Techniques

Methods for getting the most out of models

44 articles · 3 sous-catégories

Agentic red teaming — when the LLM is no longer the main target

Why auditing an agentic system means widening the perimeter beyond the LLM: the Dreadnode taxonomy of 5 fronts, the OWASP ASI framework, an audit methodology and actionable mitigations.

agenticsecurityred-teamingauditowasp-asidreadnodesupply-chainmcp

analyse Multi-Agent Orchestration

The web getting ready for AI agents — where the transition is actually happening

An empirical 2026 survey of the web facing AI agents: outbound-click disintermediation, headless bimodality between enterprise and consumer, agents ready only for bounded use cases, and fragmented bidirectional standards (MCP, NLWeb, Cloudflare Pay Per Crawl, llms.txt).

webagents-iamcpheadlesssearchtransition2026

analyse RAG & document retrieval

RAG Panorama April 2026 — Vector, Vectorless and Emerging Approaches

State of the art of Retrieval-Augmented Generation in April 2026: dense vector, vectorless/sparse/hybrid, and emerging (GraphRAG, Self-RAG, Agentic) families, plus the long-context debate. Pedagogical version.

ragretrievalembeddingsbm25colbertgraphragself-ragagentic-ragvector-databasehybrid-searchlong-context

Agentic red teaming — when the LLM is no longer the main target

The web getting ready for AI agents — where the transition is actually happening

RAG Panorama April 2026 — Vector, Vectorless and Emerging Approaches

Agentic AI — LLMs that take action

Multi-answer RLVR — training a model to explore

Agent0 — learning to improve without any human data

AI research automation — how far does it actually work?

Context graphs — the next infrastructure for AI agents?

Agent harnesses — the architecture that wraps the LLM

MemEvolve — when agent memory self-optimizes

Meta-Harness — automatically optimizing the code around the LLM

Asynchronous coding agents — effective coordination strategies

HACRL — when LLMs of different sizes train together

λ-RLM — lambda calculus to make LLM reasoning more reliable

Automated multi-agent auditing — when agents verify their own system

Chain-of-Debate — when agents argue to make better decisions

Evaluating an LLM Agent — methods and pitfalls

Multi-agent exception handling — when agents manage the unexpected

Multi-agent systems — when collaboration amplifies (or not)

Agentic AI planning — formulate, execute, adapt

Prompt chaining vs autonomous agent — which one to choose?

Reflection and self-critique — when agents correct themselves

Routing in agentic AI — how agents choose their path

Self-Consistency — running the same prompt 3 times

LLM Agents — how a language model moves from response to action

Context prompting — a practical guide by usage level

35 LLM agents in parallel — anatomy and positioning

Model distillation — transferring knowledge from large to small

LLMs and code generation — what benchmarks don't tell you

Voice models — the AI that speaks and listens

LLM inference optimization — how the industry cuts the bill

Quantization — shrinking an LLM without breaking it

Adversarial security of LLMs — attacks and defenses

Context engineering — beyond prompt engineering

Agents in production — what 1,200 real cases reveal

Cognitive dispatch — assigning the right model to the right task

Human escalation vs. auto-recovery — who corrects the agent?

174 agents, 0 failures: real reliability or measurement bias?

Fire-and-consolidate — a multi-agent orchestration pattern

The LLM/language paradox — why machines prefer machine language

The Wave pattern — orchestrating in two waves

More agents ≠ better results — the multi-agent scaling paradox

Step 0bis — challenging your own architecture before acting

The decision triangle — governing an agentic AI system