Future Proofing | Diuna Technologies

RAG Demystified: How Retrieval-Augmented Generation Improves Factuality

Plain definition RAG (Retrieval-Augmented Generation) combines a generator (LLM/SLM) with a retriever over your trusted knowledge sources; the model conditions on retrieved passages to produce grounded answers. Original formulation: Lewis et al., 2020 (NeurIPS). Why RAG helps? Reduces hallucinations , provides provenance , and enables freshness by pulling up-to-date documents at answer time. (Motivation from the original RAG paper and vendor architecture guides.) 2025 arc

Diuna Academy

13 hours ago

Red-Teaming & Continuous Assurance for Frontier Systems

Government baseline The UK/US AISI pre-deployment eval of o1 shows public sector expectations: domain-specific tests (cyber, persuasion, biosecurity), red-team procedures, and publishable summaries. Pair this with NIST AI RMF (govern–map–measure–manage) for lifecycle discipline. Your operating loop Threat model. List misuse risks by domain (sector + AISI domains). Adversarial testing. Run jailbreak and tool-use red-teams; include autonomous-agent behavior and data leakag

Diuna Academy

2 days ago

Responsible Scaling in Practice - DeepMind FSF vs Anthropic RSP vs OpenAI Preparedness (2025)

Scaling drives capability jumps; leading labs now publish thresholded safety policies . Using them as templates will lift your governance to frontier-grade. What does each framework require? Control theme DeepMind FSF (2025) Anthropic RSP / ASL OpenAI Preparedness v2 (Apr 15, 2025) Triggering events Critical Capability Levels (CCLs) incl. deceptive-alignment risk; stronger security by CCL Capability Thresholds → escalate to ASL-3 safeguards Tracked Categories (Bio/Chem, Cyber

Diuna Academy

7 days ago

Frontier Model definition

Plain definition (policy-aligned). Governments and leading labs use the frontier model to mean a highly capable, general-purpose foundation model (typically transformer-based) whose fast-scaling abilities can introduce severe, dual-use risks . The term is prominent in the UK/US AI Safety Institutes’ joint work and in lab safety policies (DeepMind FSF, Anthropic RSP, OpenAI Preparedness). How does this relate to Foundation & GPAI? Foundation model: trained on broad data, a

Diuna Academy

Nov 12