimedia.org/T343871
[3] https://phabricator.wikimedia.org/tag/outreach-programs-projects/
[4] https://www.mediawiki.org/wiki/Outreachy/Mentors
[ 5] https://www.outreachy.org/mentor/mentor-faq
Yesterday they had less than a hundred edits from just 6 individuals. The
English language Wikipedia had around 150,000 edits. Justapedia needs a
much bigger community for it to become more than a mirror.
They've switched off IP editing, one of the mistakes citizendium made, so
unless they relaunch themselves as the home of alternative medicine and
alternative facts, I agree with Galder, they are unlikely to succeed.
WSC
>
>
>
> Il giorno mar 12 set 2023 alle ore 08:15 Galder Gonzalez Larrañaga <
> galder158(a)hotmail.com> ha scritto:
>
> The logo is quite funny. According to that Information + Disinformation =
> Facts. It might be that they don't know what a Venn diagram is, or simply
> that they actually think that.
>
>
>
> Don't worry, this is just one more project that will fall into oblivion.
>
>
>
> Galder
>
>
>
>
Dear Community Members,
As we gear up for the upcoming WikiWomenCamp
<https://meta.wikimedia.org/wiki/WikiWomenCamp_2023> in India, we are
excited to announce a unique opportunity to share your incredible work and
its impact with the global community.
What We're Looking For:
We invite you to submit short (3-5 minute) videos highlighting your
projects, initiatives, or efforts that have made a significant difference
in addressing gender gaps. We know that many of you are doing remarkable
work in various corners of the world, and we want to shine a spotlight on
your achievements.
Why Short Videos:
Our camp sessions are extensive, covering a wide range of topics and
discussions. In the interest of time and with a commitment to showcasing
the amazing initiatives across our movement, we believe that short videos
are an effective way to capture and share the essence of your work.
How Your Videos Will Be Used:
The selected videos will be featured on our social media channels and
played during breaks at the camp. This provides an excellent opportunity
for you to showcase your work to a broader audience and inspire others to
get involved or collaborate with you.
Submission Guidelines:
-
Videos should be 3-5 minutes in length.
-
Please focus on the impact of your work, highlighting how it contributes
to addressing the gender gap.
-
Be creative and engage in your presentation.
-
Ensure video quality is clear and audible.
-
Narration (optional): If possible, include a brief voiceover or text
captions to provide context and explanation.
-
Language: While English is preferred for wider accessibility, videos in
other languages are also welcome. Kindly ensure that non-English videos are
accompanied by English subtitles or captions.
Submission Deadline:
To be considered for inclusion in our camp programming, please submit your
videos by Thursday, 28th September 2023. Late submissions may still be
featured later on our social media channels.
How to Submit:
To avoid any issues with file sizes, we recommend using a file transfer
service like WeTransfer <https://wetransfer.com/> or Google Drive
<https://drive.google.com/> while sharing your videos with us at
admin(a)wikiwomencamp.org. In your email, include a brief description of your
project, affiliate name / your name or username, and contact information.
We believe that your stories and experiences are powerful tools for change.
By sharing them with our community, you contribute to a collective effort
to bridge the gender gap and empower women and girls worldwide.
Thank you for being a part of the camp initiatives, and we look forward to
receiving your inspiring videos.
Warm regards,
WikiWomenCamp 2023 COT
Last month, the alternative project Justapedia was launched which is accessible through the URL https://justapedia.org. It is a complete English Wikipedia fork made sometime last year.
Unfortunately the front page showcase doesn't give it good optics because it gave the impression of trying to "whitewash" fascism. They'll have a better public opinion if they tried to fix the Holocaust distortions as revealed by Jan Grabowski and Shira Klein, and put that into the showcase.
Besides, there are alarming rumors I saw in Y Combinator that Wikipedia/Wikimedia have unreported harassment and pedophilia scandals.
*TL;DR: It was previously claimed on this list that it's generally
technically possible to attribute information in the output of a LLM-based
chatbot (such as ChatGPT) to specific parts of the LLM's training data
(such as a Wikipedia article). These claims are dubious and we shouldn't
rely on them as we continue to navigate the relations between Wikimedia
projects and LLMs.*
On Sun, Mar 19, 2023 at 12:12 PM Lauren Worden <laurenworden89(a)gmail.com>
wrote:
[...]
>
> On Sun, Mar 19, 2023 at 1:20 AM Kimmo Virtanen
> <kimmo.virtanen(a)wikimedia.fi> wrote:
> >
> >> Or, maybe just require an open disclosure of where the bot pulled from
> and how much, instead of having it be a black box? "Text in this response
> derived from: 17% Wikipedia article 'Example', 12% Wikipedia article
> 'SomeOtherThing', 10%...".
> >
> > Current (ie. ChatGPT) systems doesn't work that way, as the source of
> information is lost in the process when the information is encoded into the
> model....
>
> In fact, they do work that way, but it takes some effort to elucidate
> the source of any given output. Anyone discussing these issues needs
> to become familiar with ROME:
> https://twitter.com/mengk20/status/1588581237345595394 Please see also
> https://www.youtube.com/watch?v=_NMQyOu2HTo
>
> I sense some confusion here. That paper (ROME, http://rome.baulab.info/ )
is about attributing a model's factual claims to specific parts (weights,
neurons) of its neural network (and then changing them). It is **not**
about attribution to specific parts of its training data (such as Wikipedia
articles or other web pages), which is what Wikimedians have been
expressing concerns about.
In other words, it's entirely unclear why this should contradict what Kimmo
had said (and, separately in this thread, Galder
<https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org…>
).
(Trying to understand LLMs with analogies can be treacherous
<https://threadreaderapp.com/thread/1631491972685869056.html>. But for
people who automatically assume that neural networks "do work that way" -
i.e. preserve this kind of provenance information - and that chatbots can
be required to disclose "where [they] pulled from and how much" for a
particular answer: Imagine someone accosting you in the street and asking
you where you had originally learned that Paris is the capital of
France, say. How many of us would be able to come up with a truthful answer
like "our geography teacher told us in third grade" or "I read this in
Encyclopaedia Britannica when I was 10 years old"?)
With luck we will all have the chance to discuss these issues in
> detail on the March 23 Zoom discussion of large language models for
> Wikimedia projects:
>
> https://meta.wikimedia.org/wiki/Wikimedia_Foundation_Annual_Plan/2023-2024/…
>
> The notes from that meeting (now at
https://meta.wikimedia.org/wiki/Wikimedia_Foundation_Annual_Plan/2023-2024/…
) contain the following statements:
*"In an ideal world, the Foundation would start internal projects to
replicate ROME and RARR."*
*"The Foundation should make a public statement in support of increasing
the accuracy of attribution and verification systems such as RARR [
https://arxiv.org/abs/2210.08726 <https://arxiv.org/abs/2210.08726> ]"*
These proposals do not seem to have made it into the WMF's actual annual
plan in the end. And I realize that this thread is already a couple of
months old. However, it still seems worth resolving misconceptions in this
regard, e.g. because there have been references to such claims more
recently in other community discussion spaces.
Regarding RARR (the second project proposed in that meeting and here on
this list, as something that WMF should replicate or embrace):
RARR is indeed designed to find **a** text document supporting a given
statement produced by an LLM. But importantly, it makes no claims that the
source it finds was "the" original source used by the LLM. The first "R" in
"RARR" stands for "Retrofit[ting]" attribution - not for "restoring",
"retrieving" or such. (In fact, RARR doesn't even try to find a source in
the model's training corpus. It simply does a Google search of the entire
internet, see section 3.1 in the paper.) In other words, it too won't
"elucidate **the** source of any given output" as claimed above (my
bolding).
This is particularly relevant in light of the fact that e.g. on English
Wikipedia we generally require information to be attributable to reliable
external sources. So even if a chatbot's statement was indeed based on a
Wikipedia article, but that Wikipedia article cited the New York Times for
this information, RARR might very well pick the NYT article as source
instead of Wikipedia.
A larger issue here is that individual facts are not owned by any company
or community. Specifically, they are not copyrightable (as Wikipedians are
well aware from their daily practice: we can't enforce citing sources -
[[WP:BURDEN]] - as a legal requirement like we do for [[WP:COPYPASTE]]).
This should be kept in mind by folks who advocate for a moral or even legal
obligation for LLMs to "cite their sources'' for their output (like earlier
in this thread: "just require an open disclosure of where the bot pulled
from and how much").
Back to the technical difficulties and claims that machine learning models
"do work that way":
Folks may also be interested in a general overview paper titled "Training
Data Influence Analysis and Estimation: A Survey" (
https://arxiv.org/abs/2212.04612 ). It says e.g. that
"it can be very difficult to answer even basic questions about the
relationship between training data and model predictions; for example:
[...] Which instances in the training set caused the model to make a
specific prediction?"
Now all that said, some weeks ago, Anthropic (a startup focused on
responsible use of AI, which is researching interpretability of LLMs)
released a new research paper that actually tries to tackle this very
difficult question in case of LLMs, and do something like the kind of
attribution we are concerned with here:
"Large language models have demonstrated a surprising range of skills and
behaviors. How can we trace their source? In our new paper, we use
influence functions to find training examples that contribute to a given
model output. [...]" (
https://threadreaderapp.com/thread/1688946685937090560.html )
It looks like really interesting cutting-edge research. (They used some
advanced approximation techniques to make the required calculations
feasible in case of some LLMs that are however still much smaller than e.g.
GPT 3.5 or whatever the version of ChatGPT you use is based on.) If someone
with access to the required huge compute resources and technical skills
would apply the methods described in the paper (
https://arxiv.org/abs/2308.03296 ) to specifically investigate the case of
Wikipedia, that could be fascinating. (There's also an upcoming conference
soliciting such research: https://attrib-workshop.cc/ .)
But before anyone gets too excited: This is a statistical approach focused
on generating estimates of influence ratios only. And from the concrete
examples Anthropic shares, it seems that the relation between source and
output is typically much more diffuse and tenuous than simplistic "AI
steals from Wikipedia!!1!" type arguments would let you believe. (That's
even true for their "simple factual queries" category - see figure 42 in
the paper, for example: "Prompt: Inflation is often measured using /
Completion: the Consumer Price Index." Table 9 in the appendix describes
the sequences from the training data that were found to be most influential
for the answer in one of the examined LLMs. It observes that most of these
source texts don't actually contain the term "consumer price index",
contrary to what one might expect.)
The Anthropic authors also state generally that:
"Model outputs do not seem to result from pure memorization [...] the
influence of any particular training sequence is much smaller than the
information content of a typical sentence, so the model does not appear to
be reciting individual training examples at the token level." (
https://www.anthropic.com/index/influence-functions )
Regards, Tilman
PS 1: Of course there are still ways to make an LLM-based chatbot actually
cite sources, if one is prepared to restrict the kind of answers it can
give. Bing Chat
<https://en.wikipedia.org/wiki/Microsoft_Bing#Bing_Chat> actually
does this by default (or at least tries to), as opposed to ChatGPT,
retrieving live sources at question time. Specifically regarding Wikipedia,
one can prompt ChatGPT or other LLMs to only answer based on Wikipedia
content, and hope that it complies without hallucinating. (I summarized
three such approaches, including the Wikimedia Foundation's ChatGPT plugin,
here: https://meta.wikimedia.org/wiki/Research:Newsletter/2023/July .)
"Retrieval-augmented generation" (RAG) is a good search term for learning
more about similar approaches.
PS 2: All this is separate from questions about the overall influence of
Wikipedia (or other parts of an LLM's training data) on the general
performance of e.g. ChatGPT, with regard to its average factual accuracy,
biases etc. The answers there are also much less clear than some appear to
assume, but that's a topic for another post.
Dear List Admins,
I came to my attention that this user is subscribed to many mailing lists
and it seems just dumping random text in every mailing list and seems
irrelevant to the thread or topic raised.
Kind regards,
Butch Bustria
Admin to 2 large mailing lists
---------- Forwarded message ---------
From: Sandeep Manikpuri <iamsandeepmanikpuri(a)gmail.com>
Date: Sun, Sep 10, 2023, 12:44 PM
Subject: Talk to solve
To: <wikimedia-l-owner(a)lists.wikimedia.org>
I need a responsive, committed and paitent user or co worker to help,answer
and guide me when I will need as essential as breathing is for living. I am
a blocked ( everywhere ) user who don't know who is with me for assisting.
A nice and friendly user QuicoleJR is my need if you ask me and I don't
know other users or don't want know some harassment spealist user.
QuicoleJR is perhaps unable to receive my messages or reply due to
universal block status of mine but was helpful whenever I needed. I request
you to provide me access to QuicoleJR user and I will ask and learn then
work as well if allowed with no other users help. I have many things to
teach QuicoleJR user while I'm on learning if survive here. I need that
user only for the company and will not trust in other deletestars or
deletadmins or any Deletopidea user and personal. I hope this message will
be read and can be taken to concern of answering in rejected or accepted
both aspect of my universal block identity and my achievements of universal
records of trying and falling frequency of attempts in Wikipedia.
Thank you if you got this message.
If not ! then you are a lier.
Dear all,
One year ago, we launched Ikusgela<https://commons.wikimedia.org/wiki/Category:Ikusgela>, the Basque Wikimedians User Group effor to add high quality pedagogical videos to Wikipedia. We started with philosophy videos, and in two weeks we are launching the second season of this philosophy series. Season 2, episode 1 will be about Aristotle, one of the most referenced thinkers of all time. In order to make this launch even more interesting, we have come with the idea of having as much languages as possible via subtitles, as we know how much wikimedians like this kind of collective efforts.
That's why I'm writing all of you to seek for help with these translations. We have uploaded subtitles in 3 languages (Basque, Spanish and English) to Commons. You can see them here: https://commons.wikimedia.org/wiki/Special:PrefixIndex?prefix=Ikusgela%20-%…. What we are asking is to translate those subtitles to your language, so it is availabe when we launch the video. If you want to help, these are the steps to follow:
1. Choose the language you will use as a base from the list.
2. Click on it and choose the Edit button.
3. Copy all the text and paste it in whatever text editor you use. I will suggest using a subtitle translator so you have the structure there. https://translatesubtitles.co/ seems quite good.
4. Review the translation, save or download it and upload to Commons adding the language prefix (i.e, instead of TimedText:Ikusgela_-_Aristoteles.webm.en.srt it should be TimedText:Ikusgela_-_Aristoteles.webm.fr.srt if you are translating ot to French).
5. Save and relax. In two weeks, your subtitle will be available along the video in all platforms.
Thanks for your help, it is really appreciated.
Galder Gonzalez
Basque Wikimedians User Group
1.
Aside from the whole idea of a neutrality policy that thinks Wikipedia
wasn't neutral on either Climate Change or Covid19, and a board that is 60%
founders votes rather than the 9% of the WMF. Justapedia doesn't want
breaking news of anything that has happened in the last 7 days.
https://justapedia.org/wiki/Justapedia:Justapedia_Foundation#About
My prediction is that if the site does get any traction, and there are
millions of antivaxxers and climate change deniers who might well prefer it
to Wikipedia, the 7 day thing will bite them hard.
When famous people die, Wikipedia gets extra visitors to those pages. If
Justapedia had been around at the time of Michael Jackson's death, or of
Osama Bin Laden's, this policy alone would have led to conflict, or been
quickly abandoned. Here in the UK there will be a General Election in the
next 17 months, a large number of Wikipedia articles will be updated very
quickly with the results, probably including a new Prime minister. If
Justapedia is still around for the next UK General Election, and if it has
taken even 1% of Wikipedia's audience and community, then either this
policy will be abandoned or the people who enforce it will have to revert
loads of edits by newbies. Justapedia may not want "thought police" and may
aspire to be collegial "The Justapedia Foundation firmly believes in
maintaining a collegial environment for our volunteers"
https://justapedia.org/wiki/Justapedia:Justapedia_Foundation#Community but
this policy alone will make them a non collegial place, even if they
succeed in recruiting a community of people who share their views on
Climate Change and Covid19.
I'm aware that there are deep tensions within our community, and not just
between the on and off wiki components. I wasn't surprised to see a new
fork come forth, and if the intent, remit and design had been different
then I might well have created an account there and at the least had a foot
in both camps. As it is, I can't see Justapedia becoming something I would
like to donate my time to. But then I doubt they'd want me anyway
given that I accept the science on Climate Change and Covid19..
WSC
>
>
> ------------------------------
>
> Message: 7
> Date: Tue, 12 Sep 2023 12:00:44 +0100
> From: David Gerard <dgerard(a)gmail.com>
> Subject: [Wikimedia-l] Re: Launch of Justapedia
> To: Wikimedia Mailing List <wikimedia-l(a)lists.wikimedia.org>
> Message-ID:
> <
> CAJ0tu1E48S9LLt-GUPhkPs0dRyiRNLeeeN0fAPESLqKCj7DsJw(a)mail.gmail.com>
> Content-Type: text/plain; charset="UTF-8"
>
> what do you expect from an encyclopedia founded by world-famous
> communist Jimmy Wales
>
> -d.
>
> On Tue, 12 Sept 2023 at 08:52, Vi to <vituzzu.wiki(a)gmail.com> wrote:
> >
> > Indeed Venn diagrams are a left-wing-woke-cancel-culture propaganda.
> >
> > Vito
> >
> > Il giorno mar 12 set 2023 alle ore 08:15 Galder Gonzalez Larrañaga <
> galder158(a)hotmail.com> ha scritto:
> >>
> >> The logo is quite funny. According to that Information + Disinformation
> = Facts. It might be that they don't know what a Venn diagram is, or simply
> that they actually think that.
> >>
> >> Don't worry, this is just one more project that will fall into oblivion.
> >>
> >> Galder
> >> ________________________________
> >> From: Lauren Worden <laurenworden89(a)gmail.com>
> >> Sent: Tuesday, September 12, 2023 12:42 AM
> >> To: Wikimedia Mailing List <wikimedia-l(a)lists.wikimedia.org>
> >> Subject: [Wikimedia-l] Re: Launch of Justapedia
> >>
> >> The only specific and non-contradictory complaints about Wikipedia
> >> bias I can find on
> >> https://justapedia.org/wiki/Justapedia:Justapedia_Foundation are
> >> climate change and COVID-19, which are areas in which I think
> >> Wikipedia excels. The complaints about politics go in both directions.
> >>
> >> Perhaps Atsme can clarify?
> >>
> >> -LW
> >>
> >> On Mon, Sep 11, 2023 at 3:18 PM Andy Mabbett <andy(a)pigsonthewing.org.uk>
> wrote:
> >> >
> >> >
> >> > On Sun, 10 Sept 2023, 14:31 Rey Bueno via Wikimedia-l, <
> wikimedia-l(a)lists.wikimedia.org> wrote:
> >> >
> >> >> Justapedia
> >> >
> >> >
> >> > It's laughable, as the logo [1] that (inadvertently,?) shows
> "disinformation" and "lies" being given equal weight with "information" in
> determining "truth" suggests.
> >> >
> >> > But this [2] hints at the darker underbelly:
> >> >
> >> > "The Editorial Review Board (ERB) [...] will make binding decisions
> regarding the retention and rejection of article content, as well as serve
> to resolve content disputes, and notability issues... ERB members are
> required to have a high level, native understanding of written English"
> >> >
> >> >> Besides, there are alarming rumors I saw in Y Combinator [...]
> >> >
> >> >
> >> > Oh, please. Do better than that.
> >> >
> >> >
> >> > [1] https://justapedia.org/wiki/File:Hands-circ-sifts-sources-Sm.png
> >> >
> >> > [2] https://justapedia.org/wiki/Justapedia:Editorial_Review_Board
> >> >
> >> > _____
> *******************************************
>
.jpg){kind=link}
