Reposted by Laurence DierickxEthan Mollick @emollick.bsky.social · 03/10/2026🎵 Wave of distillation 🎵(sung to the tune of the Pixies) US labs say Chinese labs distill their models. And Europe's new “sovereign model” is fine-tuned on data generated by… GLM and Qwen. 4619
Reposted by Laurence DierickxCambridge Skeptics @cambskeptics.bsky.social · 02/10/2026Within hours of the incident aboard a Tel Aviv-bound flight, social media had supplied ‘false flag’ claims, baseless accusations about the attacker’s nationality and a fake BBC report implicating Mossad and India. The investigation was still getting started. www.abc.net.au/news/2026-10...abc.net.auFake news and deepfakes about flydubai flight spreading onlineABC NEWS Verify examines and attempts to debunk some of the fake news that's spreading online following the flydubai flight. 043
Reposted by Laurence DierickxHypervisible @hypervisible.blacksky.app · 01/10/2026“More than any other digital service, my email is a window into my brain. Now Meta has it. If that's not enough, Muse also suggested I give it my bank accounts. No thanks. I didn't let the AI run wild on my computer hard drive, either.”bbc.com'Things may get ugly': Meta's new AI Muse is about to make the internet more annoyingSomeday we'll redesign the internet for tools like this. Until then, you're in for a wild ride. 1321560
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026A Resituation of User Experience: Focusing on Slow Design dl.acm.org/doi/full/10....dl.acm.orgA Resituation of User Experience: Focusing on Slow Design | Proceedings of the 43rd ACM International Conference on Design of CommunicationYou will be notified whenever a record that you have chosen has been cited. 000
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026The former may be in the scaffolding; the latter still includes an implicit trust in the model. Building that scaffolding is itself programming work. It changes where the programming happens and who gets to do it. 100
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026The scaffolding is where much of the trustworthiness comes from, but there’s a missing step here: the system built around the model still has to rely on the model. So I’d distinguish between where trustworthiness is engineered and where trust is required. 100
Reposted by Laurence DierickxIris van Rooij 💭 @irisvanrooij.bsky.social · 30/09/2026We present a typology of traps to avoid: 1. Believing that AI systems are minds 2. Believing that AI systems are theories 3. Believing that cognitive science can be automated. Learn to recognise and avoid these traps. Failure to avoid leads to numerous problems. 18626
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026Cette histoire est aussi très symptomatique newsmachines.beehiiv.com/p/how-a-smal...newsmachines.beehiiv.comHow a small AI lab in Norway built a vibe-coding ladder for non-coding journalistsPolaris Media Vest gave reporters who can’t code a way to build interactive tools for readers — and a clear line between the throwaway widget and the service worth maintaining. 010
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026Aujourd'hui, le vibe coding et les agents permet en tout cas de faire croire que les programmeurs sont devenus au mieux substituables et au pire obsolètes. Mais qui sera sollicité quand il y aura des problèmes de sécurité, par exemple ? 110
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026C'est une excellente question, je dirais sans doute à la BBC ou dans de grandes entreprises comparable. En ce qui concerne la Belgique, un journaliste tech interrogé pendant ma thèse a souligné que les journalistes s'étaient très souvent comportés comme des divas... 110
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026De-anthropomorphizing “AI”: From wishful mnemonics to accurate nomenclature firstmonday.org/ojs/index.ph...firstmonday.org De-anthropomorphizing “AI”: From wishful mnemonics to accurate nomenclature | First Monday 000
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026Anthropomorphising AI: Two Modes, Two Errors link.springer.com/article/10.1...link.springer.comAnthropomorphising AI: Two Modes, Two Errors - Philosophy & TechnologyWhen interacting with social AI systems (SAIs), we routinely speak of what they ‘believe’, ‘want’, or ‘know’. With some exceptions, philosophers tend to treat such anthropomorphism as a single phenome... 100
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026Anthropomorphism and AI hype link.springer.com/article/10.1...link.springer.comAnthropomorphism and AI hype - AI and EthicsAs humans, we have an innate tendency to ascribe human-like qualities to non-human entities. Whilst sometimes helpful, such anthropomorphic projections are often misleading. This commentary considers ... 100
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026About anthropomorphism Constructing AI power: anthropomorphising imaginaries of generative AI by AI companies link.springer.com/article/10.1...link.springer.comAnthropomorphism and AI hype - AI and EthicsAs humans, we have an innate tendency to ascribe human-like qualities to non-human entities. Whilst sometimes helpful, such anthropomorphic projections are often misleading. This commentary considers ... 100
Laurence Dierickx @ohmyshambles.bsky.social · 01/10/2026How the data journalist Marco Maas used AI to build a second brain newsmachines.beehiiv.com/p/how-the-da... At least two things in this story: anthropomorphisation (which is obviously misleading about what a software programme is) and substitution (coders and programmers are no longer welcome).newsmachines.beehiiv.comHow the data journalist Marco Maas used AI to build a second brainMarco's agent Marvin now runs large parts of his working life and learns from every mistake. What makes it trustworthy is not the language model but the rules and checks built around it. 310
Reposted by Laurence DierickxDan Nguyen @dancow.bsky.social · 30/09/2026after the nutgraf, the first 2,500 words of this is boilerplate about plagiarism and ethics that might as well have been from chatgpt. I'm not saying that it is, but if you're going to accuse someone of massive clear-cut plagiarism, you're obligated to start off with egregious specific examples 3242
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026L'utilisation de modèles génératifs instables dans un contexte de rédaction sous pression fait peser des risques considérables sur la qualité de l'info et donc sur la confiance du lectorat. Cette incroyable histoire est à lire sur Alter Echos : www.alterechos.be/sudinfo-quan...alterechos.beSudinfo – Quand l’IA écrit des articlesEn journalisme, l’IA est un outil formidable pour la retranscription, la traduction voire la synthèse des informations. Mais en faisant écrire les articles par l’intelligence artificielle, la rédactio... 000
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Mais que dire si cet argument ne servait qu'à masquer le véritable enjeu : la pression visant à produire deux fois plus de contenu avec les mêmes ressources ? 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Si les lecteurs ne peuvent pas distinguer quand l'IA a apporté une contribution majeure à un article, peut-on encore parler de transparence ? L'argument principal est que l'IA fera gagner du temps aux journalistes. 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Le responsable de la rédaction affirme que l'IA n'est qu'un simple assistant. Pourtant, il semblerait que les instructions internes demandent à l'IA de rédiger des articles complets, incluant titres, chapeaux et intertitres. 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Ce document couvre dix scénarios de reportage courants (interviews, communiqués de presse, etc.), il associe chaque situation à une instruction (ou « prompt ») pré-rédigée, prête à l'emploi pour une IA. L'objectif est que le système IA produise l'article ! 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Bien que cette pratique ne soit pas explicitement mentionnée dans le cadre de leur travail, elle soulève la question de savoir qui rédige réellement l'information. Tout part de documents reçu par les journalistes en avril et mai dernier. 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Ca se passe dans une grande rédaction régionale belge où des journalistes sont formés à intégrer des fichiers PDF, des communiqués de presse et des transcriptions d'entretiens dans des systèmes d'IA qui génèrent des articles. 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026Using unstable generative models when writing under pressure poses huge risks to verification, nuance and editorial quality. This story was reported in the latest issue of the Belgian magazine Alter Echos. www.alterechos.be/sudinfo-quan...alterechos.beSudinfo – Quand l’IA écrit des articlesEn journalisme, l’IA est un outil formidable pour la retranscription, la traduction voire la synthèse des informations. Mais en faisant écrire les articles par l’intelligence artificielle, la rédactio... 000
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026The main argument put forward by the newsroom’s manager is that AI will save time to journalists, allowing them to do more fieldwork. However, what about if this was just an argument that hides the real challenge: putting pressure to produce twice as much content with the same resources? 101
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026The newsroom manager says that AI is merely an assistant. However, it is reported that internal prompts ask AI to produce complete articles, including headlines, intros and subheads. If readers cannot tell when AI has made a significant contribution to a story, can that be considered transparent? 110
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026In April and May 2026, a Word document started to circulate among journalists. Covering ten everyday reporting scenarios, including interviews, press releases,etc. it pairs each scenario with a pre-written prompt ready for use with an AI system. The aim is for the AI to produce the article itself! 100
Laurence Dierickx @ohmyshambles.bsky.social · 30/09/2026At Belgium's largest regional newsroom, journalists are being trained to input PDFs, press releases and interview transcripts into AI systems that produce finished articles. While this is not disclosed as part of journalists' work, it raises the question of who is actually writing the news. 110
Reposted by Laurence Dierickxrasmuskleis.bsky.social @rasmuskleis.bsky.social · 29/09/2026How do people define what is and is not news? In a new article led by Tali Aharoni based on analysis of 8 chat-based focus groups and 162 in-depth interviews, we document how not all journalism is seen as news, and not everything people call “news” is journalistic www.tandfonline.com/doi/full/10....tandfonline.com“Having Fun and Learning Something”: How Users in Four Countries Define What News Is – and Is NotNews categorizations are embedded in daily media routines. While research often relies on professional and academic definitions, users may have different understandings of what constitutes “news.” ... 1195
Laurence Dierickx @ohmyshambles.bsky.social · 29/09/2026Ha ha je crois qu'il y a des réalités différentes selon les cas d'usage et que le problème principal vient sans doute de la non rentabilité de systèmes imparfaits qui posent autant de problèmes éthiques que de problèmes liés à la qualité de l'info numérique. 010
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Je travaille depuis quelques mois sur les moteurs de recherche LLMs dans un contexte journalistique. Ici, les ressources en accès libre ne sont pas les meilleures aussi parce que les contenus de qualité sont souvent derrière des paywalls. C'est aussi pour ça que la 1ere compétence est critique. 111
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Cela ne m'étonne pas tant, on ne peut pas bien bosser avec ce genre de système sans savoir exactement ce qu'il doit faire. Reste que pour un tas d'application grand public, la qualité est variable et il y a une différence entre s'en servir pour une lettre de recommandation et un tâche à forte VA. 010
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Ces bons vieux ML et NLP restent bien indiqués dans de nombreux cas de figure. Le danger des modèles de langage, est de croire que l'on peut tout faire et tout maitriser en n'ayant pas les fondamentaux du métier. 000
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Le hype c'est aussi de ne pas s'intégrer comme prévu dans le business car ça reste une affaire économique - et ça ne prend toujours pas partout, reste à surfer sur la vague à coups de prédictions souvent apocalyptiques. www.gartner.com/en/articles/...gartner.comWhen Not to Use GenAI: Choosing the Right AI TechniqueFind out when GenAI falls short and why alternative AI techniques may be a better fit for your business needs. 101
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Le problème est vraiment les cas d'application. Ici, par exemple, une excellente expérimentation (mais pas scalable pour des questions de ressource): Can AI Outsmart Fake News? Detecting Misinformation with AI Models in Real-Time journals.sagepub.com/doi/10.1177/...journals.sagepub.com 000
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Mais il faut distinguer les usages de mr et mme Toutlemonde des usages professionnels où des modèles sont intégrés dans des chaînes métier (et là on est dans le domaine de la gestion informatique). 200
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026Ca dépend sur quelles tâches. Dans ma recherche sur les prompts en fact-checking, les modèles ont clairement du mal quand il s'agit de faire dans la nuance. De plus, ils ne sont pas entraînés pour dire le vrai mais le plausible. 201
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026AI existential risk probabilities are (still) too unreliable to inform policy www.normaltech.ai/p/p-doomnormaltech.aiAI existential risk probabilities are (still) too unreliable to inform policyHow speculation gets laundered through pseudo-quantification 000
Laurence Dierickx @ohmyshambles.bsky.social · 28/09/2026🚨From normative principles to epistemic governance: A workflow-based analysis of European journalism ethics in datafied environments - 37 European ethics codes examined: principles hold, but data practices remain largely off the radar journals.sagepub.com/doi/10.1177/...journals.sagepub.comSage Journals: Discover world-class researchSubscription and open access journals from Sage, the world's leading independent academic publisher. 000
Reposted by Laurence DierickxJay Van Bavel, PhD @jayvanbavel.bsky.social · 27/09/2026Conspiracy Believers Are 91% Likely To Hold Incoherent Beliefs Can one believe Princess Di was murdered by and faked her own death? Both can’t be true. A new study N~9,000 finds that holding 2 incompatible ideas at once isn’t a quirk, it’s practically the rule. studyfinds.com/conspiracy-b...studyfinds.comConspiracy Believers Are 91% Likely To Hold Incoherent Beliefs, Study FindsA new study finds people at the extreme end of conspiracy thinking have a 91% predicted chance of holding incompatible beliefs. 33410
Laurence Dierickx @ohmyshambles.bsky.social · 27/09/2026Absolutely, and in this case it looks like Google Scholar just scrapes HTML citation meta tags, so any site that mimics a journal's markup gets indexed, and profiles auto-attach by name matching. The ORCID string on the fake page isn't parsed as a claim at all. It's there to persuade human readers. 100
Laurence Dierickx @ohmyshambles.bsky.social · 27/09/2026The scholar says that a fabricated article was published on the website of a fake journal, bearing his name and current affiliation, and that it was then indexed in Google Scholar. I don't know more about this case, but anyone can copy-paste an ORCID number on a web page. 100
Laurence Dierickx @ohmyshambles.bsky.social · 27/09/2026Si vous ne l'avez pas encore lu, cet article est particulièrement éclairant (j'ai un immense respect pour cet immense philosophe et chercheur) : papers.ssrn.com/sol3/papers....papers.ssrn.comWhy the AI Hype is Another Tech BubbleThis article argues that the current hype surrounding artificial intelligence (AI) exhibits characteristics of a tech bubble, based on parallels with five previ 101
Laurence Dierickx @ohmyshambles.bsky.social · 27/09/2026Pas tellement étonnant, et qu'est-il possible de prévoir avec ces technologies ? Il est même possible que ces gens ne voient même pas à la semaine... qui sait ? 000
Laurence Dierickx @ohmyshambles.bsky.social · 26/09/2026More insidiously, its title and topic mimic research he has actually done, which makes the false attribution all the more believable. The harm is more than personal. Once archived and indexed, a sham paper can look legitimate, eroding trust in attribution, accountability and the scholarly record. 021
Laurence Dierickx @ohmyshambles.bsky.social · 26/09/2026Academic ghost-authorship just got weirder. A researcher has revealed on LinkedIn that Google Scholar alerted him to a paper he never wrote, in a journal that doesn't exist, published under his name and ORCID number. The article itself is AI-generated nonsense with no scientific value. 241
Reposted by Laurence DierickxLe Cheikh @lecheikh.bsky.social · 26/09/2026Grâce à l’IA, l’avenir de la recherche s’annonce radieux… 🙃 03818
Reposted by Laurence DierickxSocial Media Lab @socialmedialab.ca · 26/09/2026"Oxford lets OpenAI train its AI models on Bodleian Library. University staff voice concerns over reputational risk of partnering with company behind ChatGPT." www.theguardian.com/technology/2...theguardian.comOxford lets OpenAI train its AI models on Bodleian LibraryUniversity staff voice concerns over reputational risk of partnering with company behind ChatGPT 188
Laurence Dierickx @ohmyshambles.bsky.social · 26/09/2026Les tech IA ne sont pas nécessairement une économie : McKinsey warns that AI may get a lot more expensive for companies www.businessinsider.com/mckinsey-ai-... Ca doit aussi jouer dans la balance (reconnaissant aussi que l'économie n'est pas mon domaine donc compétences limitées là-dessus).businessinsider.comMcKinsey warns that AI may get a lot more expensive for companiesMcKinsey warns companies that AI spending may rise further and be less predictable with adoption of agents 111