Benjamin
@premium-content.bsky.social
European.
Interests: IP Law, AI, Digital Policy
Anonymous. No affiliations.
Interests: IP Law, AI, Digital Policy
Anonymous. No affiliations.
Reposted by Benjamin
seriously, all you people fantasizing about how great it would be to ban technology and stop human progress disgust me
i don’t understand how you can talk about this stuff without feeling all the meaning drain from your lives and grey nothingness consume your awareness
i don’t understand how you can talk about this stuff without feeling all the meaning drain from your lives and grey nothingness consume your awareness
December 26, 2025 at 10:26 PM
seriously, all you people fantasizing about how great it would be to ban technology and stop human progress disgust me
i don’t understand how you can talk about this stuff without feeling all the meaning drain from your lives and grey nothingness consume your awareness
i don’t understand how you can talk about this stuff without feeling all the meaning drain from your lives and grey nothingness consume your awareness
Reposted by Benjamin
the success of the user-created For You feed while the official Discover feed flops is actually a pretty major win for Bluesky — it literally is the embodiment of their goal to decentralize feed generation and give users more control/autonomy
December 26, 2025 at 3:27 PM
the success of the user-created For You feed while the official Discover feed flops is actually a pretty major win for Bluesky — it literally is the embodiment of their goal to decentralize feed generation and give users more control/autonomy
Reposted by Benjamin
The ECJ’s Russmedia ruling quietly overturns a cornerstone of EU platform law.
ERIK TUCHTFELD (@erik.d-64.social) (@mpil.de) shows how GDPR liability may replace notice-and-takedown with de facto monitoring – potentially reshaping online speech across Europe.
🔗 verfassungsblog.de/notice-and-t...
ERIK TUCHTFELD (@erik.d-64.social) (@mpil.de) shows how GDPR liability may replace notice-and-takedown with de facto monitoring – potentially reshaping online speech across Europe.
🔗 verfassungsblog.de/notice-and-t...
December 21, 2025 at 5:51 PM
The ECJ’s Russmedia ruling quietly overturns a cornerstone of EU platform law.
ERIK TUCHTFELD (@erik.d-64.social) (@mpil.de) shows how GDPR liability may replace notice-and-takedown with de facto monitoring – potentially reshaping online speech across Europe.
🔗 verfassungsblog.de/notice-and-t...
ERIK TUCHTFELD (@erik.d-64.social) (@mpil.de) shows how GDPR liability may replace notice-and-takedown with de facto monitoring – potentially reshaping online speech across Europe.
🔗 verfassungsblog.de/notice-and-t...
Reposted by Benjamin
40 Years Of #Copyright Obstruction To Human Rights And Social Justice - www.techdirt.com/2025/12/23/4... #visuallyimpaired #marrakesh
40 Years Of Copyright Obstruction To Human Rights And Social Justice
One of the little-known but extremely telling episodes in the history of modern copyright, discussed in Walled Culture the book (free digital versions available), concerns the Marrakesh T…
www.techdirt.com
December 24, 2025 at 9:26 AM
40 Years Of #Copyright Obstruction To Human Rights And Social Justice - www.techdirt.com/2025/12/23/4... #visuallyimpaired #marrakesh
Reposted by Benjamin
when i think about AI, i think about the universal approximation theorem, erdõs problems and imo medals, rate of change, and the things i can do with frontier language models like opus 4.5 and the insane internal complexity
in 2025 most ppl think of “annoying slop video” or “copilot win11 popup”
in 2025 most ppl think of “annoying slop video” or “copilot win11 popup”
December 12, 2025 at 7:16 PM
when i think about AI, i think about the universal approximation theorem, erdõs problems and imo medals, rate of change, and the things i can do with frontier language models like opus 4.5 and the insane internal complexity
in 2025 most ppl think of “annoying slop video” or “copilot win11 popup”
in 2025 most ppl think of “annoying slop video” or “copilot win11 popup”
Reposted by Benjamin
new blog post! can small, open-source models also introspect, detecting when foreign concepts have been injected into their activations? yes! (thread, or full post here: vgel.me/posts/qwen-i...)
December 21, 2025 at 12:14 AM
new blog post! can small, open-source models also introspect, detecting when foreign concepts have been injected into their activations? yes! (thread, or full post here: vgel.me/posts/qwen-i...)
Reposted by Benjamin
Quite the chart. www.ft.com/content/30a4...
December 20, 2025 at 7:47 AM
Quite the chart. www.ft.com/content/30a4...
Reposted by Benjamin
Fair use is the backbone of the U.S.’s balanced copyright ecosystem that supports innovation and creativity. @ccianet.bsky.social’s new report finds that in 2023 fair use industries made up 18% of the U.S. economy and employed 22 million employees.
Fair Use in the U.S. Economy - 2025 Edition - CCIA Research Center
ccianet.org
December 18, 2025 at 7:46 PM
Fair use is the backbone of the U.S.’s balanced copyright ecosystem that supports innovation and creativity. @ccianet.bsky.social’s new report finds that in 2023 fair use industries made up 18% of the U.S. economy and employed 22 million employees.
Reposted by Benjamin
We spent long time studying the question "does training data filtering help prevent generative models from generating unwanted content?"
TL;DR: Nope. In some cases it slightly increases difficulty but not enough to prevent even average users from finding ways to create what they want.
TL;DR: Nope. In some cases it slightly increases difficulty but not enough to prevent even average users from finding ways to create what they want.
Our new paper demonstrates that removing images of children from training datasets of text-to-image models fails to prevent the misuse of these models to generate child sexual abuse material (AI CSAM). A thread 🧵
December 16, 2025 at 10:42 AM
We spent long time studying the question "does training data filtering help prevent generative models from generating unwanted content?"
TL;DR: Nope. In some cases it slightly increases difficulty but not enough to prevent even average users from finding ways to create what they want.
TL;DR: Nope. In some cases it slightly increases difficulty but not enough to prevent even average users from finding ways to create what they want.
Reposted by Benjamin
Reposting for "digital chastity belt":
Fancy a digital chastity belt?
The government wants Apple and Google to stop you from taking or sending pics of your bits.
Intrusive scanning tech on your phone would block nudes unless you verify your age with biometrics or official ID.
It's creepy mission creep.
arstechnica.com/tech-policy/...
The government wants Apple and Google to stop you from taking or sending pics of your bits.
Intrusive scanning tech on your phone would block nudes unless you verify your age with biometrics or official ID.
It's creepy mission creep.
arstechnica.com/tech-policy/...
UK to “encourage” Apple and Google to put nudity-blocking systems on phones
Government seeks “nudity-detection algorithms” in iOS and Android, report says.
arstechnica.com
December 16, 2025 at 12:57 PM
Reposting for "digital chastity belt":
Reposted by Benjamin
This long-term thread will collect stories about Germany's absurd, over-lawyered and court-heavy bureaucracy following my Free Exchange column this week. www.economist.com/finance-and-...
Germany has a lawyer problem
Its endless bureaucratic rules trap would-be reformers
www.economist.com
December 16, 2025 at 10:45 AM
This long-term thread will collect stories about Germany's absurd, over-lawyered and court-heavy bureaucracy following my Free Exchange column this week. www.economist.com/finance-and-...
Reposted by Benjamin
Checking #parking violations is something scan cars can do very quickly. Except in Germany where data protection rules and court rulings are so tight that such cars are illegal.
No, I am not making this up. www.rbb24.de/panorama/bei...
No, I am not making this up. www.rbb24.de/panorama/bei...
Digitale Parkraumbewirtschaftung in Berlin liegt vorerst auf Eis
Parkgebühren werden nicht immer bezahlt - aber Kontrollen durch Mitarbeiter des Ordnungsamtes sind aufwändig. Sogenannte Scancars, die die Nummernschilder der Parkenden im Vorbeifahren prüfen, sollten...
www.rbb24.de
December 16, 2025 at 10:54 AM
Checking #parking violations is something scan cars can do very quickly. Except in Germany where data protection rules and court rulings are so tight that such cars are illegal.
No, I am not making this up. www.rbb24.de/panorama/bei...
No, I am not making this up. www.rbb24.de/panorama/bei...
Reposted by Benjamin
New from @jpquintais.bsky.social & me: a critical and wide analysis of the Obligations of all General-Purpose AI Model providers under the EU AI Act. We give a deep ~50 page, 300 footnote treatment and find so many tensions, loopholes, inconsistencies and more.
Link: files.michae.lv/papers/Veale...
Link: files.michae.lv/papers/Veale...
November 18, 2025 at 1:56 PM
New from @jpquintais.bsky.social & me: a critical and wide analysis of the Obligations of all General-Purpose AI Model providers under the EU AI Act. We give a deep ~50 page, 300 footnote treatment and find so many tensions, loopholes, inconsistencies and more.
Link: files.michae.lv/papers/Veale...
Link: files.michae.lv/papers/Veale...
Reposted by Benjamin
Not sure the Online Safety Acts — either UK or AUS flavours — are ready for kids setting up and running social media services themselves as regulated entities. This already happens: expect more. I ran online forums *as a child* that would now need child access assessments under the UK OSA!
December 14, 2025 at 11:45 AM
Not sure the Online Safety Acts — either UK or AUS flavours — are ready for kids setting up and running social media services themselves as regulated entities. This already happens: expect more. I ran online forums *as a child* that would now need child access assessments under the UK OSA!
Reposted by Benjamin
I wonder how much discussion this is provoking in the academic philosophy world. Probing raw idea space, seeing how language mechanistically leads to concepts and how that links concepts almost without humans to explicitly do the linking (unlike in expert systems), sounds deeply philosophical.
December 13, 2025 at 7:52 PM
I wonder how much discussion this is provoking in the academic philosophy world. Probing raw idea space, seeing how language mechanistically leads to concepts and how that links concepts almost without humans to explicitly do the linking (unlike in expert systems), sounds deeply philosophical.
Reposted by Benjamin
How the ‘Brussels effect’ backfired
How the ‘Brussels effect’ backfired
Once a model for the world, the EU’s rulemaking machine has faltered under the weight of its own ambition
giftarticle.ft.com
December 14, 2025 at 8:36 AM
How the ‘Brussels effect’ backfired
Reposted by Benjamin
Die Seite KI-vor-Gericht ist online! Prof. Heetkamp (ehem. Richter am LG) u.a. haben sie ins Leben gerufen um deutsche Halluzinations-Fälle zu sammeln.
Senden Sie Gerichtsentscheidungen zu KI-Halluzinationen an:
-> ki-vor-gericht.de
@fsteiner.bsky.social @offenenetze.de @rewis.io #TeamResopal
Senden Sie Gerichtsentscheidungen zu KI-Halluzinationen an:
-> ki-vor-gericht.de
@fsteiner.bsky.social @offenenetze.de @rewis.io #TeamResopal
December 13, 2025 at 7:30 AM
Die Seite KI-vor-Gericht ist online! Prof. Heetkamp (ehem. Richter am LG) u.a. haben sie ins Leben gerufen um deutsche Halluzinations-Fälle zu sammeln.
Senden Sie Gerichtsentscheidungen zu KI-Halluzinationen an:
-> ki-vor-gericht.de
@fsteiner.bsky.social @offenenetze.de @rewis.io #TeamResopal
Senden Sie Gerichtsentscheidungen zu KI-Halluzinationen an:
-> ki-vor-gericht.de
@fsteiner.bsky.social @offenenetze.de @rewis.io #TeamResopal
Reposted by Benjamin
OLG Hamburg weist Berufung in Sachen Kneschke ./. LAION zurück.
Anwendbarkeit der Text & Data Mining-Schranke § 44b UrhG bejaht.
Anwendbarkeit der Text & Data Mining-Schranke § 44b UrhG bejaht.
Irgendjemand Mittwoch am OLG Hamburg?
Kneschke ./. LAION e.V., die Berufung zu LG Hamburg, 27.09.2024 - 310 O 227/23 (dejure.org/2024,25822)
PS: ich bin über die "frühe" Terminierung ernsthaft irritiert, Berufung war November 2024.
Kneschke ./. LAION e.V., die Berufung zu LG Hamburg, 27.09.2024 - 310 O 227/23 (dejure.org/2024,25822)
PS: ich bin über die "frühe" Terminierung ernsthaft irritiert, Berufung war November 2024.
December 10, 2025 at 11:02 AM
OLG Hamburg weist Berufung in Sachen Kneschke ./. LAION zurück.
Anwendbarkeit der Text & Data Mining-Schranke § 44b UrhG bejaht.
Anwendbarkeit der Text & Data Mining-Schranke § 44b UrhG bejaht.
Reposted by Benjamin
I know the author is trying to be funny but as an unfunny lawyer who is watching policymakers and courts unravel our fundamental freedoms over technology they don't understand, I tend to be pretty unamused by stuff like this.
December 9, 2025 at 10:15 PM
I know the author is trying to be funny but as an unfunny lawyer who is watching policymakers and courts unravel our fundamental freedoms over technology they don't understand, I tend to be pretty unamused by stuff like this.
Reposted by Benjamin
📊 Chart of the week: #AI chatbot market shares
🔍 Could increased reliance on AI for search functions lead to new competition-distorting effects?
📑 Read the paper by Bertin Martens
🔗 buff.ly/5wq1jyd
🔍 Could increased reliance on AI for search functions lead to new competition-distorting effects?
📑 Read the paper by Bertin Martens
🔗 buff.ly/5wq1jyd
December 5, 2025 at 11:30 AM
📊 Chart of the week: #AI chatbot market shares
🔍 Could increased reliance on AI for search functions lead to new competition-distorting effects?
📑 Read the paper by Bertin Martens
🔗 buff.ly/5wq1jyd
🔍 Could increased reliance on AI for search functions lead to new competition-distorting effects?
📑 Read the paper by Bertin Martens
🔗 buff.ly/5wq1jyd
Reposted by Benjamin
All models are wrong; some are useful.
All LLM outputs are hallucinations. Some are useful.
All LLM outputs are hallucinations. Some are useful.
August 29, 2025 at 10:11 PM
All models are wrong; some are useful.
All LLM outputs are hallucinations. Some are useful.
All LLM outputs are hallucinations. Some are useful.
Reposted by Benjamin
A couple years (!) in the making: we’re releasing a new corpus of embodied, collaborative problem solving dialogues. We paid 36 people to play Portal 2’s co-op mode and collected their speech + game recordings.
Paper: arxiv.org/abs/2512.03381
Website: berkeley-nlp.github.io/portal-dialo...
1/n
Paper: arxiv.org/abs/2512.03381
Website: berkeley-nlp.github.io/portal-dialo...
1/n
December 5, 2025 at 6:54 PM
A couple years (!) in the making: we’re releasing a new corpus of embodied, collaborative problem solving dialogues. We paid 36 people to play Portal 2’s co-op mode and collected their speech + game recordings.
Paper: arxiv.org/abs/2512.03381
Website: berkeley-nlp.github.io/portal-dialo...
1/n
Paper: arxiv.org/abs/2512.03381
Website: berkeley-nlp.github.io/portal-dialo...
1/n
Reposted by Benjamin
Both sides of the Atlantic have been laying legal trenches for a year in preparation for this fight. Remember the rule that bars from the US any foreign official who limits the free speech of US corporations?
This doesn’t resolve quickly. Bluesky can start hiring more moderators who read German.
This doesn’t resolve quickly. Bluesky can start hiring more moderators who read German.
It seems reasonable to suspect that this is what yesterday‘s weird rule (barring visas for people who work on moderation or T&S) was actually about.
December 5, 2025 at 1:36 PM
Both sides of the Atlantic have been laying legal trenches for a year in preparation for this fight. Remember the rule that bars from the US any foreign official who limits the free speech of US corporations?
This doesn’t resolve quickly. Bluesky can start hiring more moderators who read German.
This doesn’t resolve quickly. Bluesky can start hiring more moderators who read German.
