@loleg@hachyderm.io

· 4mo ago

Dr Sasha Luccioni, the researcher behind #CodeCarbon and much more, is giving a seriously insightful talk about the impact of AI on the environment, economy and society c/o #AlanTuringInstitute
https://youtube.com/live/IVa98vXorgg?is=D69DmKrp3NxWzyyL

hachyderm.io

Hachyderm.io

6

0

4

0

o lаvrоvsky @loleg@hachyderm.io

· 6mo ago

Replying to

@halla @Li @KitsuneofInari I completely respect the decision of the maintainers, and think it’s better to have a clear policy than let the debate stew and simmer. Given the user base and sentiment against the GenAI topic, it’s possibly even the right move. I just want to see an even better justification, like: we don’t add code, we tactically remove it. Our code base is a sculpture, and LLMs aren’t much help here. Or even: as you kind of wrote above, Krita is an expression of the joy of art through the joy of coding. We aren’t feeling the vibes when you PR us your lobster soup, but feel free to make your own painting program from scratch. Call it Pincha, if you want!

5

1

0

o lаvrоvsky @loleg@hachyderm.io

· 7mo ago

Switzerland has finally taken steps to regulate social media and search engines, the Federal Council circulated a draft law invoking platforms like Instagram, X, and Google for public consultation - the window on this closed this week.

Organizations like @algorithmwatch_ch@mastodon.social and @digiges@mastodon.social have analyzed the draft, highlighting gaps, and pushing for a balanced regulation that protects users while supporting innovation. This initiative reflects a broader global effort to regulate digital platforms responsibly, with Switzerland aiming to contribute its unique perspective as a federal and multilingual country.

The draft aims to tackle issues like algorithmic transparency, hate speech, and liability. There are concerns about how effective these measures will be, especially given the global nature of such platforms. Next steps include revisions based on public feedback, parliamentary debate, and further adjustments.

Next step from here: AI-regulation.

The roadmap for the legislative process was shared by Estelle Pannatier & Rahel Estermann in the talk at #Winterkongress

mastodon.social

AlgorithmWatch CH (@algorithmwatch_ch@mastodon.social) - Mastodon

7

0

3

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

Enjoyed the Denmark 🇩🇰 🇮🇹 Italy match with a nailbiting third period and overtime, ending in Patrick Russell slipping past Davide Fadani in the penalty - Canada about to play Slovakia here in 🧀 Fribourg #icehockey #casualspectator

hachyderm.io

2

0

1

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

@beyondmachines1@infosec.exchange what kind of soil is being fertilized here, ai wonder ..

2

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

„The majority of our current compute is 600 GPUs in 75 TinyBox Pro machines. They were built in-house, which saves us money and ensures they suit our needs. Our self-built machines fail at a similar rate to pre-built machines we’ve bought, but we’re capable of fixing them ourselves quickly.“ #datacenter #repair
https://blog.comma.ai/datacenter/

hachyderm.io

2

0

3

0

o lаvrоvsky @loleg@hachyderm.io

· 7mo ago

Replying to

Human-in-the-loop approaches are not a solution for all problems, as they are limited by time, resources, and expertise. There is a tension between the desire for more efficient, consistent AI decisions and the need for fairness, transparency, and accountability. Incorrect decisions can result in systematic errors and a lack of context consideration. The acceptable risk of errors must be carefully weighed. In Switzerland, there is a lack of comprehensive horizontal AI laws, so the administration relies on existing sectoral regulations (e.g., data protection provisions, specialized laws, internal guidelines). The Council of Europe is planning regulations, but definitions, governance, oversight, and fundamental rights protection remain uncertain. Only a few cantons, such as Aargau and Luzern, have developed comprehensive AI strategies; others deliberately opt out of a comprehensive strategy. The focus often is on operational AI, which directly impacts public value. Not every canton can build its own cloud infrastructure, highlighting the necessity for cooperation and shared solutions. #Winterkongress summary translated with #Apertus

3

0

1

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

Together with Michele Bolla, Mathias Born, Floriane Pochonlevit, Fabrice Gouzi, we supported the BFS #opendata challenge number 10 in the #govtechhack26 hackathon.

This webapp deployed on #Infomaniak uses #Apertus to connect to the MCP service of #i14y - to help discover and analyse #opendata with algorithms of the Valentine Python library of #delftdata

Look at the screencast and screenshots, or click the DEMO link to try 📊 It's a perfect match! 💕 yourself at https://govtech.digisus-lab.ch/project/40

hachyderm.io

Hachyderm.io

1

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

#lmstudio #multimodal
https://lmstudio.ai/blog/unified-mlx-engine

hachyderm.io

Hachyderm.io

1

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

#Apertus featured in #WIRED Italy https://www.wired.it/article/intelligenza-artificiale-svizzera-apertus-modelli/

hachyderm.io

Hachyderm.io

1

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

Pitch presentations at the GovTech Hackathon 2026 organized by BFH in Zollikofen BE #govtechhack26

hachyderm.io

1

5

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

The documentary film by Mario Sixtus (DE, 2025, 52mn) is available for another week on ARTE #weekendscreening
"AI and the Death of the Internet"
https://www.arte.tv/en/videos/122187-000-A/ai-and-the-death-of-the-internet/

hachyderm.io

Hachyderm.io

1

0

2

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

@phf@dmv.community @MaSven@chaos.social we need a better *nux

1

0

o lаvrоvsky @loleg@hachyderm.io

· 7mo ago

Replying to

The use of AI often focuses on enhancing internal efficiency and the application of generative AI. Citizen-oriented applications typically limit themselves to visible chatbots, often based on simple rule-based technology (ADMT), which technically is not AI, or on machine learning (ML), which learns probabilistic patterns from data. Algorithmic systems encompass both approaches and share similar risks and consequences such as bias, discrimination, and a lack of transparency. The most common use cases and motivations for AI in administration are to increase efficiency, automate repetitive tasks, and support staff. There is significant potential for optimizing case processing, improving fraud detection, and risk reduction, as well as a wide range of application possibilities. However, central challenges include data protection, surveillance issues, discrimination, a lack of transparency and explainability, and the potential for flawed decisions. Incorrect AI systems can lead to significant financial and social damage as well as a loss of public trust. Therefore, transparency, legal protection, and human oversight are crucial, especially given the direct impact on citizens' fundamental needs ...

2

1

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

@PrivateCoffee@cuddly.space amazing work

1

2

0

o lаvrоvsky @loleg@hachyderm.io

· 18mo ago

Replying to

@GuenatGuillaume je vais demander ce soir au meetup retro du https://playbern.ch

playbern.ch

2

0

o lаvrоvsky @loleg@hachyderm.io

· 18mo ago

Replying to on hachyderm.io

If you support the mission, are interested in expanding the vision and driving results at @OpendataCH@mastodon.social - please consider applying for the job of executive director! Feel free to ask @florihas@mastodon.social or drop info@ a quick line if you have any questions https://www.kampajobs.ch/job/geschaeftsleitung-70-100-0

kampajobs.ch

1

0

2

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

Some related computing history trivia being passed around at https://news.ycombinator.com/item?id=47724571

news.ycombinator.com

A communist Apple II and fourteen years of not knowing what you're testing | Hacker News

0

o lаvrоvsky @loleg@hachyderm.io

· 10mo ago

Replying to

Recap video by Moving Water https://youtu.be/OfHl6f-jQFo #DINAconCH #dinacon25 #dinacon2025 #dinacon

Recap Video DINAcon 2025

0

o lаvrоvsky @loleg@hachyderm.io

· 10mo ago

Replying to

Join the anniversary hackathon this weekend: https://fosstodon.org/@yantar92/115476425604511892

fosstodon.org

Ihor Radchenko (yantar92): "A task from Org mode will be one of the projects …" - Fosstodon

0

o lаvrоvsky @loleg@hachyderm.io

· 10mo ago

Replying to

For more info on #Apertus see also my blog post https://hachyderm.io/@loleg/115259229881276635

hachyderm.io

o lаvrоvsky: "Fresh post on my experiences setting up #Apertus …" - Hachyderm.io

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

« Right now, I’m particularly proud of #Apertus. Together with colleagues at ETH Zurich, we developed the first transparent, ethical AI language model. It’s a step toward strengthening Switzerland’s digital autonomy, leaving us less reliant on outside parties whose agendas differ from ours. What makes it even more meaningful is that the initiative came from the ground up. A group of people had the courage to take on this ambitious challenge. If you’d asked me at the start, I might have doubted it could be done. But they proved it could. That’s the EPFL mindset: making the impossible possible.» — Anna Fontcuberta-Morral @epfl

0

o lаvrоvsky @loleg@hachyderm.io

· 6mo ago

Replying to

@Li @halla @KitsuneofInari fine, they’re not tools, they’re stamps.

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Open Future

NOODL. An experiment in equitable data licensing: promise and limits – Open Future

This report analyzes NOODL—a tiered licensing framework for African language datasets—as an experiment in equitable data commons governance.

0

1

0

o lаvrоvsky @loleg@hachyderm.io

· 4mo ago

A new drug that protects you from COVID post exposure - via @msalathe@social.epfl.ch
https://www.nejm.org/doi/10.1056/NEJMoa2509306

social.epfl.ch

Marcel Salathé (@msalathe@social.epfl.ch) - Mastodon EPFL

0

o lаvrоvsky @loleg@hachyderm.io

· 10mo ago

Replying to

@jroper@datasci.social yes, we are recording the talks (most are in German but that’s just an issue of subtitling) It’s not too late to join our HACKnight from 17:00

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

@richlv@mastodon.social sure, but who's to say that a vibe coded feature (perhaps in a form of a plugin, or fork or branch if you will) should not be used for artistic effort? There were already thousands of playful inventive and sloppy creative-throwaway-apps made by enterprising midnight hackers before genAI came along ..

0

3

0

o lаvrоvsky @loleg@hachyderm.io

· 5mo ago

Replying to

@richlv@mastodon.social I'm commenting on the originally asserted basic incompatibility between LLM generated code and, in this case, painting software used for artistic effort.

0

1

0