Articles to discover
AI · Nov 11, 2026Gemini 4 Carbon: what Google wants to entrust to an AI agentGoogle is making Gemini 4 Carbon available to new users, with different levels of access depending on the product. Its stated performance in coding, agentic tasks and search should be considered in light of the tests and safeguards required.↗
AI · Nov 10, 2026Signal: what AI has really proved about its codeGoogle DeepMind says it used AlphaProof Code to prove certain properties of Signal’s compiled code. The result sheds light on AI’s potential, but also on the precise limits of formal proof.↗
AI · Nov 8, 2026AI and research: behind the hours saved, the challenge of proofA study by Google and researchers affiliated with MIT reports time savings declared by scientists, but also hypotheses awaiting testing and a substantial need for verification. Speeding up tasks is not enough to validate discoveries.↗
Education · Nov 6, 2026AI at university: in Latin America, use is outpacing the rulesA survey of 200 institutions highlights a gap between adoption and strategy. In education, using AI is not enough to prove its effectiveness.↗
AI · Nov 4, 2026Genie 4: changing a world live puts consistency to the testGoogle is adding live instructions and sound to its generated worlds. To understand what Genie 4 offers, it is important to distinguish between a convincing setting, a useful prototype and a verifiable simulation.↗
Technology · Nov 1, 2026Gemini Notebook records audio: what it keeps, and what you need to checkThe new mobile recorder turns speech into a working source. But a transcript is not an audio archive: what you need to know before a lecture or meeting.↗
AI · Oct 25, 2026AI agents at work: what TheAgentCompany really measuresCarnegie Mellon’s test checks tasks in a simulated company. Its first scores date from 2024: they explain some limitations, but do not rank today’s AI systems.↗
View all results →