Looking for something?
A place, an idea, a question. Find the articles that interest you.
Articles to discover
Luxor bets on AI: what Bitcoin mining can bring to the tableThe company has announced a $16 million financing round to expand its computing services. Its claimed advantage lies in equipment, energy and contracts—not in magically turning mining machines into something else.
AI and scientific evidence: what Lean checks, and what still needs examiningOpenAI’s announcements have revived the question of evidence produced by AI. The logical core, assumptions and results of AlphaProof: key points for assessing them.
Perec at the BnF: how a rule can open up an entire novelA missing letter, a puzzle-like building and variations: keys to understanding Perec’s choices and getting ready for the BnF exhibition, on until 14 February 2027.
SK hynix: what its record quarter reveals about the memory market for AIResults, HBM memory and supply contracts: understanding SK hynix’s growth and the industrial risks behind AI-related demand.AI agents: what open standards allow you to keepThe AAIF welcomes four projects to make agents more portable. Methods, models, testing and security: here’s what can travel, and what still needs checking.
AI assistants at work: who really chooses the criteria?An Anthropic experiment raises questions about control over decisions entrusted to agents. For users, control cannot be limited to checking the figures.
Fujifilm is betting on magnetic tape to store AI dataThe group has announced an investment of 10 billion yen in Odawara. Behind the goal of increasing capacity fivefold lies a practical issue: storing data without confusing archiving, backup and immediate access.
Hohle Fels: what miniature art reveals about prehistoryTwo ivory birds, discovered in 2025 and presented in July 2026, add to the story of this German cave. From the carved horse to the flute, how can we interpret these objects without inventing a story for them?
Election images: ChatGPT’s filter, between illustration and photorealismThe filter reported by CNET turns some requests into stylised images. But two French tests described by Next produced different results: friction in image generation is not a check on the truth.
Core Advanced: what do the 2 GB of memory really measure?Apple’s model has 20 billion parameters, but a reading taken after a prompt shows around 2 GiB. These figures do not measure the same thing: one describes the entire model, the other a memory state observed on a particular Mac.
In Rennes, AI is not presented as the sole reason for the fineThe legal commentary on the case places the indications associated with Claude after other complaints about the application. Typography alone is therefore not enough to conclude that a tool was used — or that its use explains the penalty.
After Trappes: what can be established about hate speech and violenceIn the National Assembly, responsibility for words has become a political battleground. Yet the investigation has established no direct link with a political party: research urges us to distinguish collective influence from individual causation.