@uavster See also minute 12 of this 2008 Lift Asia talk « Robots don't have to look like robots »

New Publication (in French) : Linguistic Prostheses: Typology, Authorship, and Proletarization. In Gefen Alexandre & Huneman Philippe (Eds), Philosophies of AI: thinking and writing with LLMs, Intellectica, 81, (pp.135-161) This article proposes a classification of linguistic prostheses, which are tools, devices, or technologies that transform the writing process. The concept of linguistic prosthesis brings together a wide variety of mediation devices under a single name. What they all have in common, however, is that they open up questions of authorship and the possible proletarianization of writing skills. Questions of authorship are central to many economic and legal processes involving the rights and responsibilities of authors. Questions of proletarianization touch on social issues of great importance linked to inequalities in writing skills, indirectly affecting questions as vast as the future of work or representativeness in democracy One of the conclusions of the article is that the issue of extending or supporting linguistic capabilities with prostheses is essentially linked to the fact that the user generally doesn't own them, but at best, simply rents access to them. The behavior of the prosthesis may change depending on the subscription taken out, or just when the underlying language model is updated. In the long term, the user has no guarantee of either their accessibility or the quality of their performance. The coupled system is therefore fundamentally unstable. Given these intrinsic fragilities, it seems important that the user should be able to maintain autonomous control over his or her linguistic skills, regardless of their extension by the prostheses. The article also stresses that the use of prostheses should be studied from the more specific angle of a loss of critical knowledge and the risk of new forms of control and influence made possible by their intermediation in linguistic flows. This more subtle form of proletarianization - a control of society through a control of linguistic devices - can be understood as the implementation by machines of linguistic politics of which there are examples in history. Given the nature of the entities that today control the most linguistic capital, forms of commercially motivated bias must also be considered. If these fears are well-founded, in this second perspective, the possibility of decoupling with the prosthesis is no longer a risk, but a possible escape from a control device. — Check also the other articles (some in French, some in English) of the issue curated by Alexandre Gefen & Philippe Huneman Philosophy of AI: Thinking and Writing with LLMs by Alexandre Gefen and Philippe Huneman : On the Semantics of Large Language Models by Martin Schüle The Human and the Mechanical: logos, Veridicality Judgment, and GPT Models by Anastasia Giannakidou and Alda Mari About a Pseudo-Knowledge: Large Language Models and the Replika Hypothesis by Philippe Huneman Knowing with Large Language Models: a Paradigmatic Break by Stéphan-Eloïse Gras and Gaël Varoquaux Generative Grammars, Minimal Operations and LLMs: True and False Problems by Pierre-Yves Modicom The artificial image: Portrait of Edmond de Belamy by Alexandre Gefen Artificial Agency and Large Language Models by Maud van Lier and Gorka Muñoz-Gil Playing with à Chat(GPT)? the Ludic Side of Appropriating AI Technologies by Lionel Obadia To Understand AI Sentience, First Understand it in Animals by Jonathan Birch and Kristin Andrews

Lire Nicolas, écrire pour être lu par Nicolas, écrire avec Nicolas, écrire sur Nicolas, écrire sans Nicolas.

Generation of LOD4 models for buildings towards the automated 3D modeling of BIMs and digital twins. New article by Bryan German Pantoja-Rosero, Alexander Rusnak, Katrin Beyer and myself. One step closer to the Mirror World.

Kazuo Sejima architect of the Rolex Learning Center receives a Dr Honoris Causa from EPFL and explains that SANAA’s intention was to create for students « a new type of learning experience not only by books and teachers but also by each others » @EPFL_en

AI is about to profoundly change the way we build 3D models and manipulate them. New paper : "Automatic Removal of Non-Architectural Elements in 3D Models of Historic Buildings with Language Embedded Radiance Fields" by Alexander Rusnak, Bryan G. Pantoja-Rosero, Katrin Beyer and myself. The article explores how language-embedded radiance fields can be used for the automatic generation of point clouds from images and for the removal of non-structural elements from those point clouds. This research enables faster, simpler modeling of the interior of buildings. The research is a collaboration between the EPFL's Earthquake Engineering and Structural Dynamics Laboratory (EESD) and Digital Humanities Laboratory (DHLAB), supported by the EPFL Center for Imaging. Read the paper here :

Today, we launch "H". You can read it online or order the print version for free. “H” is first and foremost about the future of the Humanities. And, specifically for this first edition, H refers to two major cross-disciplinary transformations in the field of knowledge: "Hyperscience" and "Hyperreality". "Hyperscience". Over the next twenty years, a hyperscientific core of unified methods and knowledge will undoubtedly emerge, including the humanities and social sciences, and with research into the “science of science”. Transdisciplinarity is becoming the norm rather than the exception, recomposing a fluid and continuous knowledge landscape far removed from the old silos, forging a fertile ground for innovation and discovery. “Hyperreality”. At the same time, we are entering the regime of “Hyperreality”, a territory in which the boundaries between the real and the virtual, documentary and fiction, are blurring. Hyperreality is characterized by the gradual obsolescence of truth and falsity as foundational concepts of discourse, a continuum between the natural and the artificial, a normalization of simulation as an experience of the world, and a new realm of creative expression, a canvas onto which the future can be painted. The generation born at the dawn of the year 2000 is the one that is experiencing these two convergences. If they are introduced to these new dynamics early enough, they will not only participate in the existing global game; they may well rewrite its rules. You can read my full interview about H here :

I am participating in a discussion at the "The Future of History" event at Stanford today with @petrpridal @eltonteb and @DavidRumseyMaps . The new TimeMap/OldMapsOnline browser will be demonstrated. It is free but you must register

RT @anniehartley_: ⏰ New state-of-the-art LLM for medicine within 24h of the #Llama3 release Meet Llama-3[8B]#Meditron: The first fine-tun…

"Machine Learning for Ancient Language" is gaining traction. Sources from the inception of writing systems in ancient Mesopotamia and Egypt (3400 BCE) to the late 1st millennium CE are becoming the new precious datasets to explore.

For the 2024/25 EPFL CDH Artist-in-Residence program "Enter the Hyper-Scientific," five outstanding artists have been selected from 200+ applications: Alice Bucknell, Josua Putzke, Sahej Rahel, and the duo Matthew C. Wilson & Emilia Tapprest. Check their previous work!

What is the secret sauce ? « A lot of work is having clean data » Timothée Lacroix @MistralAI’s CTO

Meditron 70B is an example of Humanitarian foundational model. It is open source, accessible, transparent, local and has no dependency. It was trained on 128 GPUs for 13 days on EPFL clusters. It shows remarkable performances on benchmarks and was also challenged by doctors with adversarial questions. No commercial interest to stay neutral. @anniehartley_ #AMLDEPFL2024

The goal of @Replit is to build « the first native OS for AI ». Doing so one of the most valuable dataset in the world is being created with the code flux generated by million of programmers.

A l’EPFL, nous travaillons activement sur tout le patrimoine du Haut-Karabakh, notamment le patrimoine épigraphique qui est le plus susceptible d’être altéré dans un objectif de réécriture de l’histoire. Nous avons probablement aujourd’hui une des bases d’images numériques les plus importantes au monde sur ce sujet. Nous sommes à la recherche d’alliances pour valoriser cette documentation et protéger ce patrimoine de possibles destructions ou transformations.

RT @karch_tristan: I am thrilled to join the Digital Humanities Lab ( as a postdoc this month, working with @freder…

RT @RPetitpierre: Excited to announce the publication of the first paper of my doctoral research, exploring 350 years of visual evolution i…