https://news.tyber.io/en/news.tyber.io — EnglishAI news with a senior engineer's lens.2026-05-04T18:35:00.595ZJP Tibériohttps://www.linkedin.com/in/jp-tiberiohttps://news.tyber.io/en/research/ease-federated-multimodal-unlearning-via-entanglement-aware-anchor-closure/EASE: a paper on federated multimodal unlearning, and what to read it for2026-05-04T18:31:10.827Z2026-05-04T17:53:00.825ZJP TibérioA new arxiv paper proposes 'anchor closure' to make federated multimodal models forget specific data without leaking it back through the other modality. Here's the engineering shape of the problem and what to actually watch for in the results.https://news.tyber.io/en/open-source/ruvnet-ruflo/ruvnet/ruflo trends on GitHub: an agent orchestration pitch worth reading skeptically2026-05-04T18:30:00.000Z2026-05-04T17:51:24.061ZJP TibérioA repository pitching multi-agent swarms for Claude is climbing GitHub trending. The marketing is loud; the engineering questions it raises are the interesting part.https://news.tyber.io/en/research/when-rag-chatbots-expose-their-backend-an-anonymized-case-study-of-privacy-and-s/Patient-facing medical RAG chatbot gets an anonymized security audit2026-05-04T18:30:00.000Z2026-05-04T17:50:51.837ZJP TibérioA new arxiv paper documents a non-destructive security assessment of a publicly accessible medical RAG chatbot. The findings list isn't out yet — but the framing is already worth reading.https://news.tyber.io/en/industry/welcome-news-tyber-io/What this portal is — and how it stays honest2026-05-04T12:00:00.000Z2026-05-04T12:00:00.000ZJP Tibérionews.tyber.io is the public archive of the AI Radar I run privately. Here's the editorial spine — two-source rule, retraction policy, and what the LLM is and isn't allowed to do.