TimboDyne @TimboDyne

Récemment, autour d’une table, on parlait des haters. Des commentaires. Des attaques. De ce que ça fait vraiment, une fois la notification fermée. Un grand homme m’avait dit un truc simple. N’accorde jamais de crédit aux commentaires positifs. Les négatifs ne te toucheront plus. Au début, ça sonne froid. Puis X a accéléré. Quasiment 1 500 messages privés de soutien. Et, tous les jours, des gens qui viennent m’attaquer. Les deux existent. Un seul des deux mérite d’être traité comme une information. Je n’accorde de la valeur qu’aux idées. Un commentaire négatif sur Internet, pour moi, c’est le clochard bourré au coin de la rue qui t’insulte pendant que tu marches. Tu n’arrêtes pas ta journée pour débattre avec lui. Tu ne réorganises pas ta vie autour de sa phrase. Tu continues. Le soutien, c’est de la chaleur humaine. Pas une preuve que tu as raison. L’attaque, c’est du bruit. Pas un verdict. Les rageux, souvent, ne sont pas des ennemis. Beaucoup sont ignorants. Beaucoup sont dans une détresse assez profonde. Une mauvaise idée les tient. Elle leur offre un camp, un coupable, une explication gratuite à ce qui ne va pas. Ça ne rend pas l’insulte vraie. Ça rend la personne tragique. Le seul tribunal qui m’intéresse, c’est celui-ci : l’idée tient-elle ? Est-ce qu’elle décrit le réel mieux que la précédente ? Est-ce qu’elle aide à construire ? Le reste, c’est la rue. Allez construire.

USA 🇺🇸 vs EURope 🇪🇺

Les gens n’ont pas compris ce qui est en train de se passer. On nous a vendu pendant trente ans que la sécularisation était inévitable. Scientifique. Définitive. Que la religion était le stade infantile de l’humanité, et que l’Occident allait enfin raisonner sans superstition. Les 18-25 ans n’ont pas reçu le mémo. En France, les baptêmes d’adultes ont plus que triplé en dix ans. 4 100 en 2016. 13 200 à Pâques 2026. Quarante-deux pour cent ont entre 18 et 25 ans. Ce n’est pas un sondage d’ambiance. Ce sont les chiffres de la Conférence des évêques. Aux États-Unis, Gallup : la part des jeunes hommes qui disent que la religion est « très importante » est passée de 28 % à 42 % en deux ans. Leur fréquentation mensuelle est au plus haut depuis 2012. Barna mesure un rebond de l’engagement à Jésus, tiré par les plus jeunes. Les ventes de Bibles explosent. +22 % aux États-Unis, plus haut niveau en vingt-et-un ans. +87 % au Royaume-Uni depuis 2019. En France, les librairies parlent d’un « boom biblique ». Ce n’est pas un grand réveil achevé. C’est un signal. Et les élites font comme s’il n’existait pas. Parce qu’elles n’ont pas compris une chose simple. La civilisation occidentale, ce n’est pas d’abord le marché, ni la démocratie, ni même la science. C’est un logiciel. Un système d’exploitation. Ce logiciel, c’est le christianisme. Pas « une religion parmi d’autres ». Le Logos. L’idée que le réel est lisible. Que la vérité existe, et qu’elle n’est pas un rapport de force. Que chaque individu a une dignité qui ne dépend pas de sa tribu, de son sexe ou de son camp. Que la raison et la foi ne sont pas ennemies — qu’elles tiennent ensemble. On a désinstallé l’OS en croyant libérer de l’espace. Puis on s’est étonné que les applications plantent. Le droit. Le débat. L’université. La science elle-même, dès qu’elle devient un récit. La recherche de la vérité remplacée par la recherche du camp. Les jeunes ne « redeviennent pas bigots ». Ils ont grandi dans le vide qu’on leur a vendu comme une émancipation. Anxiété. Nihilisme. Idéologies qui occupent la place laissée vacante. Identité. Décroissance. Relativisme. Ce ne sont pas de mauvaises personnes. Ce sont de mauvaises idées. Et les mauvaises idées détruisent d’abord ceux qui y croient. Le retour du christianisme n’est pas un retour en arrière. C’est une génération qui cherche un cadre assez solide pour reconstruire. Un cadre où la vérité n’est pas négociable. Où la raison a un sol. Où l’on peut de nouveau bâtir des systèmes qui tiennent. Si l’Occident se redresse, ce ne sera pas par un ministère. Ce sera par des gens qui réinstallent le logiciel — et qui construisent par-dessus. La reconversion est possible. x.com/TRobinsonNewEr…

You asked for results, we made the changes necessary for this FBI to deliver- one example, DNA: We are statutorily required under the “DNA Fingerprint Act of 2005” to receive and process collected DNA samples from federally convicted offenders, arrestees, and non-U.S. citizen detainees. These samples are required to be sent to the FBI Laboratory Division (LD) for processing – we get around 50k a month. This literally solves crimes across the country, like unsolved homicides, mass shootings, sexual assaults, and other violent offenses. There had been a massive backlog at the FBI Lab for decades. On August 1, 2024, the pending DNA backlog for these samples was over 1.81 million, literally breaking federal statutory requirements. A typical example of institutionalized bureaucracy from prior leadership that left crimes across America unsolved. When we got here, we changed it. This FBI dedicated more resources, more equipment, improved IT systems, and used business analytics for DNA processing. TODAY - that DNA backlog is officially at ZERO. Just more evidence of the monumental and necessary changes made by this FBI. Thank you, FBI Lab Division. -DKP🇺🇸

Voici pourquoi, malgré l’absence de business model qui peut tenir dans le temps, Anthropic veut IPO. Ils n’ont que trois à six mois d’avance technique. Quand on wrap une commodité, les marges s’effondrent. C’est une loi, pas une opinion. Leur chiffre d’affaires peut retomber aussi vite qu’il est monté. Et c’est précisément pour ça qu’ils pressent l’introduction en Bourse. C’est extrêmement dangereux. Mais mon vrai malaise est ailleurs. Les révélations de ces derniers jours sur Cami Clark, la femme de Dario Amodei, sont troublantes. Les emails des Epstein files montrent qu’elle a sollicité Jeffrey Epstein pour investir dans sa société de « free luxury porn » destinée aux femmes. Elle l’a rencontré, lui a envoyé des scripts, lui a demandé de l’argent. Il n’y a aucune preuve qu’il ait investi. Mais le simple fait que ça existe, et qu’elle soit aujourd’hui présentée comme sounding board et conseillère stratégique d’Anthropic, laisse un goût bizarre. Ce mec-là est dans la lignée directe de SBF. Effective altruism. Un mouvement qui, sous des dehors de rationalité et de bienveillance quantifiée, ressemble de plus en plus à un néo-communisme soft : une élite morale qui décide pour les autres ce qui maximise « le bien », avec des tableaux Excel à la place du Gosplan. Anthropic, pour moi, incarne l’ancien monde. Celui du contrôle et des classes. Je pense que Dario veut créer une hyper-classe fondée sur l’IA. Une caste qui possède le code, les garde-fous, les permissions, et qui décide qui a le droit de jouer. Si cette vision gagne, on aura Altered Carbon ou 1984. Un monde verrouillé, hiérarchisé, où l’intelligence est gérée comme un privilège. Cette boîte, pour moi, est dangereuse. La vraie promesse de l’IA n’est pas de créer une nouvelle aristocratie de la sécurité. Elle est de rendre chaque individu plus capable, plus libre, plus productif. Les bâtisseurs horizontaux, ceux qui utilisent ces outils pour créer des choses qui n’existaient pas hier, finiront par l’emporter. Choisis ton camp.

I’m with many in disliking the AI watermarking. I write a lot of my own stuff, and I’ve enjoyed having it edit my work. The idea that it will both claim credit and make my work traceable is repugnant to me. Especially since they stole humanity’s knowledge to get to this point.

@j4rddvhfkv @lady_valor_07 @tedyhawgeddy Ive added it as a secret ingredient for years and nobody knew it was there and think my french toast is amazing. They don’t know what it is about my recipe… but “damn is this good!”

This is exactly why Anthropic pisses me off. You can keep saying the watermark doesn’t change ownership, doesn’t identify the user, doesn’t affect quality, doesn’t add hidden characters, and won’t look any different to readers. Fine. But that is not the whole issue. The issue is that you are still marking work that a human being may have thought through, structured, argued, and emotionally meant, just because AI helped process it. AI assistance is not authorship. If I wrote the idea, the argument, the meaning, the emotional core, and the final judgment, then it is my work. If Claude helped translate it, polish a sentence, or make it clearer, that does not suddenly make Claude the author. But a watermark absolutely changes how people perceive authorship. People are not going to say, “Claude may have been involved in processing this text.” They are going to say, “AI wrote this.” And honestly, who the hell are you to do that to my writing? Who gave you the right to stamp your model’s involvement onto work that came from my thoughts, my experience, my voice, and my judgment? You can say it does not change ownership all you want, but socially, reputationally, and creatively, it absolutely changes how the work is read. So fine. Do whatever you want. Build your watermarking system. Follow whatever framework you want. I’m not using Claude. Good luck.

Anthropic @AnthropicAI

a month ago

We’ve written an FAQ to answer some of the questions we've received about watermarking. In summary: • We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking;

2K 679 5K 11.3M 3K

@jordanurbs I canceled mine yesterday, and wont go back unless this policy is reversed.

Facts for the fake, America-hating news…

Acting Secretary of the Navy Hung Cao @SECNAV

a month ago

x.com/i/article/2088…

855 2K 11K 3.4M 497

x.com/i/article/2088…

@AnthropicAI If all those things were true then you wouldn’t be complying with the EU AI Act, as the information injected by the watermarking process would be a null set. You aren’t listening to your customers nor ex-customers.

Sarah O’Connor would say this.

Everybody loves @MarkKnopfler. It's worth thinking about why. It only starts with the fact that he's an amazing guitarist and songwriter who's been pouring his soul into recordings for nearly 50 years. I think it's because he's not fake, doesn't project a huge ego, and doesn't posture as anything he isn't. With Mark, what you see is what you get - closely observed slices of life through a lyrical and bluesy sensibility. There's an understated soulfulness about him that is deeply tasteful. He's all about the music. And the music is beautiful. Thank you, Mark, from all of us for whom your work has been an important part of the soundtrack of our lives. You're a good man, and it shows.

🎸 Rock History 🎸 @historyrock_

2 months ago

Happy 77th birthday to the legendary guitarist and singer of Dire Straits, Mark Knopfler!

152 875 8K 151K 72

"Undetectable" means you can't tell it's there. It does NOT mean nothing changed. Every watermarking scheme on Earth works the same way. It basically reaches into the sampler and nudges token selection away from what the model would have freely chosen. In open-ended prose where there are five ways to say a sentence, okay. The watermark picks one and great. Who gives a shit, right? Now do it in code. In code, the next token is not a choice. It's a consequence. One variable name. One operator. One exact API call, dictated by the logic surrounding it. When there's only one correct token and the watermark says "pick the second-best one," you don't get a stylistic quirk. You get a fucking bug. Or a freaking security hole that ships to prod. Now, of course, AI already makes lots of mistakes. So why mandate that they make more? And how about legal text? Even worse. Entire court cases have turned on the placement of a single comma. Ask the company that lost millions over an Oxford comma in a Maine overtime statute. O'Connor v. Oakhurst Dairy, 851 F.3d 69 (1st Cir. 2017), the most famous punctuation case in modern American law. Maine's overtime statute exempted workers involved in "the canning, processing, preserving, freezing, drying, marketing, storing, packing for shipment or distribution of" perishable foods. The question in the case was "packing for shipment" or "distribution" a single exempt activity, or are they two separate ones? There was no Oxford comma after "shipment." So Oakhurst Dairy argued "packing for shipment" and "distribution" were separate items in the list, meaning their delivery drivers, who distributed but didn't pack, were exempt from overtime. The drivers argued the list described one activity: "packing, whether for shipment or distribution," which they didn't do, so they were owed overtime. Judge David Barron opened his opinion for the unanimous First Circuit panel with the line that became instant legal legend: "For want of a comma, we have this case." Now imagine a compliance-mandated dice roll nudging "shall" toward "may" because the pseudorandom key felt spicy that day. The watermarking literature knows this, by the way. There's an entire subfield, entropy-filtered watermarking, that exists specifically because watermarking low-entropy text naively breaks things. They've been building escape hatches for exactly this failure mode for a generation. But those escape hatches create watermark-free zones, and the whole things turns into Swiss cheese. So who does the watermark actually catch? The lazy and the honest. Students copy-pasting homework. Bots too dumb to paraphrase. Meanwhile, the actual adversaries, the propaganda farms, the hackers, the crypto scammers, the astroturfers, the people the law was written to stop, route around it before lunch. The watermark is a lock that keeps out your grandma. It disappears when heavily edited, or paraphrased, or too short. On and on and on. That's not a provenance system. It's magical thinking and it's compliance theater.

Jonas Geiping @jonasgeiping

2 months ago

Why am I being baited by watermark misinformation on this app, is it 2023 again? A small FAQ: 1. 𝗪𝗵𝗮𝘁'𝘀 𝗮 𝘁𝗲𝘅𝘁 𝘄𝗮𝘁𝗲𝗿𝗺𝗮𝗿𝗸? -- A modification of the LLM sampling algorithm that, if there are multiple ways to write something, will pick one that agrees with a

75 116 724 117K 491

To assume a negligible or even crucial creative non-involvement of a human when a watermark is detected is quite incorrect. All this does is throw shade at someones creative work if there is any AI involvement in the human workflow creative processes. It is a way to get people to steer clear of using AI which is a quantitative and qualitative advantage unless you build out your own LLM device which doesn’t watermark. Another elite manipulation and an anti-civilizational one. I wonder if those that are content with watermarks actually use AI and are also creators?

I’ve been thinking more about AI watermarking, and what bothers me isn’t really the technology itself. It’s the social behavior that grows around it. Over the past few days, I’ve already had people tag AI detectors like Pangram under some of my posts. In some cases the result was wrong, but that barely mattered. The moment “AI detected” appeared, people were ready to dismiss the post, question my credibility, or just call me garbage. That experience changed how I look at watermarking. The problem is that people rarely stop to ask what the AI actually did. Did it translate something? Fix grammar? Improve wording? Help structure an argument? Or generate the entire thing from scratch? Those are completely different levels of involvement, but socially they can collapse into one crude label: “AI wrote this.” And that worries me, because the label can become more important than the actual content. I’m not against provenance or transparency. I just think we’re creating tools faster than people are learning how to interpret them responsibly. “AI was involved” should never automatically mean “the human contributed nothing.”

Here is my problem with this proposal for Claude. I have roughly 10,000 pages of my written work that is pre-AI, from books I have written, articles I have written, and all kinds of other material. Here is my working method. I will be working on a new article and say, “Hey, check all the stuff I wrote before and see if I have some good quotes that we can work into what I am writing today. Give me some of my best quotes.” I have a database set up for this, so it goes in and pulls out some of my best quotes. I look at them and think, “Oh yeah, that is pretty good. I can reuse that. I can reuse this other thing too.” I am not plagiarizing. It is my own work. Then I start thinking about whether I have any new thoughts on the idea. I will dictate some things and ask, “Is that new, or did I already say that?” Sometimes I repeat myself, because there is nothing wrong with that. But sometimes I have an extension of something I have been thinking about in the past, a new take or angle to the idea. Or I have learned new things about the topic. Now I want to combine my previous arguments with this new argument. So I will have the AI lay out the things I have said before alongside the things I dictated this time. Then I will say, “All right, let us put these in order so they make the final article.” That is the kind of thing I use AI for. Sometimes I write in too enormous paragraphs, so I might ask it to break those into smaller paragraphs that are easier to read on screens. Maybe there is an odd transition here or there. Maybe I need to change something from “you” to “they,” or from “us” to “you,” depending on the grammar of the piece. I might say, “We need to harmonize this grammar so it stays in the third person throughout.” Those are useful editorial functions. I am still writing the material. The quotes are mine. The arguments are mine. The new dictation is mine. I am using the AI to help retrieve, arrange, compare, and clean up what I have already written. According to the research I did, which admittedly was not that deep, I did not read the actual studies or Anthropic’s promises in detail, they are going to change what I would have written, my quotes, my dictation, and so forth, just a little bit. Just enough that they think it becomes "theirs". I do not like that. I do not want them to change it. I carefully crafted it the way I crafted it. I want the tool to help me work with my writing, not subtly rewrite it into something else.

NIK @ns123abc

2 months ago

🚨 JUST IN: Claude models will now have invisible watermarks embedded in ALL text, and ALL metadata attached to files…

keyboard_arrow_left Previous keyboard_arrow_right Next

1K 3K 30K 13.8M 14K

@___Patrice___ It is authentic, as this is the identical process for all creative work. Anyone with creative experience with AI knows without the human the end result is trash (not fit for purpose). Even the purpose / original idea has a human origin.