home.social

#provenance — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #provenance, aggregated by home.social.

  1. After scraping around 12 million artworks from Cara, the developer is now collaborating on Lantern, an open source tool helping artists track their work in public AI datasets. A fascinating turn in the debate over AI, provenance and control.

    #AIArt #Copyright #Provenance

    He Scraped All of Their Art fo...

  2. Museums need more than an AI policy: they need AI provenance. Decisions about tools should account for training data, artists’ rights, labour, environmental costs and vendor lock-in. Small, locally operated open models trained on licensed or public-domain collections could offer a path towards cultural data sovereignty.
    aam-us.org/2026/08/24/museums-
    #Museums #DigitalHeritage #CulturalData #provenance

  3. Last week someone shared with me the watermarks-remover GH repo. Working with C2Pa on just about every product I have this came as a fun challenge.

    How do I build a watermark that not only resists stripping attempts but makes them backfire on the stripper?

    Challenge accepted, received and closed.

    Introducing Provcheck Backfire

    provcheck.ai/backfire/

    #AI #watermark #EUAIACT #provenance

  4. В AI-картинках есть невидимые метки. Я проверил, как они работают и можно ли их удалить

    Вы генерируете картинку в нейросети, скачиваете PNG и открываете его. Никакого логотипа. Никакой надписи «создано ИИ». Никакой полупрозрачной плашки в углу. Обычное изображение. Но это ещё не значит, что внутри него ничего нет. Современные генераторы всё чаще добавляют в AI-изображения информацию о происхождении, которую человек глазами не увидит. Где-то это метаданные. Где-то криптографически подписанные Content Credentials. А где-то настоящий невидимый водяной знак , встроенный непосредственно в изображение. Например, Google использует SynthID, а OpenAI сейчас добавляет к поддерживаемым изображениям сразу два сигнала происхождения: C2PA и SynthID. Мне стало интересно, насколько эта маркировка действительно живучая. Можно ли определить, что изображение создано ИИ? Что именно нейросеть оставляет внутри файла? Исчезает ли маркировка после редактирования? И можно ли удалить невидимый водяной знак с AI-изображения обычным сервисом? Я попробовал несколько вариантов. Один действительно сработал. Но когда начал разбираться, понял, что вопрос «как удалить watermark с картинки» устроен гораздо сложнее, чем кажется.

    habr.com/ru/companies/syntx_ai

    #искусственный_интеллект #AIизображения #SynthID #C2PA #Content_Credentials #водяные_знаки #OpenAI #генерация_изображений #маркировка_AI #provenance

  5. Нужную книгу больше не обязательно искать: как ИИ меняет сам принцип работы с информацией

    Что, если нужную книгу больше не нужно искать — её можно собрать под собственный вопрос? Я проверил эту идею на реальном корпусе: 206 длинных видео, 160 часов 56 минут и 2,15 млн слов транскриптов . Вместо сотен summary я построил конвейер, который выделяет сущности, схлопывает дубли, отслеживает происхождение выводов, проверяет важные утверждения и понимает, когда оставшиеся источники уже не нужно читать подряд. По ходу выяснилось, почему 9 URL могут оказаться одним источником , почему векторная кластеризация не решила задачу смысловых дублей, чем такой подход отличается от RAG и почему длинная работа ИИ-агента неожиданно начинает напоминать разработку обычной программной системы. Но главный вывод оказался шире самого проекта: единицей работы с информацией может становиться не источник, а наш вопрос.

    habr.com/ru/companies/ods/arti

    #искусственный_интеллект #AIагенты #LLM #RAG #provenance #база_знаний #дедупликация #обработка_информации #управление_знаниями #fact_checking

  6. The Quotation’s Chain of Custody

    A mug sits on the desk with Marilyn Monroe's face under the sentence Laurel Thatcher Ulrich wrote: “Well-behaved women seldom make history.” The object is wrong at the name and prone to wording drift, and it keeps selling because the right sentiment can carry any name. The claim riding beside it has manners: why care whether the mug gives Monroe a line written by Ulrich, or whether a poster hands Voltaire a sentence built by Evelyn Beatrice Hall, so long as the line still stirs? Its appeal points to use over receipt. Its conclusion says the words have done good work in the world, and good work should settle the bill. […]

    bolesblogs.com/2026/08/14/the-

  7. 📢 Submit your research to swat4hcls.org by 14.Sep.2026 at app.oxfordabstracts.com/stages

    Accepted papers will have the chance to submit an extended version to the conference collection at the Journal of Biomedical Semantics

    Topics around for and

  8. #Ressource | Library of Yeshivas Chachmei Lublin. Digital catalogue and provenance database of the prewar yeshiva library | Teatr NN | 🔗 Details:jewishstudies.de/de/nachrichte

    #jewishstudies #Provenance

  9. [Edited, twice, for typos]
    [No AI was used in writing this toot — I’d rather not write at all otherwise]

    The use of #GenAISlop machines to determine whether text was provided by #LLM #TextRegurgitatingMachines is a monster of a cyclical process. Like the integrity of police officers overseen by other police officers in another branch of law enforcement. Lawyer oversighted by a professional lawyer body… and so on. #Credibility and #Transparency are unlikely to prevail in such nepotistic circumstances, corruption is everywhere but a step away.

    What are #Authors to do? Build a #TrailOfEvidence and unimpeachable record of #Provenance for every work they wish to publish? How are they to do this? Are there non-AI tools to help? This is a really f*^%#d up world the #TechBros are imposing upon us all with their inane #MarketingHype hitting the mark with many incompetant and gullible executives in charge of the #publishingIndustry. (This is all conjecture on my part based on very little but general observations of what LLM GenAI is doing — or not doing — in every field of endeavour).

    “Pangram is not infallible. I’ve used it to run some of my newsletters through it, and they came back as 80% AI when I wrote every word myself. This is one of many reasons why I disagree with Substack’s decision to partner with Pangram. Another reason is that once a tool like Pangram is available to the masses, false accusations start flying around, and careers can be ruined. Whether you admit it or not, there is A LOT of jealousy in the publishing world, so it wouldn’t surprise me if someone who disliked author X decided to use Pangram on their work and made the results public. This is not the publishing world I want to live in.” (Read more: AI Accusations and Book Publishing - Kathleen Schmidt, Aug 01, 2026). A blog post on Substack, sry. kathleenschmidt.substack.com )

    #Writing #WritingCommunity #Publishing #SlefPublishing #AuthorshipProvenance #Author #Writer #Disclosure #ProofOfAuthorship

    PS. I wonder how many *airport* books by some of the most successful authors in the *genre* would pass muster if subjected to #Pangram or similar GenAI tool generated analysis? Not many I’d wager. Further, what’s the bet Pangram is itself code produced by a GenAISlop machine. This AI ponzie scheme is just getting wilder and wilder by the day… BOOOMMM !

  10. New on the RSpace blog: research documentation in the age of AI 🧪🤖

    Our colleague Tilo reflects on a question we hear a lot: how do you let an AI agent act on your research and still retrace what it did?
    In RSpace you can treat an agent as a team member, with its own account, so everything it touches stays permissioned, versioned and in the audit trail.

    Read more and how to get started here: researchspace.com/blog/researc

    #ResearchDataManagement #OpenScience #FAIR #Provenance #OpenSource

  11. Zendaya’s earrings at the *Odyssey* London photocall have reignited debate over who should own ancient artefacts—reportedly gold from Iran’s Ziwiye hoard. Academics and activists urge Barron London to disclose full provenance and legality. 🌍🏛️ #Antiquities #CulturalHeritage #Provenance #TheOdyssey theguardian.com/culture/2026/j

  12. Die EU hat einen Verhaltenskodex zur Kennzeichnung KI-generierter Inhalte bestätigt. Ein Label ist aber noch keine Herkunftskette. Ohne maschinenlesbare Provenance, offene Standards und robuste Weitergabe zwischen Plattformen wird Transparenz beim ersten Export zur Dekoration. #AIAct #KI #Provenance #Standards #Digitalpolitik #EU digital-strategy.ec.europa.eu/

  13. The Late Edit: How Belief Turns Revision Into Revelation

    We have become a provenance culture, and I count myself an enforcer of it. Galleries now label machine-made images. Legislatures draft watermark bills, publishers add disclosure lines to copyright pages, and coders audit the licenses of every scraped repository. I am a plaintiff in litigation over the corpora used to train commercial language machines, so my commitment to the chain of custody is on file in federal court. The question a serious reader asks of any artifact in 2026 is blunt: who made this, from what, and who touched it on the way to me? That question is healthy. It is also, in one enormous and telling case, suspended. […]

    bolesblogs.com/2026/07/05/the-

  14. With the popularity of generative AI, it's becoming more and more difficult to distinguish reality from fiction. Can this problem be solved using cryptography? What are the privacy implications of the currently proposed systems?

    privacyguides.org/articles/202

    #AI #ContentCredentials #Privacy #Adobe #DigitalProvenance #Provenance #PrivacyGuides #Article #Opinion

  15. After #plw2024 I've found my #provenance rabbit hole - linking #wikidata items of people related to Achim's #loviscorinth #paintings - collectors, gallery owners ... a world of family connections! That said, you inevitably also add one Holocaust victim after another and learn more about them. Some information only exists because of #terezin #ushmm & other databases. Still hoping to find an unidentified collector so I can do my part to give a person a face – not just the "Holocaust victim" label.