AI companies are scraping content from the Wayback Machine for AI training. So if your website isn’t blocking the ia_archiver crawler, your work is being used to train AI models. Reddit is now blocking the Internet Archive because of this. Many others will do the same.

This is, without exaggerating, one of the most extraordinary things a US Treasury Secretary ever said. It should be mandatory viewing for all citizens of US "allies", Europeans first and foremost. What Bessent is saying is that the US will now treat US allies' wealth as an

Capital Account: More people turn to ChatGPT and other large language models for answers, but they don’t add to the stock of knowledge on.wsj.com/4nfNxpR

(🧵) Today, we release Meta Code World Model (CWM), a 32-billion-parameter dense LLM that enables novel research on improving code generation through agentic reasoning and planning with world models. ai.meta.com/research/publi…

We trained the whole stack. 𝗣𝗿𝗲𝘁𝗿𝗮𝗶𝗻. 𝗦𝗙𝗧. 𝗥𝗟. 𝗢𝗽𝗲𝗻 𝘄𝗲𝗶𝗴𝗵𝘁𝘀. 𝗢𝗽𝗲𝗻 𝗺𝗲𝘁𝗵𝗼𝗱𝘀. 𝗢𝗽𝗲𝗻 𝘀𝗰𝗶𝗲𝗻𝗰𝗲. From tokens to traces. From guesses to grounded. 𝗖𝗼𝗱𝗲 𝗪𝗼𝗿𝗹𝗱 𝗠𝗼𝗱𝗲𝗹 is here 50+ pages Report. Ckpts. Code. ai.meta.com/research/publi…

🪩The one and only State of AI 2025 is live! 🪩 It’s been a monumental 12 months for AI. Our 8th annual report is the most comprehensive it's ever been, covering what you *need* to know about research, industry, politics, safety and our new usage data. My highlight reel:

One Tuesday night, as my wife and I sat down for dinner, a sheriff’s deputy knocked on the door to serve me a subpoena from OpenAI. I held back on talking about it because I didn't want to distract from SB 53, but Newsom just signed the bill so... here's what happened: 🧵

Dans le projet de loi de finances que je viens de consulter, tout a été fait pour épargner Bernard Arnault et les milliardaires français. C’est donc aux autres catégories sociales qu’il est demandé de faire des efforts. Une fois de plus🧵

Paru dans l'indifférence l'année dernière, un rapport de l’Inspection Générale des Finances dresse un portrait sans complaisance du cinéma français — une industrie où l’abondance de subventions a fini par étouffer l’exigence artistique et la logique économique...⏬ Ainsi près de

Time to guillotine France’s super-rich tax breaks on.ft.com/4hb3uMd | opinion

We’re announcing a research collaboration with Commonwealth Fusion Systems, one of the world’s leading nuclear fusion companies. Together, we’re helping speed up the development of clean, safe, limitless fusion power with AI. ⚛️

Les milliardaires ne paient pas ou presque pas d’impôt sur le revenu — et les macronistes viennent de faire une première alliance avec le Rassemblement national pour maintenir intact leur privilège fiscal 🧵

En 1986, le patrimoine d’un retraité de 70 ans représentait 36 % du patrimoine d’un actif de 55 ans. Aujourd’hui, un boomer possède 118 % du patrimoine d’un actif de 55 ans.

L'effet de ces différentes propositions peut désormais être étudié grâce au formidable simulateur de @Bplanterose, chercheur au EU Tax Observatory, qui permet de quantifier les recettes fiscales des différents projets d'imposition des ultra-riches ⬇️⬇️⬇️ zucman-simulator.pages.dev

Les politiciens ne pensent qu'à leur carrière immédiate. Les militants des partis ne pensent qu'à définir ce qu'est la vraie droite ou la vraie gauche. Ils ne pensent pas, ils défendent des étiquettes. Voici les données auxquelles tout ce monde devrait penser, à mon avis.

⚠️ A Nature study found that AI chatbots agree with users far too often, with models endorsing user views 50% more than humans A key finding is prompting that asks the model to first verify the statement before proving it cuts sycophancy meaningfully, and targeted fine-tuning

AI doesn't just need agents - it needs workflows. 🚀 Today we're launching Pipelex: an open-source DSL + Python runtime for repeatable AI workflows. Think Dockerfile/SQL for LLM pipelines - declare steps, swap models, let the runtime handle execution.

Odd for OpenAI employees to claim they have more ‘stringent’ training data practices than people who train on “outright pirated data”, when they *checks notes* also train on outright pirated data

This paper from Tsinghua University and Shanghai Jiao Tong University received perfect scores (6, 6, 6, 6) at NeurIPS 2025! It aims to answer a key question: Does reinforcement learning really make large language models better reasoners? The authors study Reinforcement Learning

Today, Photoroom is turning 6, we’re releasing our second text-to-image model from scratch— and open-sourcing not only the weights, but also the full process. Photoroom, from day 1, was successful thanks to open source in the early days and it's our way to give back to the