Created
October 26, 2021 20:10
-
-
Save billfitzgerald/2fbe68a7270b744bf604398f44abebeb to your computer and use it in GitHub Desktop.
Doc 1: Rough OCR of Facebook Files released by Gizmodo: https://gizmodo.com/hey-kid-wanna-see-some-leaked-facebook-docs-1847936740
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
gateway groups qanon" | |
ey <> Ww & | |
Concerto Feedback Group a: | |
October 8, 2020: @ | |
These changes are new, small inconveniences piled atop frustrating user-expenente | |
decisions that Facebook has been making for more than a decade. But they are the latest | |
example of how Facebook tries to shape every user's experience through black box | |
algorithms—and how this approach harms not only individuals but the world at large | |
This article has actionable feedback around (many of which we may already be taking action | |
upon): | |
* Lack of Control: Seeing groups that are unsolicited (including groups they explicitly left!), | |
choices for comment ranking not being sticky, likewise even intimate conversations being | |
chained outwards through community view. | |
* Danger of Rabbit Holes: It discusses how while most are introduced to conspiracy theories | |
by word of mouth, recommendations engines and the chaining of content and communities | |
often reinforce the dangerous ideas or lead them far deeper. | |
Quotes + Potential Action Items | |
LACK OF CONTROL | |
But those are groups that we, the users, sought out and joined. Unsolicited content from | |
other, Unsubscribed groups is not always welcome. | myself noticed in recent weeks that | |
posts from groups | am not a member of appeared when | tried to use Facebook's | |
increasingly user-hostile app to engage with the handful of friends-and-family groups | do | |
regularly use. And those out-of-the-blue posts include content from two groups | explicitly | |
and intentionally left a month prior because they were making my life worse. | |
Having that kind of content also appear in your personal newsfeed (which has not yet been | |
rolled out to me) is apparently even worse. "It was creepier than | expected to see 'related | |
discussions’ hyped next to a short comments thread between my mom and my brother | |
about her latest post," tech writer Rob Pegoraro (who has occasionally written for Ars) | |
tweeted after experiencing the new feature. (He added that Facebook's obsession with | |
engagement "needs to be shot into the sun," a sentiment with which | agree.) | |
* Doublecheck that at a minimum we're not recommending groups people have’ “* —” | |
i Chats | |
REDACTED FOR CONGRESS | |
in, recommmenaations engines ana tne chaining or content ana | |
the dangerous ideas or lead them far deeper. | |
1 But those are groups that we, the users, sought out and joined. Unsolicited content from | |
} other, unsubscribed groups is not always welcome. | myself noticed in recent weeks that | |
| posts from groups | am not a member of appeared when | tried to use Facebook's | |
increasingly user-hostile app to engage with the handful of friends-and-family groups | do | |
regularly use. And those out-of-the-blue posts include content from two groups | explicitly | |
and intentionally left a month prior because they were making my life worse. | |
Having that kind of content also appear in your personal newsfeed (which has not yet been | |
rolled out to me) is apparently even worse. "It was creepier than | expected to see ‘related | |
discussions' hyped next to a short comments thread between my mom and my brother | |
about her latest post," tech writer Rob Pegoraro (who has occasionally written for Ars) | |
tweeted after experiencing the new feature. (He added that Facebook's obsession with | |
engagement "needs to be shot into the sun," a sentiment with which | agree.) | |
* Doublecheck that at a minimum we're not recommending groups people have left? | |
with me, IG has started showing follow recommendations in between stories. They provide | |
an option to see these recommendations less which | value the option to (although it | |
doesn't seem to be very effective at dialing it down at the moment) | |
ink we may need some threshold on the original post before applying adding chaining to | |
ugh CV. | can see how CV can feel different attached to a friends/family post vs. on a | |
r page's post. Maybe there needs to be a certain number people in a conversation | |
is applied? | |
Y”) | |
Y) | |
or | |
* Consider introducing a way to stop receiving in-feed unconnected content. This resonates 5 | |
LL | |
a | |
same time has introduced a slew of tweaks to the user interface on both | |
make it significantly harder to promote high-quality engagement on | |
rly in groups. First, all groups now sort by "latest activity” as their | |
er than by "recent posts." Sorting by "latest activity" drives users to | |
have comments—but every post is then sorted by "top comments," an | |
quence muddle that seems to have almost nothing to do with the | |
lves. Users can again choose to sort by "all comments" or "most | |
ices do not stick. Whether by design or by flaw, the decis*~~ *~ --* | |
<y, either, and you'll need to reselect it every single tin Chats | |
+ app | |
a | |
Facebook at the same time has introduced a slew of tweaks to the user interface on both == | |
Web and mobile that make it significantly harder to promote high-quality engagement on | |
the platform, particularly in groups. First, all groups now sort by “latest activity" as their | |
default setting rather than by "recent posts." Sorting by "latest activity" drives users to | |
posts that already have comments—but every post is then sorted by "top comments,” an | |
inscrutable, out-of-sequence muddle that seems to have almost nothing to do with the | |
conversations themselves. Users can again choose to sort by “all comments" or “most | |
recent," but those choices do not stick. Whether by design or by flaw, the decision to sort | |
by recent posts isn't sticky, either, and you'll need to reselectit every single time you post | |
a comment or navigate between posts. | |
Meaningful, thoughtful conversation—even in small, serious, well- moderated groups—has | |
become almost impossible to maintain. That, too, drives sniping, bickering, and extremism | |
7 ona small, conversational scale. | |
* Make choices on comment ranking sticky? | |
polarized discourse? How can they still reflect the group norms and standards? | |
DANGER OF RABBIT HOLES (FROM AN ARTICLE LINKED WITHIN) | |
} But the Q believers | spoke with stumbled upon these incomprehensible ideas the old- | |
} fashioned way. They say they were indoctrinated (though they don’t use that term) by their | |
, other family members, and friends, or introduced to the conspiracy through word- ' | |
mouth, rather than via the algorithms that have received the most national attention. y | |
rd-of-mouth alone isn’t enough. The ability of people sitting at home to follow the | |
Y) | |
Y) | |
LU | |
« Likewise how do we make sure the comments we are ranking aren't leading to more = | |
6 | |
) | |
r conspiracy theory and USPS phishing text scam. The Jeffrey Epstein | |
that involved an alleged sex trafficking ring and was covered by | |
as also Served as a key gateway into QAnon. | |
jorithmic recommendations seem to have compounding | |
yond evaluating our recommendations and invite system | |
DANGER OF RABBIT HOLES (FROM AN ARTICLE LINKED WITHIN) | |
But the Q believers | spoke with stumbled upon these incomprehensible ideas the old | |
fashioned way. They say they were indoctrinated (though they don’t use that term) by their | |
parents, other family members and friends, or introduced to the conspire through word- | |
of-mouth, rather than via the algorithms that have received the most nat ttention | |
But word-of-mouth al sn't enough. The ability : itti ome wv the | |
online rabbit holes downward is critical. Moms are seeing an ever-changing web ¢ f | |
trafficking conspiracy theories bounce around t ircle of mom friends | |
tain | |
debunked Wayfair conspiracy theory and USPS phishing text scam. The Jett | |
‘ | |
saga, a real-life case that involved an alleged sex trafficking ring and was c | |
reputable news outlets, has also served as a key gateway into QAnon | |
» Word-of-mouth and algorithmic recommendations seem to have compounding risks that | |
we need to consider beyond evaluating our recommendations and invite system | |
separately? | |
» | saw another post discussing the need for intentional friction in Concerto so seems like | |
we're thinking about it already! | |
Lastly | think the framing of "Engagement" is a good reminder that Concerto could easily fall | |
into the engagement trap. Using ranking to show UC, sort comments, and recommend groups | |
can lead to engagement that reflect "usage that comes from one part our brains, while the | |
User's decision of what makes your product worth their time comes from another". | |
#markdown | |
» Just Flagging and Comms Feedback | |
October 6, 2020 -® | |
is://arstechnica.com/.../facebooks-endless-quest-for | |
REDACTED FOR CONGRESS | |
>» DO! Q&A | |
How do we calculate searchability? | |
When our searchability is at 10%, what is that telling us? | |
1. 10% of returned content is violating ? | |
2. 10% of searches return violating content ? | |
3. Other? | |
— with mz: 3 others | |
©O4 | |
ent seen by 39 | |
Comment @ Share | |
C) Like | |
the answer is 3. Other:Definition: Of all of the VPVs on search results returned from | |
the essential query list (i.e. the list of queries indicating a user was searching for a | |
problem), what percent were violating according to human labeling. Source: | |
https://fb,workplace.com/notes/864276267638679 from | |
and | |
_ =e | |
https://fb.workplace.com/notes/475583250397974/ ») i | |
Search Terms used to define searchability:https |/fourl.com/unidash/8zcOh2fy | |
nd | |
Searchability: Intro and setting up the metric | |
Like Reply 2h. Edited | |
#resolved (Thank you for bearing witness to my self-help session) | |
Like Reply 1h oO 1 | |
EDACTED FOR CONGRESS | |
R | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment