Piero Bosio Social Web Site Personale

@tezoatlipoca @metin

Language models cannot reliably distinguish belief from knowledge and factAbstract-----------«As language models (LMs) increasingly infiltrate into high-stakes domains such as law, medicine, journalism and science, their ability to distinguish belief from knowledge, and fact from fiction, becomes imperative. Failure to make such distinctions can mislead diagnoses, distort judicial judgments and amplify misinformation. Here we evaluate 24 cutting-edge LMs using a new KaBLE benchmark of 13,000 questions across 13 epistemic tasks. Our findings reveal crucial limitations. In particular, all models tested systematically fail to acknowledge first-person false beliefs, with GPT-4o dropping from 98.2% to 64.4% accuracy and DeepSeek R1 plummeting from over 90% to 14.4%. Further, models process third-person false beliefs with substantially higher accuracy (95% for newer models; 79% for older ones) than first-person false beliefs (62.6% for newer; 52.5% for older), revealing a troubling attribution bias. We also find that, while recent models show competence in recursive knowledge tasks, they still rely on inconsistent reasoning strategies, suggesting superficial pattern matching rather than robust epistemic understanding. Most models lack a robust understanding of the factive nature of knowledge, that knowledge inherently requires truth. These limitations necessitate urgent improvements before deploying LMs in high-stakes domains where epistemic distinctions are crucial.»#ai #LLMs #epistemology #knowledgehttps://www.nature.com/articles/s42256-025-01113-8

Federating knowledge: exploring ways to bridge wikis and notesJoin the workshop at #39C3! NEW DATE: Day 4, 13:40 @ Free Knowledge Habitat Workshop Area.Most people and organisations have their very own way of acquiring, organising, archiving, sharing, and collaborating on knowledge repositories. A broad spectrum of opinions and approaches resulted in a diverse and rich ecosystem of knowledge management solutions. Nevertheless, this also implies scattered and disconnected knowledge sources. What would it mean to build bridges among wikis and federate knowledge?This workshop is going to be heavily centred on a twofold discussion, exploring the challenge of federated knowledge starting from two questions.What does it mean to federate knowledge repositories?Instead of pursuing a silver-bullet solution to embrace all use-cases, what would it mean to foster and enable interoperability for different software?These questions stem from years of questioning and wondering how to integrate my personal note-taking and collective, participatory knowledge management at work, in organisations, institutions, and informal collectives. Recently, I began actively researching this topic as I started playing with the MediaWiki API to cross-synchronise my local Markdown notes and the XPUB wiki, the public learning wiki of the Experimental Publishing master. I am puzzled by taking advantage of the potential of a specific software (in this case, MediaWiki) while fearing of being locked-in.Some further, more specific, insights and questions:Local-first approaches and software (e.g. Reflection)Interesting experiments based on existing protocols, such as IbisWhat do we take of semi-open and obscure yet very cool initiatives like AnytypeThe power and the limits of plain-text: how to enable collaboration on simple Markdown files and build on top of it, as Obsidian doesCc: @modal @p2panda @obsidian @wikimediaDE @dweb#knowledge #FreeKnowledge #wiki #MediaWiki #API #Obsidian #Anytype #Ibis #IbisWiki #Reflection #CCC #Federation #federatedKnowledge #docs #PKM #knowledgeManagement #personalKnowledgeManagement #collectiveKnowledgeManagement #DWeb #decentralization #ActivityPub