david
@davd.dev
tech but ethical would be nice
skipping classes @ ETH Zurich
skipping classes @ ETH Zurich
data poisoning & adversarial examples exist, but for those to work you'd need to know exactly what kind of model would be used on your data, and only works in very specific scenarios (mostly academic experiments with few exceptions).
but there's no defence against data processing
but there's no defence against data processing
February 19, 2025 at 11:43 AM
data poisoning & adversarial examples exist, but for those to work you'd need to know exactly what kind of model would be used on your data, and only works in very specific scenarios (mostly academic experiments with few exceptions).
but there's no defence against data processing
but there's no defence against data processing
theres no inherent way of preventing scrapers from collecting data from atproto (all posts & content are public by design, anyone can collect all this data)
if you're talking about preventing models being trained on this data, there is also no way of doing this generically
if you're talking about preventing models being trained on this data, there is also no way of doing this generically
February 19, 2025 at 11:36 AM
theres no inherent way of preventing scrapers from collecting data from atproto (all posts & content are public by design, anyone can collect all this data)
if you're talking about preventing models being trained on this data, there is also no way of doing this generically
if you're talking about preventing models being trained on this data, there is also no way of doing this generically
very nice :) might be nice to have the option to disable/hide content with app.bsky lexicons tho at some point, e.g. once there is enough "native" Spark content
February 19, 2025 at 11:22 AM
very nice :) might be nice to have the option to disable/hide content with app.bsky lexicons tho at some point, e.g. once there is enough "native" Spark content
if what bryan mentions is implemented, i assume the labeler would declare somewhere (DID doc?) what report types it accepts, so it could just not list any type at all in order to declare it doesn't do moderation/reports/appeals
February 17, 2025 at 7:37 PM
if what bryan mentions is implemented, i assume the labeler would declare somewhere (DID doc?) what report types it accepts, so it could just not list any type at all in order to declare it doesn't do moderation/reports/appeals
yeah through com.atproto.report.createReport, and labelers are allowed to not implement it according to this blogpost
docs.bsky.app/blog/bluesky...
i don't think that there is any visual feedback in the appview, but could be wrong
docs.bsky.app/blog/bluesky...
i don't think that there is any visual feedback in the appview, but could be wrong
Bluesky's Moderation Architecture | Bluesky
Moderation is a crucial aspect of any social network. However, traditional moderation systems often lack transparency and user control, leaving communities vulnerable to sudden policy changes and pote...
docs.bsky.app
February 17, 2025 at 7:19 PM
yeah through com.atproto.report.createReport, and labelers are allowed to not implement it according to this blogpost
docs.bsky.app/blog/bluesky...
i don't think that there is any visual feedback in the appview, but could be wrong
docs.bsky.app/blog/bluesky...
i don't think that there is any visual feedback in the appview, but could be wrong
yeah the docs sre super lacking, generally the go implementation seems to get much less love than typescript.
if u want u can check out the golang api client i'm working on, creating posts is quite straightforward with it
github.com/davhofer/bot...
if u want u can check out the golang api client i'm working on, creating posts is quite straightforward with it
github.com/davhofer/bot...
GitHub - davhofer/botsky: A Bluesky API client in Go with useful features for writing automated bots.
A Bluesky API client in Go with useful features for writing automated bots. - davhofer/botsky
github.com
February 17, 2025 at 7:30 AM
yeah the docs sre super lacking, generally the go implementation seems to get much less love than typescript.
if u want u can check out the golang api client i'm working on, creating posts is quite straightforward with it
github.com/davhofer/bot...
if u want u can check out the golang api client i'm working on, creating posts is quite straightforward with it
github.com/davhofer/bot...
and the alternative of going into the direction of fact checking algos & tools, source verification etc. comes with its own set of problems, but i'm no expert there
February 4, 2025 at 11:35 PM
and the alternative of going into the direction of fact checking algos & tools, source verification etc. comes with its own set of problems, but i'm no expert there
anyone could create an army of accounts and reverse engineer the community notes algo to gain a lot of influence.
trying to prevent this using e.g. phone number verification doesn't seem like s great solution to me...
trying to prevent this using e.g. phone number verification doesn't seem like s great solution to me...
February 4, 2025 at 11:30 PM
anyone could create an army of accounts and reverse engineer the community notes algo to gain a lot of influence.
trying to prevent this using e.g. phone number verification doesn't seem like s great solution to me...
trying to prevent this using e.g. phone number verification doesn't seem like s great solution to me...
i do think the birdwatch/community notes approach is interesting and could be applied here quite easily (while not the cleanest solution this could even be done directly with comments, a labeler, maybe hashtags).
the problem with this is that it's super vulnerable to sybil attacks and the like
the problem with this is that it's super vulnerable to sybil attacks and the like
February 4, 2025 at 11:24 PM
i do think the birdwatch/community notes approach is interesting and could be applied here quite easily (while not the cleanest solution this could even be done directly with comments, a labeler, maybe hashtags).
the problem with this is that it's super vulnerable to sybil attacks and the like
the problem with this is that it's super vulnerable to sybil attacks and the like
we've gotten so used to free services and free-to-use platforms that we've come to expect everything to be free, but that's only due to the otherwise unethical business practices (in most cases) by the companies that offer these services
February 2, 2025 at 10:55 AM
we've gotten so used to free services and free-to-use platforms that we've come to expect everything to be free, but that's only due to the otherwise unethical business practices (in most cases) by the companies that offer these services
i personally think it would be a good idea for users to pay a small fee for the infrastructure they are using. this could potentially allow independent parties to run alternative relays and other infrastructure and improve federation & decentralization...
February 2, 2025 at 10:53 AM
i personally think it would be a good idea for users to pay a small fee for the infrastructure they are using. this could potentially allow independent parties to run alternative relays and other infrastructure and improve federation & decentralization...
Reposted by david
if youre wondering if blocks work here the answer is we call them nuclear blocks and yes they work
they will block a post, a poster, seven generations of their descendants and anyone who got too close
they will block a post, a poster, seven generations of their descendants and anyone who got too close
October 17, 2024 at 4:02 AM
if youre wondering if blocks work here the answer is we call them nuclear blocks and yes they work
they will block a post, a poster, seven generations of their descendants and anyone who got too close
they will block a post, a poster, seven generations of their descendants and anyone who got too close