Luca
lucab.phd
Luca
@lucab.phd
ML Ethics, Tech Policy, Italian food, Typos
why did it take so long?
September 9, 2025 at 1:17 PM
how the email found me:
May 13, 2025 at 4:59 PM
how the email found me:
April 26, 2025 at 3:39 AM
how the email found me:
April 14, 2025 at 7:11 AM
Drawing on my own experience in auditing Twitter's systems (from the inside), I explore the crucial role business leaders can play in navigating these emerging challenges.
March 27, 2025 at 12:06 AM
Audits are going to be a more critical business practice, especially given the increasing regulatory requirements. However they are not a straightforward and linear process, rather a web of different decisions. By their nature, they don't only bring technical challenges, but organizational ones.
March 27, 2025 at 12:06 AM
how the email found me:
March 22, 2025 at 9:39 PM
i do wonder a lot if things would have gone differently if we had an union
January 29, 2025 at 8:08 PM

The next plenary meeting is going to be Monday 27th January at 5pm CET, or 8 am if you are in the West Coast!

digital-strategy.ec.europa.eu/en/policies/...
General-Purpose AI Code of Practice
The first General-Purpose AI Code of Practice will detail the AI Act rules for providers of general-purpose AI models and general-purpose AI models with systemic risks.
digital-strategy.ec.europa.eu
January 16, 2025 at 12:42 AM
☞ Energy consumption (known on estimate) is also something that regulators can demand. This really helps put into perspective the performance of models. There has been a race towards bigger models, but not every task requires the latest and biggest, (e.g. locally run tasks)
January 16, 2025 at 12:42 AM
Considering the various lawsuits that are happening in the US on this exact topic, it's going to be interesting to see what features and models are NOT going to be released in EU.
January 16, 2025 at 12:42 AM
☞ Furthermore, regulators can ask more information about how the data was filtered to remove illegal and undesirable content, and how providers have acquired the legal rights to the data.
January 16, 2025 at 12:42 AM
☞ Providers will have to disclose to regulators and downstream providers “information on data used for training, testing and validation”. This is particularly important as training data (and their provenance) is considered part of the “secret sauce” of their models
January 16, 2025 at 12:42 AM
☞ Providers will have to disclose to the regulators a list of the name of the organization that label data for them. No more hiding behind 3rd party firms
January 16, 2025 at 12:42 AM
☞ They acknowledge that Generative AI models “present unique innovation opportunities, but also challenges to artists, authors, and other creators, and to the way their creative content is created, distributed, used, and consumed” (Recital 105 AI Act)
January 16, 2025 at 12:42 AM